# AI & SEO Weekly #7: AI Training Controls, AI Mode Ads in India and Citations That Keep Changing

> Canonical HTML version: https://thejayant.in/blog/weekly/2026-11-01
> 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.

- **Google and OpenAI said on 30 October they will honour a shared robots.txt signal for AI training**, and Bing announced separate training and grounding preferences the same day. None of them is active yet.
- **Google brought ads in AI Mode to India and nine other countries on 28 October**, and launched ads on follow-up questions in the US.
- **Ahrefs found that half the websites cited in AI Overviews on day one were gone on day two**, across a 20-day test of five AI tools.
- **Semrush said some automated Google searches now return results real users do not see**, which can make rank trackers show drops that did not happen.
- **OpenAI released GPT-6.1 Terra on 26 October** at $1 per million input tokens, half the price of GPT-6.1 Sol, in ChatGPT Work, Codex and the API.

This is issue #7 of the weekly roundup covering AI, SEO, Google, Meta, the model labs and, at the very end, gaming. It covers **26 October to 1 November 2026** and follows [issue #6, covering 11 to 17 October](https://thejayant.in/blog/weekly/2026-10-17). There was no issue for 18 to 25 October; one documentation change from that week is listed separately at the end, and the rest of that week is not covered here.

The thread this week is control and measurement. Publishers are getting more specific ways to say how AI may use their pages, advertisers in India have a new place to appear, and two studies showed that both AI citations and rank-tracker data move more than a single check suggests. Before changing anything, decide where you want to appear, find out which control actually affects that, and measure it more than once.

Every item has the same shape: what happened, why it matters, and a link to the source. Every claim carries a label. Confirmed means the company or maintainer said it themselves; performance claims inside an announcement are still the company's claims. Reported means a named outlet reported it. Observed means a named source measured or saw something in a defined sample or product, and the scope of their sample matters.

## The five that actually matter

If you read nothing else, read these. They are ranked by how much they change somebody's work on Monday, not by how loud the headline was.

1. **[Publisher controls for AI are getting more specific.](#ai-training-signal)** Google and OpenAI will honour a shared training signal, and Bing is splitting training from grounding. Dates and effects differ by provider, so read each one's rules before you change yours.
2. **[Ads in AI Mode reached India.](#ai-mode-ads-india)** Indian advertisers should check eligibility and reporting in their own accounts before setting a test budget.
3. **[AI citations changed sharply from one day to the next.](#ahrefs-citation-stability)** Measure AI visibility repeatedly with a fixed question set, not with one screenshot.
4. **[Rank-tracker data needs a second check.](#semrush-poisoned-serps)** Confirm an unexpected drop in Search Console and with real searches before you rewrite or roll back a page.
5. **[GPT-6.1 Terra is a cheaper model for Work, Codex and the API.](#gpt-6-1-terra)** OpenAI also canceled its planned Pro 200 allowance cut. Reprice repeatable agent work with the current model and plan.

- **Mistral Large 4 weights:** answered. Mistral published the weights under its own licence. [See Other major releases](#mistral-large-4-weights).
- **The 16 October ranking volatility:** still unconfirmed. Google's Search Status Dashboard showed no newly announced ranking update when we checked for this issue. [See SEO](#semrush-poisoned-serps) for one reason tracker data can mislead.
- **Personal Agent Protocol v0.1 and Meta's business terms:** not re-checked for this issue. We will come back to both next week.

## AI 2 stories

### Anthropic filed with the US Securities and Exchange Commission on 26 October for a proposed initial public offering

Confirmed 26 Oct · [Anthropic](https://www.anthropic.com/news/s1-sec)

Anthropic filed a Form S-1 registration statement for a proposed IPO of its Class A common stock and plans to list on the Nasdaq Global Select Market under the symbol "ANT". The number of shares and the price range have not been set, and the registration statement has not yet taken effect. A filing starts the process; it does not set a trading date.

**Why it matters:** buyers judging Anthropic as a long-term supplier now get a formal disclosure process to read alongside product announcements.

### Google DeepMind previewed Genie 4 on 27 October, a world model that generates explorable scenes with sound

Confirmed 27 Oct · [Google DeepMind](https://deepmind.google/blog/genie-4-a-new-generation-of-world-simulation/)

Genie 4 is a research preview for trusted testers. It generates explorable environments with music, ambient sound and dialogue, and lets people change a running scene with text, such as the weather or the camera. Wider access is planned for Google AI Ultra subscribers in Project Genie and for early-access partners of the Gemini API and AI Studio. DeepMind says objects and physics can drift over longer sessions.

**Why it matters:** creative teams get another way to try a scene before paying for production, but access and consistency still limit what it can deliver.

## AI SEO 2 stories

### Ahrefs found that 50% of the websites cited in AI Overviews on day one were gone on day two

Observed 29 Oct · [Mateusz Makosiewicz, Ahrefs](https://ahrefs.com/blog/how-stable-are-ai-citations/)

Ahrefs asked five AI tools the same questions every day for 20 days and tracked about 37,500 websites cited on day one. By day two, 50% of the websites cited in AI Overviews were gone, 58% in AI Mode and 59% in ChatGPT. These figures describe Ahrefs' sample of cited websites, not the odds that your page will disappear.

**Why it matters:** for generative engine optimisation (GEO) and AI optimisation (AIO), repeated checks are a better basis for reporting than one prompt that worked. Save the prompt, tool, date, location and cited URL, and report visits and leads separately. Our [website facts test](https://thejayant.in/blog/can-ai-understand-your-business) covers the content work underneath.

### Bing announced separate preferences for AI training and AI grounding on 30 October, replacing NoArchive and NoCache for those uses

Confirmed 30 Oct · [Bing Search Blog](https://blogs.bing.com/search/2026/10/your-content-your-choice-clearer-preferences-for-training-and-grounding/)

Microsoft added two independent preferences: AI Training, set in robots.txt or on a page, and AI Grounding, set per page, which decides whether a page can inform generated answers. Sites using NoArchive or NoCache can record temporary grounding preferences in Bing Webmaster Tools from **15 December 2026**. The new preferences take effect as Microsoft crawls from **1 February 2027**, and NoArchive/NoCache sites must publish their choices by **1 March 2027**. Microsoft says the choices do not change eligibility for ordinary Bing results; restricting grounding can reduce AI visibility.

**Why it matters:** publishers can now choose training and answer inclusion separately, but sites with old NoArchive or NoCache tags need a planned transition, because those tags stop counting after 1 March.

## AI Automations 1 story

### n8n released the 3.0 beta on 27 October with breaking changes and no way back to 2.x

Confirmed 27 Oct · [n8n release notes](https://docs.n8n.io/changelog/release-notes.md#n8n30) · [n8n announcement, 23 Oct](https://community.n8n.io/t/announcing-n8n-version-3-0/324016)

n8n 3.0 removes legacy nodes such as Function, Cron and the old OpenAI node, tightens security defaults and drops npm installs in favour of Docker for self-hosting. The stable release is planned for early December. n8n says moving to 3.x is one way, and recommends running the Migration Report in 2.39 or later, taking a full backup and testing on a separate instance first.

**Why it matters:** agencies running client automations should list the affected nodes and test workflows on a copy before scheduling any migration.

## SEO 3 stories

### Google and OpenAI each said on 30 October they will honour a shared robots.txt signal for AI training

Confirmed 30 Oct · [Google Search Central](https://developers.google.com/search/blog/2026/10/ai-training-control) · [OpenAI](https://openai.com/index/helping-websites-express-ai-training-preferences/)

Both announcements draw on the IETF AI Preferences working group: one token in robots.txt states a preference about foundational generative AI training to every participating provider. Google aims to support it in 2027; OpenAI says it will implement it in the coming months. Existing controls stay. One catch on Google's side: a site that opts out with the shared token and does not name Google-Extended will be treated as if Google-Extended were disallowed, which also removes it from grounding in Gemini Apps.

**Why it matters:** one shared signal could simplify robots.txt, but it is not active yet and each provider's implementation decides the result. Before changing robots.txt, write down what you want: training permission, search discovery, answer inclusion or crawl access. If an assistant cannot reach an important page, start with [why ChatGPT cannot read a website](https://thejayant.in/blog/why-chatgpt-cant-read-my-website).

### Google updated its spam guidance on 26 October to help platforms keep link-shared pages out of Search

Confirmed 26 Oct · [Search Central documentation updates](https://developers.google.com/search/updates#october-2026) · [User-generated spam guide](https://developers.google.com/search/docs/monitor-debug/prevent-abuse)

The guide now covers platforms where people create shareable pages, generated outputs or interactive tools. It asks operators to separate pages meant for search from pages meant for direct sharing or collaboration, to block indexing of link-shared content, and to watch for abuse in Search Console. This is a documentation update, not a ranking update.

**Why it matters:** public sharing features can quietly create thousands of indexable URLs. Check their indexing defaults before spam or thin pages build up.

### Semrush said on 27 October that some automated Google searches return results that real users do not see

Observed 27 Oct · [Semrush](https://www.semrush.com/blog/poisoned-serps/) · [Search Engine Land](https://searchengineland.com/semrush-confirms-poisoned-google-results-499199)

Semrush calls them "poisoned SERPs": pages that are clearly off-topic, or that look normal at first but contain loosely related results further down. Its example is a Brazilian search for a baby stroller that returned a dental supply store. Semrush describes it as an industry-wide issue for data providers. Google has not commented, and the cause is not confirmed.

**Why it matters:** a sudden drop in a third-party tool needs confirming in Search Console and with real searches before it becomes a reason to rewrite, roll back or alarm a client. Google's [Search Status Dashboard](https://status.search.google.com/summary) showed no newly announced ranking update this week, which does not rule out unannounced or site-specific changes.

## Marketing AI & Ads 2 stories

### Google expanded ads in AI Mode to India and nine other countries on 28 October

Confirmed 28 Oct · [Google](https://blog.google/products/ads-commerce/ads-in-ai-mode-global/)

The new markets are Australia, Canada, India, Kenya, Malaysia, New Zealand, Nigeria, Pakistan, the Philippines and Singapore, with Brazil, Mexico and the UK planned for Q4. Google also launched multiturn ads in the US, which can appear on follow-up questions. Google's claim that sponsored and organic product listings get similar click-through rates comes from its own US data for 20 August to 3 September.

**Why it matters:** Indian advertisers have a new placement to test; follow-up-question ads are a separate, US-only rollout. Do not apply the US click-through figure to India.

### ChatGPT Ads opened a multi-image ad format in beta on 26 October

Confirmed 26 Oct · [OpenAI Help Center](https://help.openai.com/en/articles/20001212-creating-ads-for-chatgpt-ads)

Advertisers can upload four to six square images, all linking to one destination URL, with "Learn more" as the only call to action for now. Single-image and multi-image ads both work with Reach, Clicks and Conversions campaigns. The help page does not name new markets for the format.

**Why it matters:** build a coherent set of images around one landing page and measure what that page converts before spending more on creative.

## Google 2 stories

### Google opened a preview of its Connected Apps Platform on 28 October, one place to build apps for Search and Gemini

Confirmed 28 Oct · [Google](https://blog.google/innovation-and-ai/technology/developers-tools/connected-apps-one-place/)

Developers can connect through a hosted Model Context Protocol (MCP) server or Android AppFunctions, and choose Search, Gemini or both before review. Google says more than 50 Connected Apps have launched; Adobe, Experian, Localiza and Peloton are live in Gemini, with Search support expected in the coming weeks. Access opens in phases.

**Why it matters:** businesses with a useful interactive service have another distribution route to weigh alongside their website. For the marketing side of MCP, see our [comparison of marketing data MCP tools](https://thejayant.in/blog/best-marketing-mcp-for-claude-chatgpt).

### Sundar Pichai said on 28 October that one in five AI Mode searches uses voice, images or documents

Confirmed 28 Oct · [Alphabet Q3 remarks](https://blog.google/company-news/inside-google/message-ceo/alphabet-earnings-q3-2026/)

In Alphabet's third-quarter remarks, Pichai also said AI Mode queries are on average three times longer than traditional searches. These are Google's figures about AI Mode only. They do not mean one in five of all Google searches is multimodal, and they say nothing about website traffic.

**Why it matters:** make product details, specifications and visuals understandable together, so a page can answer a detailed task rather than only match a short phrase.

## Google Analytics 0 stories

Nothing shipped. The latest entry in [GA4's release notes](https://support.google.com/analytics/answer/9164320) is still 5 October, and we found no GA4 release dated 26 October to 1 November. That is a statement about the public release log, not a promise that nothing changed in any account.

## Search Console 0 stories

Nothing shipped. We found no new Search Console feature in [Search Central's updates](https://developers.google.com/search/updates) or the [Search Central blog](https://developers.google.com/search/blog) for this window. The updated spam guide under SEO does point to Search Console for spotting abuse; use it when you investigate unusual indexed pages, but it is guidance, not a new report.

## Meta 1 story

### NVIDIA and Meta expanded their work on Meta's Muse models on 26 October, and NVIDIA will invest $3 billion in Meta's equity offering

Confirmed 26 Oct · [NVIDIA Newsroom](https://nvidianews.nvidia.com/news/nvidia-and-meta-to-expand-collaboration-on-muse-and-muse-family-of-models-amid-robust-growth)

The companies will work on training and inference performance for the Muse family, and Meta plans to deploy more NVIDIA Vera Rubin systems and Vera CPUs, along with Spectrum-X networking and BlueField-4. They are also considering work on NVIDIA's agent safety tools, OpenShell and Sentry; those are exploratory, not deployed products.

**Why it matters:** Meta is committing more money and hardware to its own models, but what that does for its apps still has to show up in the products.

## WhatsApp 2 stories

### WhatsApp added four group-chat features on 29 October, including messages that stay pinned forever

Confirmed 29 Oct · [WhatsApp Blog](https://blog.whatsapp.com/never-miss-what-matters-in-your-group-chats)

Groups can now pin a message with no expiry, send recent message history to someone after they join, share group status updates that disappear after 24 hours, and filter chats to show only mentions and replies. These are group-chat changes; the announcement does not add anything to the WhatsApp Business Platform.

**Why it matters:** community managers can keep recurring instructions at the top and make onboarding easier as the features reach their groups.

### WABetaInfo saw profile cover photos rolling out in the WhatsApp Android beta on 30 October

Observed 30 Oct · [WABetaInfo](https://wabetainfo.com/whatsapp-is-rolling-out-profile-cover-photos-on-android-beta/)

In beta 2.26.43.11, some Android testers can add a landscape banner above their profile picture. Visibility can be set to everyone, contacts, contacts except some, or nobody. This is beta evidence, so check the specific account rather than assuming it is in the stable app.

**Why it matters:** business profiles may gain another visual element, but campaign or support instructions should wait until it reaches the accounts involved.

## Claude 1 story

### Anthropic described its work with six drug-development organisations on 27 October

Confirmed 27 Oct · [Anthropic](https://www.anthropic.com/research/early-progress-drug-development)

The partners include Bristol Myers Squibb, Generate Biomedicines and BigHat Biosciences, using Claude for tasks from antibody design to reviewing manufacturing documents and assessing clinical trial plans. Anthropic also described its own discovery programmes. These are research programmes and early case studies, not evidence of clinical effectiveness or regulatory approval.

**Why it matters:** the useful question for any team is which defined task it can validate, and how specialists check the output before it influences a real decision.

## ChatGPT 1 story

### ChatGPT added iCloud Mail and Calendar connections on 27 October for all plans

Confirmed 27 Oct · [ChatGPT release notes](https://help.openai.com/en/articles/6825453-chatgpt-release-notes)

iCloud users can connect Mail and Calendar to find emails, catch up on an inbox, check a schedule and update calendar events. OpenAI lists it for all ChatGPT plans on the web, the desktop app and iOS. What ChatGPT can read or change still depends on the access you grant.

**Why it matters:** people who run client work through iCloud can now bring those sources into ChatGPT workflows.

## Gemini 1 story

### Google announced Gemini 4 Carbon on 28 October, starting with a small group of cyber defenders

Confirmed 28 Oct · [Google](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/)

Google calls Carbon its strongest model, focused on cybersecurity and long tasks. Access starts with cyber defenders and widens to trusted partners through the Fairwind Program; developers, businesses and consumers get it later, with no date. Google lists an introductory API price of $2 per million input tokens and $10 output. The benchmark results are Google's launch claims.

**Why it matters:** follow the evaluations, but keep production plans tied to models you can actually access and test.

## Codex 3 stories

### OpenAI released GPT-6.1 Terra on 26 October at $1 per million input tokens, half the price of GPT-6.1 Sol

Confirmed 26 Oct · [OpenAI](https://openai.com/index/introducing-gpt-6-1-terra/)

Terra is a fast model for visual reasoning, computer use and writing. API pricing is $1 per million input tokens, $0.05 cached and $5 output, 50% lower than GPT-6.1 Sol. It is rolling out to Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex. OpenAI's capability comparisons are its own evaluations.

**Why it matters:** test repeatable tasks on Terra and compare the cost of a successful task, including retries and review time, before moving a workflow.

### OpenAI canceled the planned Pro 200 Work and Codex allowance cut on 29 October and doubled Pro 100 and Pro 500

Confirmed 29 Oct · [ChatGPT release notes](https://help.openai.com/en/articles/6825453-chatgpt-release-notes)

Pro 100 and Pro 500 plans now include twice their previous Work and Codex allowance, and Business Premium plans also get more. The Pro 200 reduction announced in September is canceled: subscribers who had the higher allowance keep it, and those on the lower allowance are raised to the same limit.

**Why it matters:** recheck your current allowance before buying more capacity because of the old reduction notice.

### Codex CLI 0.166.0 added `codex doctor --network-only` for connection problems

Confirmed 30 Oct · [Codex release 0.166.0](https://github.com/openai/codex/releases/tag/rust-v0.166.0)

The release, published late on 29 October UTC, adds focused checks for configuration, connectivity and WebSockets, keyboard-driven transcript selection and GPT-6.1 Terra in the model catalogue. It also keeps approved permission settings when a conversation is resumed or forked.

**Why it matters:** when a session fails to connect, the new command is a narrower place to start than a full reinstall.

## Claude Code 1 story

### Claude Code 2.1.316 set 30 November as the date fast mode for Opus 4.8 stops working

Confirmed 31 Oct · [Claude Code release 2.1.316](https://github.com/anthropics/claude-code/releases/tag/v2.1.316)

The release adds a setting to turn off the ultrathink keyword, a default effort level for subagents and pooled group budgets in the Claude apps gateway. Its notes say fast mode for Opus 4.8 has been deprecated since 30 October and will be removed on **30 November 2026**. After that, fast-mode requests on Opus 4.8 return an error instead of falling back; the notes point to Opus 5.5 for the speed-up.

**Why it matters:** teams with pinned model settings have a dated change to test before unattended jobs start failing.

## Mobile AI & Devices 1 story

### ChatGPT began rolling out account switching on iOS and Android on 30 October

Confirmed 30 Oct · [ChatGPT release notes](https://help.openai.com/en/articles/6825453-chatgpt-release-notes)

Mobile users can stay signed in to up to two ChatGPT accounts and switch without signing out. Chats, memory, settings and subscriptions stay separate for each account. The feature is rolling out, so the option may not appear for everyone yet.

**Why it matters:** check which account is active before pasting client information or starting work on your phone.

## Other Major Releases 1 story

### Mistral released the weights of Large 4 under its own modified MIT licence

Confirmed 29 Oct · [Mistral AI](https://mistral.ai/news/mistral-large-4-oss/) · [Hugging Face](https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B)

Large 4, nicknamed "Le Chonk", takes multimodal input and has three reasoning modes. It is released under the Mistral Ecosystem Open-Weight licence, which allows use, fine-tuning and commercial use, except that businesses offering the model as a service need a separate commercial licence. Downloadable weights are not the same as a standard MIT licence.

**Why it matters:** organisations looking at private deployment have another option, with licence terms and hardware needs to assess alongside quality.

## Off the clock: gaming 1 story

### Avatar Legends: The Fighting Game received its Avatar State patch on Xbox on 30 October

Confirmed 30 Oct · [Xbox Wire](https://news.xbox.com/en-us/2026/10/30/avatar-legends-the-fighting-game-enters-the-avatar-state/)

Gameplay Group's update for Xbox Series X|S reworks Avatar Aang and Nightmare Korra, adds Ranked Mode 2.0, redesigns the menus and match screens, and rebalances every fighter. It is an update to the existing game, not a new release.

**Why it matters:** returning players have a real patch to learn before their next ranked session.

- **Google clarified its URL guidance on hyphens and underscores (23 October).** The [update log](https://developers.google.com/search/updates#october-2026) says the edit to the [URL structure guide](https://developers.google.com/search/docs/crawling-indexing/url-structure) brings it in line with Google's video series. It is a clarification, not a ranking change: use it for new URLs and weigh migration costs before renaming existing ones.

1. **Audit your AI permissions.** Record your current crawler rules, Google-Extended choices and any NoArchive or NoCache tags. Give each planned change a specific purpose and a review date, and note 15 December for Bing.
2. **Check unexpected ranking movements.** Compare the affected queries and pages with Search Console, actual visits and a few real searches before changing the site.
3. **Repeat your AI visibility checks.** Use a fixed question set and record citations over time. Report mentions, citations, visits and conversions as separate measures.
4. **Review AI Mode ad eligibility in India.** Check the account and documentation, then define a small test with a business outcome and a spending cap.
5. **Reprice repeatable agent work.** Compare Terra with the model you use now, counting corrections, failed runs and human review, and check your current subscription allowance. For product sites, our [product-catalogue accuracy audit](https://thejayant.in/blog/can-ai-read-product-catalogue-correctly) pairs well with these checks.

## FAQ

### What were the biggest AI and SEO news stories of 26 October to 1 November 2026?

Google and OpenAI said on 30 October they will honour a shared robots.txt signal for AI training, and Bing announced separate training and grounding preferences. Google expanded ads in AI Mode to India and nine other countries on 28 October. Ahrefs found that half the websites cited in AI Overviews disappeared the next day in its sample, Semrush said some automated Google results differ from what users see, and OpenAI released GPT-6.1 Terra.

### Does the shared AI training signal remove a site from AI search?

No, not by itself. It states a preference about foundational model training, and it is not active yet: Google aims to support it in 2027 and OpenAI in the coming months. Grounding and crawl access are separate. On Google, opting out with the shared token without naming Google-Extended is treated as disallowing Google-Extended, which also removes grounding in Gemini Apps. Microsoft's new settings separate training from grounding explicitly.

### Are ads in AI Mode available in India?

Google named India in its 28 October expansion. That does not mean every advertiser, campaign or query is eligible immediately. Ads on follow-up questions launched only in the US.

### Does a missing AI citation mean my AI visibility strategy failed?

No. Ahrefs' study shows citations change a lot from day to day. Look for a sustained pattern across comparable checks, then look at visits and leads. One missing citation proves nothing.

### Should a team move to n8n 3.0 now?

Not in production. The 27 October release is a beta with breaking changes, the stable release is planned for early December, and the move cannot be undone. Run the Migration Report, take a full backup and test on a separate instance first.

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

Issue #8 covers 2 to 8 November 2026. It will check the Personal Agent Protocol and Meta's business terms, which this issue did not re-check, and how the Bing and Google training controls develop.