# AI & SEO Weekly #2: Apple's First Foldable, OpenAI's Agents API, and Meta's Muse

> Canonical HTML version: https://thejayant.in/blog/weekly/2026-09-13
> Author: Jayant Solanki — https://thejayant.in/
> This Markdown file is a plain-text twin of the article at the URL above. Same content, no page furniture. It is public, not bot-only.

- **Apple shipped its first foldable.** The iPhone Duo, a 7.6-inch inner display and no Face ID, launched alongside an Apple Watch that listens to your conversations and recaps them.
- **OpenAI had a developer wave on 10 September** — the Agents API in public beta, GPT‑Live‑1 voice in the API, a Data agent for ChatGPT Work, and ChatGPT for Financial Services. Then it paused $200 Pro sign-ups because Astra demand broke capacity.
- **Meta launched Muse, a personal AI agent that connects to your email, calendar and payments.** Two days later it was the **No. 2 app in the US**.
- **Anthropic named names.** Its September threat report alleges sustained distillation campaigns by Alibaba, Moonshot AI and DeepSeek against Claude's agentic, coding and reasoning capabilities.
- **Google Analytics finally shipped Dashboards** — drag-and-drop KPI reports on one canvas, confirmed in the release notes on 9 September.

This is issue #2 of a weekly roundup covering AI, SEO, Google, Meta, the model labs — and, at the very end, whatever I have been playing. Everything below happened between **6 and 12 September 2026**. [Issue #1 is here](https://thejayant.in/blog/weekly/2026-09-05) if you missed it.

It was a heavy week. Apple held its fall event, OpenAI ran what amounted to a developer day, Meta entered the agent race, and Anthropic published the most pointed threat report any lab has released. One hundred and two stories below, grouped by section.

Every item follows the same shape: what happened, why it matters, and a link to the source that reported it. Every claim carries a label — Confirmed means the company said it themselves, Reported means a named outlet reported it, Observed means somebody watched it happen in the product and it may not be a deliberate launch at all. Where a section was genuinely quiet, it says so and names the source I checked.

## The five that actually matter

If you read nothing else this week, read these. Ranked by how much they change what somebody has to do on Monday, not by headline volume.

## AI 14 stories

### Mistral raised €3 billion at a €21 billion valuation, the largest European tech equity round ever

Reported 8 Sep · [TechCrunch](https://techcrunch.com/2026/09/08/mistral-raises-e3b-as-sovereign-ai-becomes-big-business/)

![Mistral raised three billion euros as sovereign AI becomes big business — image published by TechCrunch](https://techcrunch.com/wp-content/uploads/2026/03/GettyImages-2264771189.jpg?resize=1200,800)
_Image: [TechCrunch](https://techcrunch.com/2026/09/08/mistral-raises-e3b-as-sovereign-ai-becomes-big-business/)_

The French lab confirmed a **€3 billion** (about $3.58 billion) Series D at a post-money valuation above **€21 billion** (about $24.39 billion), led by Samsung Electronics with EQT's Scaleup Europe Fund and PSG Equity as co-leads. Mistral called it the largest equity round ever completed by a European technology company.

**Why it matters:** "sovereign AI" — a model your government can point at and call domestic — is now a fundable category on its own, which means European and Indian buyers get a third option that is not American or Chinese.

### Cognition hit a $48 billion valuation, signalling that AI coding is not winner-take-all

Reported 8 Sep · [TechCrunch](https://techcrunch.com/2026/09/08/cognition-hits-48b-valuation-signaling-investors-believe-ai-coding-is-far-from-a-winner-take-all-market/)

![Cognition CEO Scott Wu — image published by TechCrunch](https://techcrunch.com/wp-content/uploads/2026/08/Scott-Wu-of-Cognition.jpg?w=1024)
_Image: [TechCrunch](https://techcrunch.com/2026/09/08/cognition-hits-48b-valuation-signaling-investors-believe-ai-coding-is-far-from-a-winner-take-all-market/)_

Cognition, maker of the Devin coding agent, reached a **$48 billion** valuation — in the same week Anthropic, OpenAI and Google all ship coding agents of their own.

**Why it matters:** investors are pricing in several durable winners in AI coding rather than one. If you are choosing tooling, you are choosing for a market that will still have competitors in two years.

### Harvey reached a $15.5 billion valuation, months after hitting $11 billion

Reported 9 Sep · [TechCrunch](https://techcrunch.com/2026/09/09/harvey-hits-15-5b-valuation-months-after-reaching-11b/)

The legal AI company's valuation rose from $11 billion to **$15.5 billion** in a matter of months.

**Why it matters:** vertical AI — one profession, deeply served — is compounding faster than horizontal assistants, which is the opposite of what the "one model does everything" narrative predicts.

### An Anthropic researcher quit, warning that self-improving AI is "gambling with our lives"

Reported 9 Sep · [TechCrunch](https://techcrunch.com/2026/09/09/gambling-with-our-lives-anthropic-researcher-quits-warns-against-self-improving-ai/)

![An Anthropic researcher quit warning against self-improving AI — image published by TechCrunch](https://techcrunch.com/wp-content/uploads/2026/09/GettyImages-2063425288.jpg?resize=1200,818)
_Image: [TechCrunch](https://techcrunch.com/2026/09/09/gambling-with-our-lives-anthropic-researcher-quits-warns-against-self-improving-ai/)_

A researcher left Anthropic publicly, arguing against the industry's push toward models that improve themselves. TechCrunch followed with a podcast on 11 September noting the warning arrived "at a very interesting time" — days before the company's CEO published his own framing.

**Why it matters:** safety departures used to be quiet. Public resignations with a stated technical objection are a different signal, and they are now coming from inside the lab that markets itself on caution.

### Anthropic's CEO published a plan to "pace the frontier"

Reported 12 Sep · [TechCrunch](https://techcrunch.com/2026/09/12/anthropic-ceo-outlines-plan-to-pace-the-frontier/)

![Anthropic CEO Dario Amodei — image published by TechCrunch](https://techcrunch.com/wp-content/uploads/2026/03/Dario-Amodei-Anthropic-1.jpg?w=1024)
_Image: [TechCrunch](https://techcrunch.com/2026/09/12/anthropic-ceo-outlines-plan-to-pace-the-frontier/)_

Dario Amodei set out an argument for controlling the speed of frontier development, published three days after a researcher resigned over the same question.

**Why it matters:** "pace" is a commercial position as much as a safety one. A lab that slows down needs everyone else to slow down too, which is why this is being argued in public rather than in a policy document.

### OpenAI added Paul Christiano, a prominent AI-risk researcher, to its Foundation Board

Confirmed 9 Sep · [OpenAI](https://openai.com/index/paul-christiano-joins-openai-foundation-board/) · [TechCrunch](https://techcrunch.com/2026/09/09/openai-adds-a-prominent-ai-doomer-to-its-board-of-directors/)

OpenAI confirmed Christiano's appointment to the Foundation Board on 9 September. TechCrunch's framing — "a prominent AI doomer" — captures why the appointment was noticed.

**Why it matters:** board composition is the only safety lever with actual authority behind it. Whether it is used is a separate question, but the appointment is a real one, not a blog post.

### OpenAI paused new $200-a-month Pro subscriptions because Astra demand was straining its infrastructure

Reported 10 Sep · [TechCrunch](https://techcrunch.com/2026/09/10/openai-puts-pro-subscriptions-on-hold-due-to-astra-demand/)

![OpenAI paused Pro subscriptions due to Astra demand — image published by TechCrunch](https://techcrunch.com/wp-content/uploads/2026/09/openai-getty.jpg?resize=1200,800)
_Image: [TechCrunch](https://techcrunch.com/2026/09/10/openai-puts-pro-subscriptions-on-hold-due-to-astra-demand/)_

Product lead Thibault Sottiaux announced on X that Pro sign-ups were being disabled because the tier "puts the most strain" on OpenAI's systems. The API, Go and Plus plans stayed open. He described it as "the smallest step that allows us to continue giving the broadest access possible".

**Why it matters:** a frontier lab turning away $200/month customers is a capacity statement, not a pricing one. If you are building on Astra, plan for rate limits rather than assuming headroom.

### Moonshot AI, the lab behind Kimi, is targeting $2 billion in annual revenue

Reported 11 Sep · [TechCrunch](https://techcrunch.com/2026/09/11/kimi-maker-moonshot-ai-targets-2-billion-in-annual-revenue/)

![Moonshot AI, maker of Kimi, targets two billion dollars in annual revenue — image published by TechCrunch](https://techcrunch.com/wp-content/uploads/2026/07/GettyImages-2286500505.jpg?resize=1200,800)
_Image: [TechCrunch](https://techcrunch.com/2026/09/11/kimi-maker-moonshot-ai-targets-2-billion-in-annual-revenue/)_

Moonshot set a **$2 billion** annual revenue target — in the same week Anthropic accused it of distilling Claude.

**Why it matters:** the Chinese labs are no longer competing only on price and openness. A revenue target at that scale means enterprise contracts, which means they will show up in procurement shortlists outside China.

### Y Combinator's Garry Tan said US open-weight labs should distill frontier models too

Reported 11 Sep · [TechCrunch](https://techcrunch.com/2026/09/11/y-combinators-garry-tan-wants-u-s-open-weight-ai-labs-to-distill-frontier-models-too/)

Tan argued American open-weight labs should use the same distillation techniques Anthropic spent the week condemning.

**Why it matters:** the norm against distillation is not settled, and it is being contested by people with capital rather than only by the labs being distilled. Expect terms-of-service enforcement to become the actual battleground.

### AI agents are flooding public services with automated requests

Reported 10 Sep · [TechCrunch](https://techcrunch.com/2026/09/10/ai-agents-are-flooding-public-services-with-new-requests/)

![AI agents are flooding public services with new requests — image published by TechCrunch](https://techcrunch.com/wp-content/uploads/2026/06/GettyImages-2264911748.jpg?resize=1200,705)
_Image: [TechCrunch](https://techcrunch.com/2026/09/10/ai-agents-are-flooding-public-services-with-new-requests/)_

Public bodies are seeing volumes of agent-generated requests — records requests, form submissions, queries — that their processes were never sized for.

**Why it matters:** this is the first at-scale example of what agentic traffic does to a system that assumed a human on the other end. Any site with a form, a quote request or a support queue is next, and rate limiting a customer's agent is a customer-service decision, not just an engineering one.

### Suno replaced its AI models with one trained on licensed music as copyright suits piled up

Reported 9 Sep · [TechCrunch](https://techcrunch.com/2026/09/09/suno-replaces-its-ai-models-with-a-new-one-trained-on-licensed-music-as-copyright-suits-pile-up/)

Suno swapped out its generation models for a new one trained on licensed material, with litigation mounting.

**Why it matters:** the first major generative company to rebuild its model on licensed data under legal pressure. If it works commercially, it becomes the template every rights-holder points at.

### Jensen Huang explained why Nvidia expects to grow 70% next year

Reported 10 Sep · [TechCrunch](https://techcrunch.com/2026/09/10/jensen-huang-explains-why-nvidia-will-grow-an-astounding-70-next-year/)

Nvidia's CEO set out the case for **70%** growth in the coming year, a fortnight after confirming the Hugging Face acquisition covered in [issue #1](https://thejayant.in/blog/weekly/2026-09-05).

**Why it matters:** compute pricing follows Nvidia's capacity assumptions. A 70% growth plan is the clearest available signal that inference costs are not about to fall off a cliff.

### India's Pocket FM doubled its revenue run rate to $500M, with AI producing 93% of its audio

Reported 10 Sep · [TechCrunch](https://techcrunch.com/2026/09/10/indias-pocket-fm-doubles-revenue-run-rate-to-500m-as-ai-powers-93-of-audio-content/)

![India's Pocket FM doubled its revenue run rate to 500 million dollars — image published by TechCrunch](https://techcrunch.com/wp-content/uploads/2026/09/pocket-fm.jpg?resize=1200,800)
_Image: [TechCrunch](https://techcrunch.com/2026/09/10/indias-pocket-fm-doubles-revenue-run-rate-to-500m-as-ai-powers-93-of-audio-content/)_

The Indian audio-series platform doubled its run rate to **$500 million**, with AI now generating **93%** of its audio content.

**Why it matters:** the clearest published example of AI production economics working at consumer scale — and it is an Indian company, in a category Western media has barely noticed.

### AI spend per employee fell at top firms in August

Reported 9 Sep · [TechCrunch](https://techcrunch.com/2026/09/09/ai-spend-per-employee-slumped-at-top-firms-in-august-summer-doldrums-or-a-warning-sign/)

Per-employee AI spending dropped at leading firms during August. TechCrunch frames the open question honestly: summer doldrums, or the first sign of budget fatigue.

**Why it matters:** one month is not a trend, but it is the first down month in a long run of up ones. Worth watching in September's data before drawing anything from it.

## AI SEO, GEO & AIO 8 stories

### Google DeepMind published research on replacing search's two-stage ranking architecture with a single LLM

Reported 10 Sep · [Search Engine Journal](https://www.searchenginejournal.com/google-deepmind-develops-new-ai-search-ranking-model/589062/)

![Google DeepMind's Autoregressive Ranking research — image published by Search Engine Journal](https://cdn.searchenginejournal.com/wp-content/uploads/2026/09/google-autoregressive-ranking-620.jpg)
_Image: [Search Engine Journal](https://www.searchenginejournal.com/google-deepmind-develops-new-ai-search-ranking-model/589062/)_

DeepMind's paper proposes **Autoregressive Ranking** in place of the standard two-stage setup. Today a Dual Encoder turns queries and documents into vectors and retrieves candidates cheaply, then a Cross Encoder — far more accurate, far more expensive — ranks that shortlist. The proposal collapses both into one LLM that ranks directly.

**Why it matters:** this is a research paper, not a launch, and should be read as direction rather than deployment. But every "how ranking works" mental model you use rests on the two-stage architecture, and the team that builds Google's models is publishing on replacing it.

### A University of Washington paper estimates AI Overviews cut search referrals to English Wikipedia by about 5%

Reported 11 Sep · [Search Engine Journal](https://www.searchenginejournal.com/what-wikipedia-reveals-about-ai-overviews-and-web-traffic/589042/)

![What Wikipedia reveals about AI Overviews and web traffic — image published by Search Engine Journal](https://cdn.searchenginejournal.com/wp-content/uploads/2026/09/wikipedia-ai-overviews-featured.jpg)
_Image: [Search Engine Journal](https://www.searchenginejournal.com/what-wikipedia-reveals-about-ai-overviews-and-web-traffic/589042/)_

Mehrzad Khosravi and Hema Yoganarasimhan estimate a **~5%** drop in monthly search referrals to English Wikipedia after AI Overviews became the US default in 2024. The working paper was last updated 2 September and **has not been peer reviewed**; it revises down a larger estimate the same authors published in February. Google disputes whether Wikimedia's combined referral metric can isolate the feature at all.

**Why it matters:** this is the most-cited number in the "AI is killing traffic" argument, and it just got smaller and more contested. If you are making a business case either way, quote the 5% with its caveats or do not quote it.

### Le Monde says AI Overviews have not hurt its audience so far

Reported 10 Sep · [Search Engine Journal](https://www.searchenginejournal.com/le-monde-says-ai-overviews-havent-hurt-its-audience-so-far/589018/)

![Le Monde says AI Overviews have not hurt its audience so far — image published by Search Engine Journal](https://cdn.searchenginejournal.com/wp-content/uploads/2026/09/le-monde-ai-overviews-featured-1600x840-1.jpg)
_Image: [Search Engine Journal](https://www.searchenginejournal.com/le-monde-says-ai-overviews-havent-hurt-its-audience-so-far/589018/)_

The French newspaper reported no measurable audience damage from AI Overviews to date — a direct counterweight to the Wikipedia finding published a day later.

**Why it matters:** a reference site and a news publisher are affected differently, because the queries that reach them are different. "AI Overviews cost traffic" is not one question with one answer; it is a question about your query mix.

### ChatGPT Shopping recommendations from product feeds jumped from 8.26% to 61.54% in a single day

Reported 9 Sep · [Search Engine Journal](https://www.searchenginejournal.com/chatgpt-shopping-results-lean-hard-on-product-feeds/589000/)

![ChatGPT Shopping results lean hard on product feeds — image published by Search Engine Journal](https://cdn.searchenginejournal.com/wp-content/uploads/2026/09/chatgpt-shopping-product-feeds-featured.jpg)
_Image: [Search Engine Journal](https://www.searchenginejournal.com/chatgpt-shopping-results-lean-hard-on-product-feeds/589000/)_

Profound data across **1,757,723** tracked prompt runs in July found the share of ChatGPT Shopping product picks classified as feed-integrated rose from **8.26% to 61.54% on 10 July**. Products now arrive predominantly through feed retrieval rather than web search.

**Why it matters:** this is the single most actionable number in the issue for anyone selling online. If your catalogue is not in a connected feed, the majority of ChatGPT's product recommendations are now drawn from a pool you are not in. I have written the longer version in [AI shopping and product feeds](https://thejayant.in/blog/ai-shopping-seo).

### Google documented a partner-only API that returns full-web Search results as JSON

Reported 11 Sep · [Search Engine Journal](https://www.searchenginejournal.com/google-documents-partner-only-api-for-full-web-search-results/589194/)

![Google documents a partner-only API for full-web search results — image published by Search Engine Journal](https://cdn.searchenginejournal.com/wp-content/uploads/2026/09/google-partner-api-featured-1600x840-2.jpg)
_Image: [Search Engine Journal](https://www.searchenginejournal.com/google-documents-partner-only-api-for-full-web-search-results/589194/)_

Google refreshed the developer docs for its **Web Search Service API** on 9 September. It returns Google Search results as JSON, and access requires a Google Cloud project, an API key and a client ID issued under a partner agreement. Custom Search JSON API users who need full-web results must migrate **by January**.

**Why it matters:** if your rank tracking or research tooling depends on the Custom Search JSON API, you have a migration deadline and a gatekeeping problem — the replacement is not self-serve.

### Google Ads data shows queries getting longer since AI Mode launched

Reported 11 Sep · [Search Engine Land](https://searchengineland.com/google-ads-data-shows-query-length-shift-post-ai-mode-458162)

Jason Tabeling's updated analysis, extended through August 2026, finds longer and more specific searches gaining impression and conversion share while short queries keep losing ground — continuing a shift he first measured a year earlier.

**Why it matters:** it confirms with paid-search data what most people only assert: conversational search habits are bleeding into classic search. Exact-match keyword strategies are describing a shrinking share of demand.

### Duane Forrester on why everyone quotes the same AI-visibility number and gets different answers

Reported 10 Sep · [Search Engine Journal](https://www.searchenginejournal.com/everyone-is-quoting-the-same-number-and-getting-different-answers/588659/)

Forrester picks apart why the same headline statistic produces contradictory conclusions across the industry — different sampling, different definitions of a citation, different platforms measured.

**Why it matters:** GEO measurement has no agreed methodology yet, which means vendor dashboards are not comparable. Read the method before you read the number — which is the whole reason my own [visibility benchmark](https://thejayant.in/blog/ai-search-visibility-benchmark) published its methodology before its results.

### A study of 120,000 AI mentions across five models identified four signals that drive multi-location visibility

Reported 10 Sep · [Search Engine Journal](https://www.searchenginejournal.com/ai-signals-multi-location-uberall-spa/587988/)

Uberall analysed **120,000** mentions across five AI models to isolate what actually moves visibility for businesses with many physical locations.

**Why it matters:** multi-location brands have the hardest GEO problem — one brand, hundreds of entities, and models that resolve them inconsistently. Sample-based studies are currently the only evidence available.

## AI Automations 4 stories

### OpenAI launched the Agents API in public beta, exposing the Codex harness to developers

Confirmed 10 Sep · [OpenAI](https://openai.com/index/introducing-the-agents-api/)

![OpenAI introduces the Agents API — image published by OpenAI](https://images.ctfassets.net/kftzwdyauwt9/430gA5qBJZHD8Dd4uGmm6J/f91d9df145eab0f0f981f60958026813/og-image.png?w=1200&h=675&fit=fill)
_Image: [OpenAI](https://openai.com/index/introducing-the-agents-api/)_

The Agents API gives developers the same harness and infrastructure that runs Codex: context management, efficient tool use, subagent coordination, and environments that keep agents running reliably **for days** with files, code execution and saved intermediate results. Customer Ciridae reported its evaluation score rising from **0.71 to 0.85** and a **4× latency reduction** from built-in subagent support.

**Why it matters:** the hard part of agents was never the model — it was the scaffolding around it. Renting that scaffolding removes months of work, and it means far more agents are about to exist. See the earlier point about [public services drowning in agent traffic](#agents-public-services).

### OpenAI brought GPT-Live-1, its full-duplex voice model, to the API

Confirmed 10 Sep · [OpenAI](https://openai.com/index/introducing-gpt-live-1-in-the-api/)

![OpenAI brings GPT-Live-1 to the API — image published by OpenAI](https://images.ctfassets.net/kftzwdyauwt9/7Mz7MwoSyFLdOMk26N1hCA/b66e84b656c1d666fc2adc1c56ff67f1/introducing-gpt-live-1-green-seo.png?w=1200&h=675&fit=fill)
_Image: [OpenAI](https://openai.com/index/introducing-gpt-live-1-in-the-api/)_

GPT-Live-1 listens and speaks at the same time — full duplex — and can delegate deeper reasoning to the models and tools it is paired with. The API release adds controls for steering how voice agents speak and act, building on its interruption handling.

**Why it matters:** full-duplex voice is what makes a phone agent tolerable rather than infuriating. Expect a wave of voice front-ends on support and booking flows within a quarter.

### OpenAI introduced a Data agent in ChatGPT Work that builds dashboards from company data on request

Confirmed 10 Sep · [OpenAI](https://openai.com/index/put-data-to-work/)

![OpenAI introduces a Data agent in ChatGPT Work — image published by OpenAI](https://images.ctfassets.net/kftzwdyauwt9/2UxvYcTeyyfNcGIxHFKW06/82ff8662f3a80f6b7319afb2a25dc6ae/put-data-to-work--seo-v001.png?w=1200&h=675&fit=fill)
_Image: [OpenAI](https://openai.com/index/put-data-to-work/)_

The Data agent connects to company data, investigates what changed, and builds shareable interactive dashboards — refined in conversation, without writing queries or learning an analytics tool.

**Why it matters:** it aims squarely at the "wait for the analyst" bottleneck. It also aims squarely at the BI vendors, in the same week Google Analytics shipped [its own drag-and-drop dashboards](#ga-dashboards).

### n8n 2.39.0 added Git push and pull for projects through the public API

Confirmed 8 Sep · [n8n release notes](https://github.com/n8n-io/n8n/releases/tag/n8n%402.39.0)

Fourteen n8n releases landed between 6 and 11 September. The feature release, **2.39.0** on 8 September, added source-control push, pull and status endpoints to the public API, a workflow version endpoint, folder exploration for Instance AI, an instance reporting module, and opt-in concurrency limits for Instance AI runs.

**Why it matters:** automations become reviewable infrastructure once Git operations are API-driven. If your team treats workflows as production code, this is the release that makes CI over n8n practical.

## SEO 7 stories

### No confirmed Google ranking update ran this week

Confirmed 12 Sep · [Google Search Status Dashboard](https://status.search.google.com/summary)

The Search Status Dashboard shows no ranking or serving incident in the window. The most recent confirmed update remains the **August 2026 spam update**, which ran 18–21 August. Nothing has been announced since.

**Why it matters:** if your rankings moved this week, the cause was your site, your competitors or normal volatility — not an update. That is a genuinely useful thing to be able to rule out before you start changing pages.

### Google said the old "position 1 to 10" model is hard to map onto modern results

Reported 10 Sep · [Search Engine Roundtable](https://www.seroundtable.com/google-position-1-10-hard-map-42053.html)

![Google says the old position 1-10 model is hard to map for site owners — image published by Search Engine Roundtable](https://images.seroundtable.com/google-serp-hero-ghUdMYys.jpg)
_Image: [Search Engine Roundtable](https://www.seroundtable.com/google-position-1-10-hard-map-42053.html)_

Google acknowledged that the classic ten-blue-links ranking positions are difficult to map for site owners now that results carry AI Overviews, AI Mode entry points and mixed modules.

**Why it matters:** it is confirmation from the source that average position is decreasingly meaningful. If your reporting still leads with it, the number is measuring a layout that no longer exists.

### A survey of SEO practitioners put content and backlinks at the top of the ranking-factor list

Reported 10 Sep · [Search Engine Roundtable](https://www.seroundtable.com/google-ranking-factors-expert-survey-42057.html)

![Google ranking factors expert survey — image published by Search Engine Roundtable](https://images.seroundtable.com/google-rank-survey-hero-uwK3uH5v.jpg)
_Image: [Search Engine Roundtable](https://www.seroundtable.com/google-ranking-factors-expert-survey-42057.html)_

An expert survey ranked content quality and backlinks as the most important factors — the same two that have topped these surveys for a decade.

**Why it matters:** a survey measures what practitioners believe, not what Google does. Useful as a read on industry consensus; not evidence about the algorithm.

### Search Central Live is coming to Bogotá and Mexico City

Confirmed 8 Sep · [Google Search Central](https://developers.google.com/search/blog/2026/09/search-central-live-mexico-and-colombia)

Google announced Search Central Live events in Colombia and Mexico — the only post on the Search Central Blog during the window.

**Why it matters:** Search Central Live is where Google's search relations team answers questions on record. The Latin America dates are worth knowing if you work those markets; the quiet blog is worth knowing either way.

### Scammers are embedding phone numbers in Google Maps business photos

Observed 10 Sep · [Search Engine Roundtable](https://www.seroundtable.com/google-maps-phone-number-in-photo-scam-42056.html)

![Google Maps phone number in photo scam — image published by Search Engine Roundtable](https://images.seroundtable.com/broken-phone-google-maps-D3V02zAI.jpg)
_Image: [Search Engine Roundtable](https://www.seroundtable.com/google-maps-phone-number-in-photo-scam-42056.html)_

A scam is circulating in which a fraudulent phone number is baked into an image uploaded to a Maps business listing, bypassing the checks applied to the listing's phone field.

**Why it matters:** if you manage local listings, audit your uploaded photos as well as your NAP fields. Text inside an image is not moderated the same way and customers cannot tell the difference.

### Google added post view counts to Business Profiles

Reported 10 Sep · [Search Engine Journal](https://www.searchenginejournal.com/google-business-profile-post-view-counts/589128/)

![Google adds post view counts to Business Profiles — image published by Search Engine Journal](https://cdn.searchenginejournal.com/wp-content/uploads/2026/09/google-business-profile-post-view-counts-featured.jpg)
_Image: [Search Engine Journal](https://www.searchenginejournal.com/google-business-profile-post-view-counts/589128/)_

Business Profile posts now show a view count, giving local businesses their first direct measure of whether posting is doing anything.

**Why it matters:** Business Profile posts have always been an act of faith. A view count lets you decide whether to keep making them with evidence rather than habit.

### Self-reported content marketing success fell to a 12-year low

Reported 10 Sep · [Search Engine Land](https://searchengineland.com/content-marketing-success-falls-survey-488049)

A survey found marketers rating their content marketing as successful at the lowest rate in twelve years of the same study.

**Why it matters:** the cheapest explanation is AI-driven volume — everyone publishing more while the same amount of attention gets divided. It is a self-assessment, not an outcome measure, but the direction is consistent with what search results look like now.

## Marketing AI & Ads 9 stories

### Google shipped a batch of data and measurement tools for advertisers

Confirmed 10 Sep · [Google](https://blog.google/products/ads-commerce/data-strength-updates/)

![Google's new advertiser data and measurement tools — image published by Google](https://storage.googleapis.com/gweb-uniblog-publish-prod/images/Header_EnnLoZX.width-1300.png)
_Image: [Google](https://blog.google/products/ads-commerce/data-strength-updates/)_

Google announced updates aimed at first-party data strength and measurement, framed around driving profitable growth rather than volume.

**Why it matters:** every measurement release from Google in the past two years pushes the same way — more first-party data in, more modelled attribution out. Plan your data collection for that direction, not against it.

### Google launched Meridian GeoX globally, moving geo experiments out of beta

Reported 10 Sep · [Search Engine Journal](https://www.searchenginejournal.com/google-launches-meridian-geox-globally/589030/)

![Google launches Meridian GeoX globally — image published by Search Engine Journal](https://cdn.searchenginejournal.com/wp-content/uploads/2026/09/meridian_geox_1600x840_09092026-281.png)
_Image: [Search Engine Journal](https://www.searchenginejournal.com/google-launches-meridian-geox-globally/589030/)_

**Meridian GeoX** reached general availability worldwide, letting marketers run geo experiments and feed incrementality results into marketing mix modelling. Meridian also gained capabilities for modelling longer-term brand effects.

**Why it matters:** geo holdout tests are the most credible measurement available to most advertisers, and they are now free and open-source. If you have been arguing about attribution, this is the tool that ends the argument with an experiment.

### Google expanded Data Manager and added uplift measurement

Reported 10 Sep · [Search Engine Journal](https://www.searchenginejournal.com/google-expands-data-manager-and-adds-uplift-measurement/589025/)

Data Manager gained wider connectivity, and uplift measurement was added to the reporting set.

**Why it matters:** uplift answers "would this have happened anyway", which last-click attribution cannot. It is the number to move your reporting toward.

### Advertisers report Google Ads promotional credits being invalidated after they spent to qualify

Reported 11 Sep · [Search Engine Land](https://searchengineland.com/advertisers-report-google-ads-credits-being-revoked-after-spending-488330)

PPC consultant David Melamed reported two cases in quick succession. In one, an advertiser spent the required **$3,200** to earn a **$3,200** promotional credit, and the credit was marked "Invalidated" more than a month later. Melamed said the advertiser likely would not have spent that way without the offer.

**Why it matters:** if you have ever raised budget to hit a credit threshold, check whether the credit actually landed. Treat promotional credits as unconfirmed until they appear on an invoice.

### Google Ads added reporting segments for loyalty programme members

Observed 11 Sep · [Search Engine Roundtable](https://www.seroundtable.com/google-ads-loyalty-program-members-segments-42050.html)

New segmentation lets advertisers break reporting down by loyalty programme membership.

**Why it matters:** separating existing members from new customers is the difference between paying to acquire and paying to re-reach someone you already had.

### Google Ads opened a video campaign group beta

Observed 11 Sep · [Search Engine Roundtable](https://www.seroundtable.com/google-ads-video-campaign-group-42027.html)

A beta appeared allowing video campaigns to be organised into groups.

**Why it matters:** grouping is usually the precursor to shared budgets and shared bidding across campaigns, which changes how video is planned.

### Google Shopping ads are showing an "X days left" attribute on sponsored results

Observed 11 Sep · [Search Engine Roundtable](https://www.seroundtable.com/google-ads-x-days-left-42063.html)

A countdown attribute is appearing on sponsored shopping results, surfacing promotion end dates in the ad itself.

**Why it matters:** urgency is being rendered by the platform rather than written by you. Make sure your feed's promotion dates are accurate, because they are now visible.

### Google Merchant Center added product videos

Observed 11 Sep · [Search Engine Roundtable](https://www.seroundtable.com/google-merchant-center-add-product-videos-42064.html)

Merchants can now attach product videos to their feed items.

**Why it matters:** combined with [ChatGPT Shopping's shift to feed-integrated retrieval](#chatgpt-feeds), the product feed is becoming the single most valuable asset in ecommerce marketing — and it now carries video.

### TikTok rejected Meta ads urging rivals to join a child safety settlement

Reported 11 Sep · [Search Engine Land](https://searchengineland.com/tiktok-rejects-meta-ads-pushing-rivals-to-join-child-safety-settlement-488324)

TikTok declined to run Meta advertising that pressed competitors to sign on to a child safety settlement.

**Why it matters:** platforms refusing each other's advocacy advertising is a reminder that ad inventory is editorial. If your campaign takes a position, approval is not guaranteed on any network.

## Google Product News 8 stories

### Google committed €13 billion to AI infrastructure in Finland

Confirmed 9 Sep · [Google](https://blog.google/innovation-and-ai/infrastructure-and-cloud/global-network/google-ai-commitment-to-finland/)

![Google's 13 billion euro AI infrastructure commitment to Finland — image published by Google](https://storage.googleapis.com/gweb-uniblog-publish-prod/images/Google_FI_Announcement_Helsinki_2026_0556-2.width-1300.jpg)
_Image: [Google](https://blog.google/innovation-and-ai/infrastructure-and-cloud/global-network/google-ai-commitment-to-finland/)_

Google announced a **€13 billion** investment in Finnish AI infrastructure, published alongside a separate blueprint for clean energy growth in the country.

**Why it matters:** data centre siting is now an energy-policy story. Where compute gets built determines which regions get low-latency inference — and which get the power bills.

### Chrome moved to a two-week release cycle, starting with version 153

Reported 8 Sep · [TechCrunch](https://techcrunch.com/2026/09/08/chrome-is-now-shipping-updates-every-2-weeks-as-ai-changes-the-security-landscape/)

![Chrome moves to a two-week release cycle — image published by TechCrunch](https://techcrunch.com/wp-content/uploads/2026/02/google-chrome-GettyImages-2151457378.jpg?w=1024)
_Image: [TechCrunch](https://techcrunch.com/2026/09/08/chrome-is-now-shipping-updates-every-2-weeks-as-ai-changes-the-security-landscape/)_

Chrome now ships every two weeks rather than every four, a change TechCrunch attributes to AI accelerating both vulnerability discovery and exploitation.

**Why it matters:** browser testing windows just halved. If you maintain anything that breaks on Chrome updates — an extension, a canvas-heavy app, a payment flow — your regression cadence needs to match.

### Google Labs opened Dreambeans, its hyper-personalised story feed, to all US accounts

Confirmed 10 Sep · [Google](https://blog.google/innovation-and-ai/models-and-research/google-labs/dreambeans-expansion-september-2026/)

![Google Labs expands Dreambeans to all US accounts — image published by Google](https://storage.googleapis.com/gweb-uniblog-publish-prod/images/Dreambeans_social.max-1440x810.png)
_Image: [Google](https://blog.google/innovation-and-ai/models-and-research/google-labs/dreambeans-expansion-september-2026/)_

Dreambeans — daily AI-generated stories built for the individual reader — became available to every US account.

**Why it matters:** a fully generated content feed from Google, with no publisher in it at all. Worth watching as a signal of where Discover-style surfaces could go.

### DeepMind released AlphaGenome Atlas, a high-resolution map of human DNA

Confirmed 8 Sep · [Google DeepMind](https://blog.google/innovation-and-ai/models-and-research/google-deepmind/alphagenome-atlas/)

![DeepMind releases AlphaGenome Atlas — image published by Google](https://storage.googleapis.com/gweb-uniblog-publish-prod/images/AlphaGenome_Atlas_herosocial.width-1300.png)
_Image: [Google](https://blog.google/innovation-and-ai/models-and-research/google-deepmind/alphagenome-atlas/)_

AlphaGenome Atlas maps the human genome at high resolution, extending the AlphaFold line of work into DNA regulation.

**Why it matters:** not a marketing story, but the clearest current example of AI producing scientific output rather than text about science.

### Google Search added live game feeds and fantasy football advice for the NFL season

Confirmed 9 Sep · [Google](https://blog.google/products-and-platforms/products/search/football-features-google-search/)

![Google Search adds football features for the NFL season — image published by Google](https://storage.googleapis.com/gweb-uniblog-publish-prod/images/Football_on_Search_blog_header.width-1300.png)
_Image: [Google](https://blog.google/products-and-platforms/products/search/football-features-google-search/)_

Search gained Live Game Feeds and fantasy football advice as the NFL season opened. Google also published race-training features for runners on 10 September.

**Why it matters:** every seasonal feature Google builds into the results page is a category of query that stops reaching publishers. Sports and fitness content sites should assume this pattern continues.

### Google's revived nuclear plant secured a $1.9 billion US government loan

Reported 8 Sep · [TechCrunch](https://techcrunch.com/2026/09/08/googles-revived-nuclear-power-plant-gets-1-9b-loan-from-us-government/)

The nuclear plant Google is restarting to power data centres received a **$1.9 billion** federal loan.

**Why it matters:** AI infrastructure is now drawing sovereign-scale energy financing. That is the real constraint on model scaling, and it is being solved with public money.

### Google Cloud signed an Accenture deal to catch up in AI deployment

Reported 8 Sep · [TechCrunch](https://techcrunch.com/2026/09/08/google-cloud-races-to-catch-up-in-the-ai-deployment-wars-with-accenture-deal/)

Google Cloud partnered with Accenture to accelerate enterprise AI deployments, an area where it trails Microsoft and AWS.

**Why it matters:** enterprise AI is won by systems integrators, not by benchmarks. Whoever the consultancies are trained on becomes the default in large organisations.

### Google Photos cut its trash retention from 60 days to 30

Observed 8 Sep · [9to5Google](https://9to5google.com/2026/09/08/google-photos-trash-bin-now-deletes-media-after-30-days-down-from-60-days/)

Deleted media now leaves the Google Photos bin after **30 days**, halved from 60.

**Why it matters:** a quiet halving of your recovery window. If you rely on the bin as an undo, you now have half the time to notice.

## Google Analytics 2 stories

### Google Analytics launched Dashboards, a drag-and-drop canvas for KPI reporting

Confirmed 9 Sep · [Google Analytics release notes](https://support.google.com/analytics/answer/9164320) · [Search Engine Land](https://searchengineland.com/google-analytics-launches-customizable-dashboards-488343)

![Dashboards available in Google Analytics — image published by Search Engine Roundtable](https://images.seroundtable.com/google-analytics-explode-hero-FVUHcKAC.jpg)
_Image: [Search Engine Roundtable](https://www.seroundtable.com/google-analytics-dashboards-42075.html)_

The 9 September release notes confirm Dashboards: a grid-based canvas where you drag dimensions and metrics on directly, then position, resize and align visualisations by hand. **Six visualisation types** are available at launch, including scorecards and tables, and a dashboard can be published straight into the Reports navigation without going through the library.

**Why it matters:** this is the feature GA4 has been missing since the Universal Analytics migration — a single screen you can put in front of a client without exporting to Looker Studio first.

### Context: custom conversion windows landed in August and are still worth configuring

Confirmed 11 Aug · [Google Analytics release notes](https://support.google.com/analytics/answer/9164320)

Outside the window, but adjacent to Dashboards and under-adopted: since 11 August, engaged-view conversion windows accept any integer from **1 to 30 days** (previously fixed at 3) and click-through windows any integer from **1 to 90** (previously preset values only). Configure under Advertising > Conversion management > Settings.

**Why it matters:** default attribution windows rarely match a real sales cycle. This is a five-minute change that makes every conversion report more honest.

## Search Console 1 story

### Search Console's Indexing report is missing several days of June 2026 data, permanently

Confirmed 11 Sep · [Search Engine Land](https://searchengineland.com/google-search-console-indexing-report-missing-june-data-488261)

![Google Search Console indexing report missing old June data — image published by Search Engine Roundtable](https://images.seroundtable.com/google-data-chart-fire-broken-axmksapV.jpg)
_Image: [Search Engine Roundtable](https://www.seroundtable.com/google-search-console-indexing-report-missing-data-42073.html)_

Page indexing reports show gaps across several June 2026 days, affecting every property. John Mueller confirmed the cause: "This is likely from the time in June where the data was delayed — there just isn't any data for that period, the page indexing report wasn't updated. **We don't back-fill indexing data.**"

**Why it matters:** it is a reporting gap, not an indexing problem — do not panic, and do not let a client panic either. But it is permanent, so if June is in a reporting period you are comparing against, annotate it now.

Nothing else shipped in Search Console this week — the Search Central Blog's only post was the Latin America events announcement, and there were no new reports or report changes.

## Meta 4 stories

### Meta launched Muse, a personal AI agent that connects to your email, calendar and payments

Confirmed 8 Sep · [Meta Newsroom](https://about.fb.com/news/) · [TechCrunch](https://techcrunch.com/2026/09/08/meta-debuts-its-muse-ai-agent-will-consumers-trust-it/)

![Meta launches Muse, its personal AI agent — image published by TechCrunch](https://techcrunch.com/wp-content/uploads/2026/09/Muse-Neswsroom-Banner.jpg?resize=1200,675)
_Image: [TechCrunch](https://techcrunch.com/2026/09/08/meta-debuts-its-muse-ai-agent-will-consumers-trust-it/)_

Muse launched in the US on 8 September as a personal agent that plugs into email, calendars, payments, and apps for health, smart home, dining, shopping, music and events. Meta says it can send emails, book travel, lower bills, fill out forms, build plans, turn recipe reels into grocery lists and send party invitations. It arrived **less than two weeks after Meta agreed an $18 billion multistate settlement** over social media consumer harms.

**Why it matters:** this asks for more personal data than any Meta product ever has, from the company with the least accumulated trust to spend. Whether people grant it is the most interesting consumer-AI question of the quarter.

### Muse became the No. 2 app in the US within two days

Reported 10 Sep · [TechCrunch](https://techcrunch.com/2026/09/10/metas-ai-agent-muse-is-now-the-no-2-app-in-the-us/)

![Meta's AI agent Muse reaches No. 2 in the US app charts — image published by TechCrunch](https://techcrunch.com/wp-content/uploads/2026/09/meta-muse-app-GettyImages-2293792137.jpg?w=1024)
_Image: [TechCrunch](https://techcrunch.com/2026/09/10/metas-ai-agent-muse-is-now-the-no-2-app-in-the-us/)_

Two days after launch, Muse was the **No. 2** app in the US charts.

**Why it matters:** it answers the trust question faster than anyone expected — at least for downloads. Retention is the number that will actually matter, and it is a month away.

### Instagram's head said a chronological feed is "much worse"

Reported 11 Sep · [9to5Google](https://9to5google.com/2026/09/11/instagram-chronological-feed-much-worse/)

Adam Mosseri argued publicly that a purely chronological feed produces a worse experience — 9to5Google's read was that he said the quiet part out loud.

**Why it matters:** the ranked-feed debate is settled inside the company, whatever regulators think. For anyone posting organically, reach remains an algorithmic allocation, not a function of posting time.

### Meta ran ads pressing rivals to join a child safety settlement; TikTok refused them

Reported 11 Sep · [Search Engine Land](https://searchengineland.com/tiktok-rejects-meta-ads-pushing-rivals-to-join-child-safety-settlement-488324)

Meta bought advertising urging competitors to sign a child safety settlement. TikTok declined to carry it.

**Why it matters:** Meta is using paid media as a policy instrument against competitors. Expect more of this as AI and child-safety regulation converge.

## WhatsApp 6 stories

### WhatsApp is testing guest calls for people without an account

Observed 6 Sep · [WABetaInfo](https://wabetainfo.com/whatsapp-is-testing-guest-calls-for-people-without-an-account/)

![WhatsApp testing guest calls for people without an account — image published by WABetaInfo](https://wabetainfo.com/wp-content/uploads/2026/09/WA_GUEST_CALLS_WITHOUT_ACCOUNT_FEATURE_WEB.webp)
_Image: [WABetaInfo](https://wabetainfo.com/whatsapp-is-testing-guest-calls-for-people-without-an-account/)_

A beta feature lets someone without a WhatsApp account join a call via a link.

**Why it matters:** it removes the last requirement that everyone on a call be a WhatsApp user — which matters commercially in markets like India where WhatsApp is the default business channel.

### WhatsApp ended support for older Android versions

Reported 7 Sep · [WABetaInfo](https://wabetainfo.com/whatsapp-ends-support-for-older-android-versions-today/)

Support ended for older Android releases on 7 September.

**Why it matters:** in markets with long device lifecycles, a support cut-off strands real customers. If WhatsApp is a support channel for you, expect a small cohort to go quiet.

### WhatsApp rolled out its full Liquid Glass interface broadly

Observed 9 Sep · [WABetaInfo](https://wabetainfo.com/whatsapp-is-widely-rolling-out-the-full-liquid-glass-experience/)

![WhatsApp rolling out the full Liquid Glass experience on iOS — image published by WABetaInfo](https://wabetainfo.com/wp-content/uploads/2026/09/WA_FULL_LIQUID_GLASS_CHAT_INTERFACE_EXPERIENCE_IOS.webp)
_Image: [WABetaInfo](https://wabetainfo.com/whatsapp-is-widely-rolling-out-the-full-liquid-glass-experience/)_

The full Liquid Glass chat interface is now reaching users broadly on iOS.

**Why it matters:** if you produce WhatsApp creative or documentation with screenshots, they are about to look dated.

### WhatsApp is rolling out a redesigned navigation interface for iPad

Observed 8 Sep · [WABetaInfo](https://wabetainfo.com/whatsapp-is-rolling-out-a-redesigned-navigation-interface-for-ipad/)

The iPad app received a reworked navigation layout.

**Why it matters:** WhatsApp on tablets is increasingly where small businesses run support. Layout changes move where staff click.

### WhatsApp is rolling out media thumbnail previews in the chat list

Observed 8 Sep · [WABetaInfo](https://wabetainfo.com/whatsapp-is-rolling-out-media-thumbnail-previews-in-the-chat-list/)

Chats now show a thumbnail of the most recent media directly in the list.

**Why it matters:** the first image in a broadcast now has a preview slot in the chat list. That is a small but real creative consideration for WhatsApp marketing.

### WhatsApp is rolling out longer music clips for status updates

Observed 7 Sep · [WABetaInfo](https://wabetainfo.com/whatsapp-is-rolling-out-longer-music-clips-for-status-updates/)

Status updates can carry longer music clips than before.

**Why it matters:** Status continues to be rebuilt as a Stories competitor. For brands in India and Brazil, it is an underused organic surface.

## Claude 4 stories

### Anthropic accused Alibaba, Moonshot AI and DeepSeek of running distillation campaigns against Claude

Confirmed 10 Sep · [Anthropic](https://www.anthropic.com/threat-intelligence-report-september-2026) · [TechCrunch](https://techcrunch.com/2026/09/10/anthropic-details-distillation-campaigns-from-alibaba-moonshot-ai-and-deepseek/)

![Anthropic's September 2026 threat intelligence report — image published by Anthropic](https://cdn.sanity.io/images/4zrzovbb/website/7a4426f8ffe57e7de23ff36906fb1cc3efe2a82b-1200x630.jpg)
_Image: [Anthropic](https://www.anthropic.com/threat-intelligence-report-september-2026)_

Anthropic's September threat report covers activity disrupted between **December 2025 and August 2026** across seven harm areas: cyber operations, influence operations, surveillance, scams and fraud, biological misuse, conventional weapons development, and distillation. On distillation it states that "unauthorized labs have developed increasingly sophisticated methods to circumvent our defenses and harvest the capabilities of US frontier models", targeting Claude's agentic and tool-use capabilities, coding and data analysis, and logical reasoning. Claude Haiku, Sonnet and Opus models were involved; no Fable or Mythos-class models except in one distillation case.

**Why it matters:** the first time a frontier lab has named specific competitors in a published threat report. It moves distillation from an open secret to a documented allegation — and sets up the export-control and terms-of-service fights that follow.

### Anthropic found that rogue AI agents get stuck on CAPTCHAs

Reported 10 Sep · [TechCrunch](https://techcrunch.com/2026/09/10/anthropic-reveals-rogue-ai-agents-hate-captchas-just-like-you/)

A detail from the same threat report: malicious agentic operations were frequently blocked by CAPTCHAs.

**Why it matters:** a genuinely useful finding for site owners. The cheapest existing defence still works against the current generation of malicious agents — while also blocking the legitimate ones your customers may soon be sending.

### Hackers are draining Claude subscribers' token allowances through stolen sessions

Reported 8 Sep · [TechCrunch](https://techcrunch.com/2026/09/08/hackers-are-stealing-claude-tokens-from-subscribers/)

![Hackers are stealing Claude tokens from subscribers — image published by TechCrunch](https://techcrunch.com/wp-content/uploads/2026/06/GettyImages-2278736523.jpeg?resize=1200,798)
_Image: [TechCrunch](https://techcrunch.com/2026/09/08/hackers-are-stealing-claude-tokens-from-subscribers/)_

Independent AI consultant Grant De Swardt watched his Claude Max 20x token usage climb from **45% to 55%** during a controlled interval in which he did no work, with scheduled tasks paused, cloud execution disabled and no active local session.

**Why it matters:** a concrete follow-on from the browser-session hijacking Anthropic warned about in [issue #1](https://thejayant.in/blog/weekly/2026-09-05). If your usage graph has unexplained climbs, that is now a known attack pattern rather than a billing mystery.

### Anthropic argued in public for pacing frontier development

Reported 12 Sep · [TechCrunch](https://techcrunch.com/2026/09/12/anthropic-ceo-outlines-plan-to-pace-the-frontier/)

Covered in full [above](#pace-the-frontier): Dario Amodei's case for controlling the speed of frontier development, published three days after a researcher resigned over self-improving AI.

**Why it matters:** Anthropic spent the week both accusing competitors of stealing its capabilities and arguing everyone should slow down. Those positions are compatible, but they are not a coincidence.

## ChatGPT 5 stories

### OpenAI launched ChatGPT Images 2.5 with four new creative tools

Confirmed 8 Sep · [OpenAI](https://openai.com/index/introducing-chatgpt-images-2-5/)

![OpenAI's announcement artwork for ChatGPT Images 2.5 — image published by OpenAI](https://images.ctfassets.net/kftzwdyauwt9/6J668sj93QnQ8PWP7epUDv/3dbd2d2350053d96b03477d0fdb2e2c6/images2point5_16-9c.png?w=1200&q=85&fm=webp)
_Image: [OpenAI](https://openai.com/index/introducing-chatgpt-images-2-5/)_

Images 2.5 shipped on 8 September with up to **50% lower generation latency** than Images 2.0, better reference-photo fidelity, and four new tools — Sketch, Templates, Comments and prompt sharing. Two API models came with it: GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst.

**Why it matters:** the tools matter more than the model — they move image generation from describing to directing. I have written the full breakdown with OpenAI's own samples in [ChatGPT Images 2.5: everything new](https://thejayant.in/blog/chatgpt-images-2-5).

### OpenAI positioned GPT-6 Astra as the next generation of intelligence for work

Confirmed 9 Sep · [OpenAI](https://openai.com/index/gpt-6-astra-next-generation-work/)

A follow-up to the 3 September Astra launch, framing the model specifically around professional work.

**Why it matters:** the positioning shift from "most capable model" to "model for work" is how pricing gets justified. My [full Astra breakdown](https://thejayant.in/blog/gpt-6-astra) and the [benchmark caveats](https://thejayant.in/blog/gpt-6-astra-benchmarks) cover what is actually behind it.

### OpenAI introduced ChatGPT for Financial Services

Confirmed 10 Sep · [OpenAI](https://openai.com/index/introducing-chatgpt-financial-services/)

![OpenAI introduces ChatGPT for Financial Services — image published by OpenAI](https://images.ctfassets.net/kftzwdyauwt9/2Tj0j0dTcedAZ2PMSbTg4j/f1155b4361dcd8f4c85c12833233ebba/OAI_Finances_4_16x9-_SEO.png?w=1200&h=675&fit=fill)
_Image: [OpenAI](https://openai.com/index/introducing-chatgpt-financial-services/)_

A vertical package aimed at financial services, launched alongside the Agents API and Data agent on 10 September.

**Why it matters:** OpenAI is going after regulated industries directly, where Harvey and others built businesses on serving one profession properly. Horizontal is coming for vertical.

### OpenAI published how it scaled storage for over 1 billion ChatGPT users

Confirmed 11 Sep · [OpenAI](https://openai.com/index/scaling-storage-one-billion-users-part-one/)

An engineering post on scaling online storage to serve over **1 billion** ChatGPT users.

**Why it matters:** the user figure is the story. One billion users is search-engine scale, which is exactly why the EU designated ChatGPT a Very Large Online Search Engine — covered in [issue #1](https://thejayant.in/blog/weekly/2026-09-05).

### Pro plan sign-ups were disabled mid-week

Reported 10 Sep · [TechCrunch](https://techcrunch.com/2026/09/10/openai-puts-pro-subscriptions-on-hold-due-to-astra-demand/)

Detailed [above](#openai-pro-pause) — the $200 tier closed to new customers while API, Go and Plus stayed open.

**Why it matters:** if you were planning to put a team on Pro this month, check availability before you budget for it.

## Gemini 4 stories

### The Gemini app launched on Windows with an Alt + Space shortcut

Confirmed 10 Sep · [Google](https://blog.google/innovation-and-ai/products/gemini-app/gemini-app-now-on-windows/)

![The Gemini app is now available for Windows — image published by Google](https://storage.googleapis.com/gweb-uniblog-publish-prod/images/Gemini_for_Desktop_Thumbnail_2096x1182.width-1300.png)
_Image: [Google](https://blog.google/innovation-and-ai/products/gemini-app/gemini-app-now-on-windows/)_

Gemini now installs as a Windows desktop app, summoned with **Alt + Space**, able to draft documents, manage tasks, create images and pull information from Google apps without leaving the current window.

**Why it matters:** a keyboard shortcut on the OS is the most valuable real estate in consumer AI, and Google has just claimed it on Windows — the platform Microsoft owns.

### Gemini's Daily Brief became free in the US

Observed 8 Sep · [9to5Google](https://9to5google.com/2026/09/08/gemini-daily-brief-free/)

![Gemini app makes Daily Brief free in the US — image published by 9to5Google](https://9to5google.com/wp-content/uploads/sites/4/2026/09/Gemini-Daily-Brief-cover.jpg?quality=82&strip=all&resize=1200,628)
_Image: [9to5Google](https://9to5google.com/2026/09/08/gemini-daily-brief-free/)_

The personalised Daily Brief, previously a paid feature, opened to free US users.

**Why it matters:** another daily summary surface that answers before a publisher gets a click. Same pattern as Dreambeans, different entry point.

### Google unified the Gemini side panels across Workspace

Observed 9 Sep · [9to5Google](https://9to5google.com/2026/09/09/gemini-workspace-side-panel-upgrade/)

Every Gemini side panel in Workspace now carries the same feature set, rather than differing by app.

**Why it matters:** the inconsistency was the main reason Workspace users never learned what Gemini could do. Uniform capability makes it teachable.

### Google updated its AI plans with new to-do features

Confirmed 9 Sep · [Google](https://blog.google/products-and-platforms/products/google-one/fall-2026-ai-plan-updates/)

Google's autumn AI plan update added task-management features across its subscription tiers.

**Why it matters:** the AI subscription bundle keeps absorbing adjacent categories. Standalone to-do apps are the current casualty.

## Codex 4 stories

### Codex CLI 0.154.0 added "max" and "ultra" reasoning-effort levels

Confirmed 9 Sep · [Codex changelog](https://developers.openai.com/codex/changelog/)

Release 0.154.0 introduced `max` and `ultra` reasoning-effort values, plus `ExternalMessage` support in synchronous and asynchronous `run()`. A matching Python SDK 0.154.0 followed on 10 September.

**Why it matters:** effort levels are the cost dial on agentic coding. Two new tiers at the top means longer, more expensive runs are now a deliberate choice rather than an accident.

### ChatGPT desktop gained Quick Chat from the floating Pets controls, and Appshots on Windows

Confirmed 11 Sep · [Codex changelog](https://developers.openai.com/codex/changelog/)

Build 26.908 lets you type a quick chat from the floating Pets controls on macOS and Windows — `@` for context, `$` for a skill, **Option+Space** on macOS or **Windows+Alt+P** on Windows. Appshots came to Windows: press both Alt keys to share the frontmost app window, screenshot plus available text, with ChatGPT.

**Why it matters:** same land-grab as the Gemini Windows app, from the other direction. Whoever owns the global shortcut owns the habit.

### OpenAI published how GPT-5.6 Sol helps run quantum computing experiments

Confirmed 8 Sep · [OpenAI](https://openai.com/index/codex-quantum-computing-experiments/)

A case study on Codex and GPT-5.6 Sol operating quantum computing experiments.

**Why it matters:** coding agents are moving from writing software to operating instruments. That is a different risk profile and a different value proposition.

### OpenAI published a researcher's use of Codex to search for new antimicrobial molecules

Confirmed 10 Sep · [OpenAI](https://openai.com/index/using-codex-chatgpt-to-search-for-new-antimicrobials/)

A case study covering Codex and ChatGPT applied to antimicrobial discovery.

**Why it matters:** alongside DeepMind's AlphaGenome Atlas the same week, both major labs are now publishing science output rather than benchmark scores. That is the pitch to governments.

## Claude Code 3 stories

### Claude Code 2.1.269 added `claude plugin eval` for scored, reproducible plugin testing

Confirmed 11 Sep · [Claude Code CHANGELOG](https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md)

The 11 September release added `claude plugin eval`, which runs a plugin's eval suite and produces scored, reproducible results as JSON and an HTML report. The same release added `/output-style` for listing and switching output styles, including in headless and cloud sessions, and a Bash-tool setting that returns a diff of the files a command changed.

**Why it matters:** plugins were untestable, which meant nobody could tell a good one from a confident one. A scored eval suite is what turns a plugin ecosystem into something an organisation can adopt deliberately.

### Six Claude Code releases shipped between 6 and 11 September

Confirmed 11 Sep · [npm release history](https://www.npmjs.com/package/@anthropic-ai/claude-code)

Versions **2.1.263 through 2.1.269** shipped across the window — roughly one a day. Notable additions beyond plugin eval: a `maxEffortLevel` setting that caps effort across every provider including Bedrock, Vertex and Foundry; `--json` output for the plugin install, uninstall, update, enable and disable commands; and browser-tab icons for published artifacts.

**Why it matters:** `maxEffortLevel` is the one to know about if you manage spend — it is an administrative ceiling on how expensive any single run can get.

### The Claude apps gateway gained pricing propagation so /cost matches the spend meter

Confirmed 10 Sep · [Claude Code CHANGELOG](https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md)

With `pricing:` set in `gateway.yaml`, signed-in clients now receive the same rates through managed settings, so `/cost` and telemetry agree with the actual spend meter. The release also added startup warnings when a gateway's `access_control.allow_cidrs` is empty.

**Why it matters:** self-hosted gateway users have been reporting cost figures that did not match their bills. This closes that gap, which matters if you are charging clients for agent time.

## Mobile AI 7 stories

### Apple unveiled the iPhone Duo, its first foldable, at a 9 September event

Reported 9 Sep · [TechCrunch](https://techcrunch.com/2026/09/09/apple-unveils-its-first-foldable-the-iphone-duo/)

![Apple unveils its first foldable, the iPhone Duo — image published by TechCrunch](https://techcrunch.com/wp-content/uploads/2026/09/iPhone-duo.jpeg?resize=1200,675)
_Image: [TechCrunch](https://techcrunch.com/2026/09/09/apple-unveils-its-first-foldable-the-iphone-duo/)_

Announced at Apple's "Surprise and Shine" event in Cupertino, the iPhone Duo is the first marquee device under new CEO John Ternus. It uses a wide form factor — a **7.6-inch** inner display, **5.4-inch** outer — with an under-display camera, a custom nanotexture finish to hide the crease, and a grade 5 aluminium hinge built from over **100** components. Notably it ships with **Touch ID only, no Face ID**, though it can be unlocked with an Apple Watch.

**Why it matters:** a new dominant screen aspect ratio, from the vendor whose users convert best. If you have not tested your site at a 7.6-inch tablet-ish ratio, that is now a real audience.

### Apple's Watch Series 12 and Ultra 4 can listen to your conversations and recap them

Reported 9 Sep · [TechCrunch](https://techcrunch.com/2026/09/09/apple-watchs-new-feature-listens-to-your-chats-and-recaps-them/)

The new Watches add an AI feature that listens to conversations and produces a recap of the day. TechCrunch's follow-up argued the feature normalises the idea that technology is always listening.

**Why it matters:** ambient recording on a mainstream wrist device is a consent problem for everyone in the room, not just the wearer. Expect meeting-room policies to become a real question in enterprises.

### Apple added a way to prove iPhone photos are not AI-generated

Reported 9 Sep · [TechCrunch](https://techcrunch.com/2026/09/09/apple-has-a-new-way-prove-your-iphone-photos-arent-ai-slop/)

![Apple's new way to prove iPhone photos are not AI-generated — image published by TechCrunch](https://techcrunch.com/wp-content/uploads/2026/09/Apple-iPhone-18-Pro-Photos-app-Apple-Reference-Image-260909_big.jpg.large_2x.jpg?resize=1200,800)
_Image: [TechCrunch](https://techcrunch.com/2026/09/09/apple-has-a-new-way-prove-your-iphone-photos-arent-ai-slop/)_

The iPhone 18 Pro Photos app gained a mechanism for attesting that an image came from the camera rather than a model.

**Why it matters:** provenance is being attacked from the opposite end. Instead of marking what is generated — which [strips easily](https://thejayant.in/blog/chatgpt-images-2-5) — Apple is marking what is real, from a device you already trust.

### Apple says Siri handles 2.5 billion requests every day

Reported 10 Sep · [Search Engine Roundtable](https://www.seroundtable.com/apple-siri-2-5-billion-requests-daily-42060.html)

Apple disclosed a figure of **2.5 billion** Siri requests per day.

**Why it matters:** that is query volume at search-engine scale on a surface with no SEO discipline attached to it at all. Worth holding next to the 1 billion ChatGPT users figure from the same week.

### The Google Pixel Tablet has been discontinued

Reported 9 Sep · [9to5Google](https://9to5google.com/2026/09/09/google-pixel-tablet-discontinued/)

![The Google Pixel Tablet has been discontinued — image published by 9to5Google](https://9to5google.com/wp-content/uploads/sites/4/2026/01/Pixel-Tablet-on-Dock.jpg?quality=82&strip=all&resize=1200,628)
_Image: [9to5Google](https://9to5google.com/2026/09/09/google-pixel-tablet-discontinued/)_

The Pixel Tablet is out of stock everywhere and has been discontinued.

**Why it matters:** Google has again exited a hardware category rather than iterating. Android tablet share consolidates further around Samsung.

### Android began rolling out passkey transfers between password managers

Confirmed 10 Sep · [9to5Google](https://9to5google.com/2026/09/10/android-passkey-password-transfer/)

![Android rolling out passkey transfers between password managers — image published by 9to5Google](https://9to5google.com/wp-content/uploads/sites/4/2026/09/Android-passkey-password-transfer.jpg?quality=82&strip=all&resize=1200,628)
_Image: [9to5Google](https://9to5google.com/2026/09/10/android-passkey-password-transfer/)_

Passkeys can now move between password managers on Android, with Google publishing a guide to switching managers the same week.

**Why it matters:** passkey lock-in was the main reason people avoided adopting them. Portability removes the strongest objection to turning them on for your users.

### Arm announced new CPU cores and GPUs promising desktop-class gaming and more on-device AI

Reported 8 Sep · [9to5Google](https://9to5google.com/2026/09/08/arms-new-cpu-cores-and-gpus-for-android-chips-promise-desktop-class-games-more-ai-power/)

![Arm's new CPU cores and GPUs for Android chips — image published by 9to5Google](https://9to5google.com/wp-content/uploads/sites/4/2026/09/arm-c2-mali-g2-header.jpg?quality=82&strip=all&resize=1200,628)
_Image: [9to5Google](https://9to5google.com/2026/09/08/arms-new-cpu-cores-and-gpus-for-android-chips-promise-desktop-class-games-more-ai-power/)_

Arm's new C2 cores and Mali G2 GPUs target desktop-class games and increased on-device AI capability for Android chips.

**Why it matters:** more on-device inference means fewer round trips to a server — and fewer signals in your analytics. Plan for measurement getting harder, not easier.

## Other major releases 5 stories

### Microsoft patched a record 972 vulnerabilities, 112 of them critical

Reported 8 Sep · [Ars Technica](https://arstechnica.com/security/2026/09/microsoft-patches-a-record-972-vulnerabilities-112-of-them-critical/)

September's Patch Tuesday covered a record **972** vulnerabilities, **112** rated critical.

**Why it matters:** patch volume at this scale is the other side of the [Chrome two-week cycle](#chrome-2week) — AI is finding bugs faster than release trains were designed to ship fixes. Patch windows are becoming a continuous process.

### ClickFix attacks infecting both PCs and Macs are spreading rapidly

Reported 11 Sep · [Ars Technica](https://arstechnica.com/security/2026/09/clickfix-attacks-infecting-pcs-and-macs-are-going-viral/)

The ClickFix technique — tricking users into pasting a malicious command into their own terminal or run dialog — is spreading across both platforms.

**Why it matters:** it bypasses every download protection by making the victim the delivery mechanism. Worth a line in your team's security briefing, because technical staff are the target.

### Four groups were caught using the same Chrome and Windows exploit kit

Reported 9 Sep · [Ars Technica](https://arstechnica.com/information-technology/2026/09/4-groups-caught-using-the-same-chrome-and-windows-exploit-kit/)

Four separate threat groups were found sharing one exploit kit targeting Chrome and Windows.

**Why it matters:** shared tooling means a single patch protects against four actors at once — which is the argument for patching on the day rather than the quarter.

### Automattic's board forced Matt Mullenweg into a leave of absence

Reported 9 Sep · [TechCrunch](https://techcrunch.com/2026/09/09/automattics-board-forces-ceo-matt-mullenweg-into-leave-of-absence/) · [Search Engine Journal](https://www.searchenginejournal.com/wordpress-co-founder-matt-mullenweg-ousted-from-automattic/589009/)

![Automattic's board forces CEO Matt Mullenweg into a leave of absence — image published by TechCrunch](https://techcrunch.com/wp-content/uploads/2024/10/matt-mullenweg-bling-watch.jpg?resize=1200,799)
_Image: [TechCrunch](https://techcrunch.com/2026/09/09/automattics-board-forces-ceo-matt-mullenweg-into-leave-of-absence/)_

The WordPress co-founder was placed on leave by Automattic's board.

**Why it matters:** WordPress still runs a large share of the web, and its governance has been contested for two years. Anyone with client sites on WordPress should be watching who ends up controlling the project's direction.

### A hacker stole $340 million in a crypto heist, then returned most of it

Reported 8 Sep · [TechCrunch](https://techcrunch.com/2026/09/08/a-hacker-stole-340m-in-a-crypto-heist-then-returned-most-of-it/)

A **$340 million** theft was followed by the return of most of the funds.

**Why it matters:** included because the pattern — steal, negotiate, return for a bounty — is now a recognised business model rather than an anomaly.

## Off the clock: gaming 7 stories

### Nintendo held a Zelda 40th Anniversary Direct and revealed Ocarina of Time remake gameplay

Reported 8 Sep · [Eurogamer](https://www.eurogamer.net/everything-announced-legend-of-zelda-40th-anniversary-nintendo-direct)

![Everything announced at Nintendo's Legend of Zelda 40th Anniversary Direct — image published by Eurogamer](https://assetsio.gnwcdn.com/zelda-direct-link-scream-lower-header.jpg?width=1200&height=630&fit=crop&enable=upscale&auto=webp)
_Image: [Eurogamer](https://www.eurogamer.net/everything-announced-legend-of-zelda-40th-anniversary-nintendo-direct)_

The first of two Directs planned for the week included a full gameplay reveal for the Ocarina of Time Switch 2 remake, which drew a mixed reception — "Dawg it's Unreal Engine Zelda" was the pithiest version. The live-action film, it turns out, is simply called _The Legend of Zelda_.

**Why it matters:** forty years of one series, and the internet still cannot agree on whether a remake should look like memory or like hardware.

### Almost 1,900 Blizzard workers ratified a union contract with protections against layoffs and generative AI

Reported 9 Sep · [Eurogamer](https://www.eurogamer.net/1900-blizzard-vote-historic-contract-union-layoffs-ai)

![Blizzard workers ratify a historic union contract — image published by Eurogamer](https://assetsio.gnwcdn.com/overwatch-new-heroes.jpg?width=1200&height=630&fit=crop&enable=upscale&auto=webp)
_Image: [Eurogamer](https://www.eurogamer.net/1900-blizzard-vote-historic-contract-union-layoffs-ai)_

Around **1,900** unionised Blizzard workers ratified their first contract after years of negotiation. Blizzard must now bargain with staff over the use of generative AI in game development, and workers gained stronger layoff protections.

**Why it matters:** the first major contract in which generative AI use is a mandatory bargaining subject. It is a template every creative industry union will now reach for.

### Grand Theft Auto 6 worldwide unlock times were revealed

Reported 8 Sep · [Eurogamer](https://www.eurogamer.net/gta-6-worldwide-release-unlock-times-ps5-xbox-series-x-s)

![GTA 6 worldwide release unlock times revealed — image published by Eurogamer](https://assetsio.gnwcdn.com/nz-gta.jpg?width=1200&height=630&fit=crop&enable=upscale&auto=webp)
_Image: [Eurogamer](https://www.eurogamer.net/gta-6-worldwide-release-unlock-times-ps5-xbox-series-x-s)_

Unlock times went out for PS5 and Xbox Series X|S, and it appears Xbox owners can still use the old New Zealand region trick to start early. Separately, the final hearing in the Rockstar unfair-dismissal tribunal began this week.

**Why it matters:** if you want the detail, I have written far too much about it: [everything we know](https://thejayant.in/blog/gta-6-everything-we-know), [whether it uses AI](https://thejayant.in/blog/does-gta-6-use-ai), and [how smart the NPCs could be](https://thejayant.in/blog/gta-6-npc-ai).

### Experts tested Nvidia's DLSS 5 and found it really does tank frame rates

Reported 7 Sep · [Eurogamer](https://www.eurogamer.net/nvidia-dlss-5-tested-frame-rates)

![Experts test Nvidia's DLSS 5 and find it tanks frame rates — image published by Eurogamer](https://assetsio.gnwcdn.com/nba-2k7-dlss-5.jpg?width=1200&height=630&fit=crop&enable=upscale&auto=webp)
_Image: [Eurogamer](https://www.eurogamer.net/nvidia-dlss-5-tested-frame-rates)_

Independent testing of Nvidia's graphics-altering DLSS 5 confirmed a heavy frame-rate cost. "This is super, super heavy" was the verdict.

**Why it matters:** a rare case of independent benchmarking arriving fast enough to matter before purchase decisions. The same discipline the [AI benchmark world could use](https://thejayant.in/blog/gpt-6-astra-benchmarks).

### China's games industry is booming while the West struggles, a new report shows

Reported 7 Sep · [Eurogamer](https://www.eurogamer.net/china-games-market-50-billion-dollars-value)

![China's games market reaches 50 billion dollars in value — image published by Eurogamer](https://assetsio.gnwcdn.com/phantom-blade-zero-blog_header.png?width=1200&height=630&fit=crop&enable=upscale&auto=webp)
_Image: [Eurogamer](https://www.eurogamer.net/china-games-market-50-billion-dollars-value)_

China's games market reached **$50 billion** in value as Western studios continue to contract.

**Why it matters:** same week Chinese AI labs were accused of distilling US frontier models and Moonshot set a $2B revenue target. The centre of gravity in two different technology industries is moving in the same direction.

### The Blood of Dawnwalker sold a million copies in three days

Reported 6 Sep · [Eurogamer](https://www.eurogamer.net/blood-of-dawnwalker-1-million-sales-rebel-wolves)

Rebel Wolves' debut sold **1 million** copies in three days. Its most popular mod, amusingly, nullifies the game's main hook.

**Why it matters:** a new studio from ex-CD Projekt developers hitting a million in a weekend is the clearest counter-argument to "the games industry is finished".

### Saber Interactive said it will not change its communications strategy after an AI writer controversy

Reported 6 Sep · [Eurogamer](https://www.eurogamer.net/saber-interactive-comms-strategy-ai-writer-controversy-willits)

Saber's chief creative officer confirmed no change to how the studio communicates, following backlash over AI-written material.

**Why it matters:** read alongside the [Blizzard contract](#blizzard-union) in the same week — one studio negotiating AI use with its workers, another declining to change course after a public reaction. Both are viable positions right now, which is the point.

## FAQ

### What were the biggest AI news stories of the week of 6–12 September 2026?

Five stood out. OpenAI launched the Agents API in public beta along with GPT-Live-1 voice in the API and a Data agent for ChatGPT Work. Apple unveiled its first foldable, the iPhone Duo, plus an Apple Watch that recaps overheard conversations. Meta launched Muse, a personal AI agent that reached No. 2 in the US app charts within two days. Anthropic published a threat report accusing Alibaba, Moonshot AI and DeepSeek of distilling Claude. And Google Analytics launched Dashboards.

### What is OpenAI's Agents API?

The Agents API, launched in public beta on 10 September 2026, gives developers the same harness and infrastructure that powers Codex: context management, efficient tool use, subagent coordination, and environments that keep agents running reliably for days with file access, code execution and saved intermediate results. One early customer, Ciridae, reported its evaluation score rising from 0.71 to 0.85 and a 4× latency reduction from the built-in subagent support.

### Did Google release a core update in September 2026?

No. The Google Search Status Dashboard shows no ranking or serving incident between 6 and 12 September 2026. The most recent confirmed ranking update remains the August 2026 spam update, which ran from 18 to 21 August. If your rankings moved during this week, the cause was not an announced Google update.

### What did Anthropic accuse Chinese AI labs of doing?

Anthropic's September 2026 threat intelligence report, published 10 September, alleges that Alibaba, Moonshot AI and DeepSeek ran distillation campaigns against Claude. The report states that unauthorized labs developed increasingly sophisticated methods to circumvent Anthropic's defences and harvest capabilities from US frontier models, targeting Claude's agentic and tool-use capabilities, coding and data analysis, and logical reasoning. It covers activity disrupted between December 2025 and August 2026 across seven harm areas.

### What is Meta Muse?

Muse is Meta's personal AI agent, launched in the US on 8 September 2026. It connects to a user's email, calendar, payments and apps for health, smart home, dining, shopping, music and events, and can send emails, book travel, lower bills, fill out forms, create plans, turn recipe reels into grocery lists and send party invitations. It became the No. 2 app in the US within two days of launch.

### What are Google Analytics Dashboards?

Dashboards launched on 9 September 2026 as a drag-and-drop, grid-based canvas for bringing KPIs and visualizations onto a single report. You drag dimensions and metrics directly onto the canvas rather than building each report separately, then position, resize and align the visualizations manually. Six visualization types are available at launch, including scorecards and tables, and a dashboard can be published straight to the Reports navigation without going through the library.

### Why is ChatGPT Shopping suddenly favouring product feeds?

Profound data across 1,757,723 tracked prompt runs in July 2026 found that the share of ChatGPT Shopping product recommendations classified as feed-integrated jumped from 8.26% to 61.54% on 10 July. In practice this means most product picks now come from retailers with a connected product feed rather than from general web search, so stores without a feed integration are largely absent from that pool.

### Do AI Overviews reduce website traffic?

The evidence is mixed and contested. A University of Washington working paper, last updated 2 September 2026 and not peer reviewed, estimates AI Overviews reduced monthly search referrals to English Wikipedia by roughly 5% after the feature became the US default in 2024 — a revision downward from the same authors' larger February estimate. Google disputes whether the combined referral metric can isolate the feature. In the same week, Le Monde reported that AI Overviews have not hurt its audience so far. The effect appears to depend heavily on a site's query mix.

### When is the next issue of AI & SEO Weekly?

Issues publish weekly, covering Monday to Sunday. The next one will cover 13 to 19 September 2026. You can subscribe to the RSS feed at /blog/weekly/feed.xml.