You changed your business address, updated your prices or stopped offering a service. Your website reflects the change. ChatGPT still gives customers the old information.
The natural assumption is that the assistant has not updated its knowledge. That is one possible explanation, but it is not the only one—and it is not always the most useful place to start.
ChatGPT can show old business information because the conversation contains outdated context, a retrieved source still carries the old fact, the current page is not accessible, or the answer does not use current web evidence. Even when search is used, an answer can cite an outdated page or misinterpret an old offer as current.
The practical response is to identify where the old fact appears and which version the assistant can verify. Changing the publication date on your homepage is not a universal fix.
This guide explains how to investigate stale business details, what you can update, and how to monitor the outcome without relying on promises of an instant refresh.
What should you check first?
Start with the exact outdated claim, not a general question about whether AI knows your business.
| Outdated detail | First places to inspect | Important distinction |
|---|---|---|
| Old address | Contact page, business profiles, branch pages | Moved office versus a different active branch |
| Old phone number | Footer, contact page, directory listings | Current public number versus historical campaign tracking |
| Previous price | Product page, price table, catalogue, promotional page | Current price versus a dated offer |
| Retired service | Service pages, navigation, older articles | No longer offered versus temporarily unavailable |
| Former business name | About page, official profiles, old domain | Same company after rebrand versus a separate entity |
| Old product availability | Exact variant page and relevant product feed | Current stock versus a past catalogue entry |
Save the answer and any citations before asking ChatGPT to correct itself. You need that evidence to compare later results.
Is this a knowledge problem, a source problem or a conversation problem?
Treat these as separate diagnostic questions.
1. The answer did not establish current evidence
An assistant may generate a description without showing that it checked today's public information. OpenAI documents web search, but not every business question necessarily results in a successful current-page retrieval.[1]
Request current evidence in a fresh conversation. If the answer becomes accurate, record that difference. Do not conclude that the entire platform has permanently updated.
2. A public source still carries the old detail
A directory may list your former address. A downloadable brochure may show last year's prices. Your old service page may remain accessible even after it disappears from the menu.
If the answer cites that source, you have a concrete correction route. If you discover it independently, label it a possible source rather than claiming certainty about what the assistant used.
3. The current page is not the version being received
Your CMS preview may be updated while logged-out visitors receive a cached page. The current information could also be hidden behind a selection, interaction or failed request.
Check the public page and the retrieved evidence. Do not assume that clicking Save in the editor changed every public copy.
4. The conversation supplies an older version
An earlier message, uploaded brochure or project document may contain the old fact. Asking another question in the same conversation does not remove that context.
A fresh conversation is a useful control. Review relevant memory, project files and connected sources where available. You do not need to delete your account history to begin a controlled comparison.
5. The answer adds a false time assumption
A source may accurately state that a business offered a promotion in the past. The generated answer may drop the date and present it as an ongoing offer.
In that situation, the source and the answer need different treatment. Do not rewrite accurate history simply because the assistant lost its context.
Step 1: create a dated record of the change
Write down what changed, when it changed and where the current fact is documented.
| Field | Example entry |
|---|---|
| Business | Cedar Event Hire, a fictional example |
| Old fact | Delivery included in a particular summer package |
| Current fact | Delivery quoted separately for current bookings |
| Effective date | The date the promotion expired |
| Current public evidence | Current delivery-policy URL |
| Historical evidence | Archived or dated promotion URL |
| Responsible person | The manager who approves delivery terms |
The effective date matters. A customer asking about a booking made during the promotion may need the old terms. A customer booking today needs the current terms.
The reference should therefore say more than “old = wrong.” It should establish when and where each statement applies.
Step 2: compare three kinds of answer
Use separate conversations for these checks so one answer does not supply the correction for the next.
Check A: the normal customer question
Ask the original question as a customer would. For example:
What are the delivery charges for [BUSINESS] in [CITY]?Save the answer, citations, date and visible mode. This captures the problem you are trying to monitor.
Check B: current web verification
Search current public information about [BUSINESS] in [LOCATION].
What is its current [ADDRESS / PRICE / DELIVERY POLICY / SERVICE]?
Give the supporting URL and a short relevant excerpt. Distinguish
current information from dated promotions or historical descriptions.
If sources disagree or the current fact cannot be established, say so.This is a more controlled diagnostic prompt. It should not be merged into the same results table as the neutral customer question without labelling the difference.
Check C: direct-page extraction
Retrieve this exact public page: [CURRENT URL].
What does it state about [SPECIFIC FACT]?
Use this page only. Include a short supporting excerpt.
Do not substitute a different source or infer missing terms.
If retrieval fails or the fact is absent, state that clearly.If Check C is accurate while Check B repeats an old directory, investigate source discovery and consistency. If Check C fails, investigate access. If Check C receives the correct wording but changes its meaning, preserve the interpretation error.
These patterns narrow the investigation. They do not expose every internal cause of the answer.
Step 3: verify your live page outside the editor
Open the current URL while logged out. Look for the exact fact you changed.
Check the visible text, downloadable documents and relevant structured data. If the fact is commercially important, confirm that it is not present only in a rotating banner or an image.
Ask your developer to compare the public response with the content expected from the CMS. Check cache layers deliberately rather than repeatedly clearing every cache without evidence.
For a price change, confirm the unit and currency. For an address change, confirm the correct branch. For a product change, check the exact variant. An apparent freshness problem may actually be a scope mismatch.
If the assistant cannot access the corrected content, follow the website retrieval troubleshooting guide.
Step 4: find remaining public copies of the old fact
Search your own website and the sources most relevant to the answer.
Start with:
- The specific cited page.
- Contact information repeated in headers and footers.
- Pricing, policy, product and service pages.
- PDFs, brochures and older campaign pages.
- Business profiles and directory records you manage.
- Important external articles containing the disputed information.
An exact search for the old phone number or a distinctive phrase can help locate copies. Search results are incomplete, so use your CMS or a crawl for a more systematic review of your own site.
Prioritise evidence. Ten irrelevant directory edits are not automatically more useful than correcting the exact page the answer cites.
Do not demand removal of every historical reference. A news article reporting your former office address at the time of an event may be accurate. A current contact directory showing that address as today's location is a different problem.
Step 5: publish the change clearly
Your corrected page should let a customer understand what applies now.
For a changed policy, state the current condition beside the relevant offer. If an effective date matters, include it plainly.
For a fictional example:
For bookings placed from 1 November, delivery and setup are quoted separately. Earlier bookings follow the delivery terms in their confirmed quotation.
That is clearer than an undated “terms updated” note. Use wording approved for your actual business; do not copy the example date into a real policy.
For a move, identify the new customer-facing location and whether the old location is closed. For a rebrand, explain the relationship between names without claiming that two unrelated entities are the same business.
Keep public facts concise enough to verify, but preserve conditions. Oversimplifying an offer can create the next inaccurate answer.
Should you delete an old page or keep it with a notice?
Choose according to the page's purpose and the customer's needs.
| Page situation | Reasonable action |
|---|---|
| A current policy page contains old terms | Update it and preserve version history internally where needed |
| A past promotion has historical value | Clearly mark its period and expiry; link to current options |
| A replaced page has a close equivalent | Consider a permanent redirect to that equivalent |
| A retired page has no relevant replacement | Use an appropriate retirement response rather than an unrelated redirect |
| An external historical article is accurate | Request clarification only if its current presentation is misleading |
For URL changes, use the domain migration and rebrand guide. Do not turn every old URL into a redirect to the homepage simply to remove a warning from an audit.
Step 6: update profiles and request external corrections
Correct public profiles you manage using their normal editing processes. Verify the published outcome rather than assuming that a submitted edit is already live.
If an external publisher carries an outdated current-business description, provide the exact page, old statement, correct replacement and official evidence.
A request such as “please update our delivery information” leaves work for the editor. A precise sentence and source make the correction easier to assess.
For a complete evidence and outreach process, see how to trace and correct wrong business information in AI answers.
Can you ask ChatGPT to update its information?
You can provide corrected information in a conversation, ask it to verify a current source and submit feedback on an inaccurate response.
Those actions have different outcomes:
| Action | What it can establish | What it does not establish |
|---|---|---|
| Paste the correct fact into chat | The assistant can use supplied information in that task | Independent public discovery of the correction |
| Supply a current URL | A test of that URL in that retrieval path | That every future answer will use it |
| Submit response feedback | The provider receives a report through the available route | A guaranteed publication or refresh deadline |
| Update your source page | The public source is corrected once verified live | Immediate changes across all downstream systems |
Avoid anyone promising to “retrain ChatGPT with your new address” merely by publishing a page or submitting one prompt. Ask what system is actually being changed and what observable result they will verify.
How long does it take for new information to appear?
There is no universal timeline across assistants, search systems and source types.
Google provides a recrawl request process for eligible site owners and notes that crawling can take days to weeks without guaranteeing inclusion. That request concerns Google's systems, not every AI assistant.[2]
A direct retrieval can also behave differently from an answer that selects information through search. Treat them as separate monitoring tasks.
Use a schedule you can maintain. My suggested starting point is to verify the live correction immediately, check direct retrieval, then revisit independent discovery after several days and again later if the issue persists. These are operating choices, not platform service levels.
Repeatedly asking the same question every few minutes is unlikely to produce a useful change record. Record meaningful intervals and any relevant deployment or source changes.
What about Gemini and Claude?
Apply the same source-first process, but record each product's actual configuration.
Anthropic documents web search and direct URL retrieval. Google documents source and related links in Gemini Apps. Neither fact means that all three assistants will choose the same evidence or refresh simultaneously.[3][4]
Keep separate logs for ChatGPT, Gemini and Claude. Do not use one corrected response in Claude as evidence that the ChatGPT issue has disappeared.
Also keep consumer-app tests separate from API tests, research modes and uploaded-document tasks. Different tools and supplied context make them different experiments.
A simple monitoring log
Issue ID:
Business and relevant location/product:
Old fact:
Current approved fact:
Effective date:
Current public evidence:
Old public copies identified:
Corrections requested and published:
Platform and visible model/mode:
Prompt type: customer / current search / direct URL
Exact prompt:
Test timestamp and timezone:
Full response and citations:
Outcome: current / outdated / mixed / uncertain / retrieval failed
Next action and owner:Report source corrections and answer outcomes separately. “Website updated, external listing pending” is more informative than “AI fixed.”
If you automate this process, failed test collection should trigger a review rather than being treated as an absence of misinformation. Your monitoring system must not turn an unverified model response into the new official fact.
What not to do
Do not change dates without changing facts. A recent-looking timestamp does not resolve an outdated brochure or contradictory delivery policy.
Do not repeatedly correct the same chat and call it a public retest. The conversation now contains your correction.
Do not publish dozens of near-identical updates. Maintain the relevant pages and fix actual contradictions.
Do not assume every old reference is harmful. Historical accuracy and current accuracy can coexist when dates and scope are clear.
Do not turn one good answer into a guarantee. Report what happened in the recorded tests.
Frequently asked questions
Why is ChatGPT still using my old business address?
Check the answer's citations, your public contact pages, business listings and conversation context. The old address may remain public, refer to another branch or appear without current retrieval evidence. Establish which case applies before choosing a fix.
Will updating my website automatically update ChatGPT?
It updates the source once the change is live. It does not guarantee that ChatGPT immediately retrieves or selects that source for every question. Verify direct access and independent discovery separately.
Can I request an immediate refresh of my company information?
You can use available feedback channels and ask for current source verification in a conversation. There is no universal owner-controlled refresh command for every generated business answer.
Should I remove an expired promotion?
If it remains useful historical information, clearly identify the expired period and direct readers to current options. If it has been replaced or retired, choose an appropriate URL treatment. Decide according to purpose, not merely whether an assistant cited it once.
Does a newer date make an assistant trust a page more?
A truthful effective date helps a reader interpret time-sensitive facts. It is not a guarantee of source selection or accuracy. Keep dates meaningful and consistent with the actual change.
Why does the assistant get the fact right only when I provide the URL?
That suggests a difference between direct retrieval and independent evidence selection. Investigate the sources selected in the independent test and whether the corrected fact is clearly discoverable. The pattern alone does not prove a specific indexing problem.
How do I measure whether the correction worked?
Retain the original prompt, use fresh conversations and log repeated outputs. Record whether the fact is current and whether the citation supports it. Use the full business-facts framework for a broader audit.
Make the current answer easy to verify
Choose one outdated fact. Establish the correct scope and effective date, inspect the evidence behind the answer and correct the public copies that actually need changing.
That gives you a reviewable task and a meaningful retest. It also improves the information available to customers who never use an AI assistant.
If you need help tracing stale sources and checking the live implementation, request an AI-ready website audit.
- Can AI understand your business? – the website facts test: reference sheet, three test tracks and scoring
- Why can't ChatGPT read my website? – diagnose access, redirects, firewalls and missing content
- AI gets your business details wrong – trace the source and correct the facts
- Why does ChatGPT show old information? (this guide) – find the outdated source and track the correction
- Domain changes, rebrands and AI search – preserve sources, test old links and identity after a move
- Can AI read your product catalogue? – variants, prices, units and availability, checked
- Does AI know where you operate? – offices, service areas, delivery and remote markets
Still seeing old facts in AI answers?
I'm Jayant Solanki, an SEO, GEO and automation strategist. I trace stale sources, check the live implementation and cache layers, and give you a prioritised list of the copies that actually need changing.
Ranked #1 for "metal buildings" · +30% YoY organic traffic · Evidence-labelled research
Sources
Documentation checked on 17 October 2026. Diagnostic prompts, monitoring intervals and prioritisation are editorial recommendations. Examples are fictional; no measured refresh speed or platform-wide correction is claimed.
- OpenAI: Searching the web with ChatGPT.
- Google Search Central: Ask Google to recrawl your URLs.
- Anthropic: Enable and use web search.
- Google: View related sources from Gemini Apps.
Jayant Solanki
AI-Ready SEO, GEO & AIO strategist based in Indore, India. He works across technical SEO, analytics, website development and marketing automation for eCommerce, local-service and global businesses.
More about Jayant →