ChatGPT has outdated information about my company: why and how to fix
ChatGPT showing old prices, former executives or discontinued products? Why AI models lag behind your updates, and how to replace stale facts at the source.
- AI shows old facts because models learn from a frozen snapshot and search answers still find old pages that were never updated or removed.
- Fix the stale page, not just your new one: archived pricing, old press releases, directories and partner pages carry most outdated facts.
- Make the current fact dated, explicit and repeated on sources engines trust, then re-test on a schedule.
ChatGPT shows outdated information about your company for two reasons: the model's training data is a frozen snapshot, and when it searches, it still finds old pages that were never updated, redirected or removed. To fix it, find every source still stating the old fact, update or retire it, publish the current fact with a date, and repeat it on sources the engines trust.
This article covers stale facts specifically: old prices, former executives, discontinued products and old addresses. For the general correction workflow, see how to fix wrong AI information.
01Why does ChatGPT still show our old information?
There are two separate mechanisms, and they need different fixes.
The first is model memory. A model learns from data collected up to a cutoff, then is used for months or years. OpenAI lists GPT-5's knowledge cutoff as September 30, 2024;[1] newer OpenAI models list cutoffs in 2026.[2] Anything that changed after a model's cutoff is invisible to it unless it searches.
The second is retrieval of old pages. When ChatGPT searches, it rewrites the question and can send it to partner providers such as Microsoft.[3] If the top results include a 2022 press release announcing your old CEO or a cached pricing page, the answer will repeat them.
02Which outdated facts cause the most damage?
Not every stale fact is equally risky. Prioritize the ones that change a buying decision or create a compliance problem.
| Stale fact | Typical source | Business risk |
|---|---|---|
| Old pricing or plan names | Archived pricing pages, comparison sites, old reviews | Lost deals, angry customers expecting old prices |
| Former executives | Old press releases, conference bios, directories | Credibility with investors and press |
| Discontinued products | Legacy product pages, marketplace listings | Support load, wrong expectations |
| Old addresses or offices | Business listings, map profiles, old PDFs | Missed visits, local-search confusion |
| Old policies or terms | Cached help pages, forum threads | Legal exposure if customers rely on them |
The legal side is not hypothetical. In 2024 a Canadian tribunal held Air Canada responsible for wrong refund information its own website chatbot gave a customer, rejecting the argument that the chatbot was a separate entity.[4] That case involved a company's own bot, not a third-party engine, but it shows how seriously stale or wrong policy information can be treated.
03How do you find where the stale fact is coming from?
Start from the answer. Ask the question several times, note any sources the engine cites, and then search the web for the exact old fact: the old price, the former CEO's name with your company name, the discontinued product name.
Make a list of every page that still states it. You will usually find more on your own domain than you expect: old blog posts, PDF brochures, careers pages, investor decks and help-center articles.
Record each one with its URL, owner, the exact wording and whether you control it. That list is your work queue, and it doubles as evidence if a regulator, journalist or customer later asks what you did to correct the record.
04How do you update the facts at the source?
Work through the list in order of influence: pages engines cite most, then pages that rank for your brand name, then the long tail.
- 1Update or retire owned pagesWeek 1Edit pages that should stay; redirect or remove ones that shouldn't. For history pages, label old facts clearly as past, with dates.
- 2Publish a dated current-facts pageWeek 1One plain page with current pricing, leadership, products and locations, each with an 'as of' date. Link it from your footer.
- 3Fix profiles you controlWeeks 1–2Business profiles, review platforms, marketplaces, app stores, LinkedIn. Change every field, not just the description.
- 4Ask third parties to updateWeeks 2–6Send short, sourced requests to directories, partners and publishers. Link to the dated facts page.
- 5Clear stale search resultsWeeks 2–6Use your search consoles to recrawl updated pages; use Google's outdated-content tool for removed third-party pages.
- 6Re-test on a scheduleOngoingRe-run the same prompts weekly and log which engines have flipped.
Two details make the difference. First, state the change explicitly: "Since March 2026, [Name] is our CEO, succeeding [Name]." A sentence that names both old and new helps engines connect them. Second, avoid leaving only a deleted page behind; a page that says what replaced a discontinued product is more useful to an engine than a 404.
Google offers a Refresh Outdated Content tool for pages you do not own that have been removed or significantly changed but still show old content in results.[5] It does not edit live pages, so it only helps after the third party updates theirs.
05What works for each type of stale fact?
The general steps apply to everything, but each fact type has its own trap.
Old pricing. Comparison sites and reviews quote prices with no date, so engines cannot tell old from new. Put an "as of" date next to every price on your own pages, and ask review platforms to update plan names as well as numbers. If prices vary by market, say so on the page, or engines will mix currencies and regions.
Former executives. Leadership pages get updated quickly, but bios on conference sites, podcast pages and old press releases do not. Publish one dated announcement that names both the outgoing and incoming person, update LinkedIn company pages and Wikidata where applicable, and ask event organizers to update speaker pages that rank for your brand.
Discontinued products. Deleting the product page often makes things worse, because the third-party pages remain and nothing on your site contradicts them. Keep a short page that says the product was retired, when, and what replaced it. Marketplace and app-store listings need the same treatment.
Old addresses. Business listings and map profiles are the main source here. Update the verified profile first, then directories and data aggregators, then PDFs and footers on regional sites. Check suggested edits regularly, as the next section explains.
06Why does Google Business Profile matter here?
Address, hours and phone numbers are often pulled from business listings. Google lets owners edit these details and reviews changes before they go live.[6] Google's help documentation also now gives owners four days to accept or reject edits that others suggest before Google may publish them, according to a September 2026 report.[7]
If you moved offices and someone suggests your old address, an unmonitored profile can quietly revert. Assign an owner to check suggested edits.
07How long until ChatGPT shows the new facts?
It depends on which mechanism produced the old fact. Search-based answers can change once engines recrawl the corrected sources. Memory-based answers change only when a new model is deployed, which you do not control.
You can't edit model memory, but you can make sure that when an engine searches, every result it finds agrees on the new fact. Consistency across your site, profiles and third-party sources is what tips the answer.
You can also send feedback through each platform's tools, which we cover in how to report incorrect information to ChatGPT, Gemini and Perplexity. Treat feedback as a supplement; it does not change the sources the engine finds.
08How do you stop facts going stale again?
Most stale facts come from change events nobody connected to the web: a price change, a leadership change, a product retirement, an office move. Add an "external facts" step to each of those processes.
Illustrative example: the price change that never left 2023. A software company raises prices and updates its pricing page. Its old price still sits in a comparison article, two review profiles and a partner page. Assistants keep quoting the old price for months, and sales spends calls explaining the difference. A one-page checklist attached to the price-change process would have caught all four.
For companies with many markets and languages, the list of places to update grows quickly. OwnlyReal runs this as an ongoing process inside the client's CMS and approvals, tracing stale claims to their sources and fixing them; see how it works.
Questions people also ask
Why does ChatGPT have outdated information about my company?
Either the model learned the old fact before its training cutoff, or when it searches it finds old pages that still state it. The second is usually fixable within weeks.
How do I update my company information in ChatGPT?
You can't edit ChatGPT directly. Update or retire every page that states the old fact, publish a dated current-facts page, fix your profiles and listings, and ask third parties to update theirs.
Why does ChatGPT still name our former CEO?
Old press releases, bios and directories still name them, and older models learned it before the change. Publish an explicit leadership-change statement and update every profile that lists executives.
Will deleting old pages fix outdated AI answers?
It helps if engines were citing them, but redirects or clearly dated history pages are often better than plain deletion, because they tell engines what replaced the old fact.
How long does it take for AI to show updated information?
Search-based answers can change after corrected sources are recrawled; facts in model memory change only with new model versions. Re-test weekly to track it.
