Correcting Outdated AI Information About Your Company Past a Knowledge Cutoff

Every AI answer about your company comes from one of two places.

The first is trained knowledge. This is what the model learned before its knowledge cutoff. OpenAI's model documentation, as checked in August 2026, listed a cutoff of February 16, 2026 for the GPT-5.6 family and August 31, 2025 for chat-latest, the Instant model ChatGPT uses. So "ChatGPT" is not one knowledge state. A routing change can move the cutoff by months. Anything that happened after the cutoff does not exist for an answer given without search.

The second is retrieval. The engine searches, fetches pages and writes from what it read. As of October 2026, retrieval fires per query and at the engine's discretion. ChatGPT decides when to search. Claude.ai web search is a user toggle. Google AI Overviews often do not trigger at all.

Say you run a SaaS company and changed your pricing in May 2026. A retrieval-off answer will quote the old price, because that is all the model learned. A retrieval-on answer can quote the new price, but only if it finds and reads a page that states it.

For more background, read why AI knows your old news. If you are not sure whether an answer is stale or simply invented, start with how to tell a hallucination from an outdated answer.

Be plain with your team about the limits.

• Trained knowledge is frozen. Nothing you publish today reaches a retrieval-off answer from the current model. It changes only when the vendor ships a model with a later cutoff.
• GPTBot is not a shortcut. OpenAI documents GPTBot as its training crawler. Allowing it does not put your new facts into today's model, and nobody has measured its effect on future ones. Blocking it does not erase what the model already learned.
• You don't choose when search runs. The engine decides, or the user does.
• You don't choose what gets cited. Retrieval and citation are run by the engine. You can only make sure the right page is eligible.

What is in your control is narrower and more reliable: whether a current, fetchable, readable statement of the fact exists. Older press coverage is a separate problem, covered in whether old news stories shape AI answers. If you want to know what vendors offer for direct corrections, see whether you can ask OpenAI or Google to fix wrong information.

Work through these in order. Each step removes a block that would stop an engine from reading the current fact.

1. List what changed. Write down each fact that moved, with the date: prices, plan names, product names, leadership, location, ownership, discontinued features.
2. Pick one home page per fact. State the fact in a single plain sentence near the top, for example "The Team plan costs $40 per user per month." A specific stated fact on a fetchable page gives the engine a sentence it can quote.
3. Remove the old version from your own site. Search your pricing page, about page, footer, help docs, old blog posts and PDFs. If two pages disagree, an engine can pick either.
4. Put the fact in the raw HTML. Vercel found that ChatGPT and Claude crawlers fetch JavaScript files but don't execute them, so they can't read client-side rendered content. View the page source or fetch it with JavaScript off. If the sentence only appears after the page loads, render it on the server. Googlebot renders JavaScript, so Google may see a fact that other engines miss.
5. Check robots.txt per bot. Rules are per crawler. OpenAI states that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers. Confirm OAI-SearchBot, PerplexityBot, Claude-SearchBot, Googlebot and bingbot are allowed.
6. Check the firewall and CDN. OpenAI's help centre says eligibility needs both: robots.txt permission and a host or CDN that allows its published crawler IPs. From 2026-09-15, Cloudflare blocks Training and Agent crawlers by default on ad-displaying pages for some accounts, while Search stays allowed. The AI Crawler Checker requests a page as each AI crawler and shows which ones get turned away.
7. Check index and snippet eligibility. Google says a page must be indexed and eligible to show with a snippet to appear as a supporting link in AI Overviews or AI Mode. Make sure noindex, nosnippet, max-snippet:0 or data-nosnippet is not sitting on the fact. Check Google Search Console and Bing Webmaster Tools separately.
8. Fix third-party pages. Update directory listings, marketplace profiles, partner pages and review profiles you control. When engines cite other sites instead of yours, being named correctly on those sites is associated with appearing in the answer.

A minimal robots.txt that allows the main search crawlers looks like this:

User-agent: OAI-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Claude-SearchBot
Allow: /

Run through this list before you test any AI answer. Each item is a yes or no.

• Every changed fact has one page that states it in a single sentence.
• No other page on your site states the old value.
• The sentence appears in the page source with JavaScript off.
• robots.txt allows OAI-SearchBot, PerplexityBot, Claude-SearchBot, Googlebot and bingbot.
• Each of those crawlers gets a 200 response, not a 403 or a challenge page.
• robots.txt itself is reachable by those crawlers.
• The page has no noindex, nosnippet or max-snippet:0, and no data-nosnippet wraps the fact.
• The page is indexed in Google and in Bing.
• Third-party profiles you control show the current fact.
• You logged the date of each change, so you can compare before and after.

Test with search on, logged out, and with memory off, so personal context doesn't skew the answer. Ask the question a buyer would ask, such as "How much does the Team plan cost?" Then read the answer and its citations.

Without visible citations, you can't know whether search ran. Treat an uncited answer as unknown, not as proof your fix failed. Judge each engine on its own, because they draw on different indexes. For the ChatGPT side specifically, see fixing ChatGPT errors in trained knowledge versus search. For a quick snapshot before you start, the free checker on this site runs up to three buyer questions and scores the answers.

Correcting facts is not a one-time job. Repeat it on these triggers:

• Every time a fact changes. Run steps 2 to 7 the day you publish the change.
• Once a month. Spot-check your top buyer questions with search on, across each engine you care about.
• After a model release or default change. Cutoffs move, and OpenAI's documentation warns its model snapshots "will be regularly updated." Retest retrieval-off answers to see what the new model learned.
• After any site, CDN or firewall change. A migration or a new bot rule can silently block crawlers that were working last month.

Keep a simple log: the fact, the date it changed, and what each engine said on each check. That log turns vague complaints into a list of specific fixes. On timing expectations, read how long it takes AI to update what it says about your brand.

Much of the advice in circulation has no evidence behind it.

• Bumping dates. A dateModified or lastmod value is self-declared. No engine states it weights it, and changing a date without changing the fact proves nothing.
• Adding schema to get cited. Google says structured data isn't required for generative AI search. Ahrefs' matched test found AI Overview citations fell 4.6% after pages added schema. Existing markup does no harm, so there is no need to remove it. It just isn't a lever.
• Publishing an llms.txt. Google says Search ignores it, and no other engine documents reading it.
• Rewriting for an AI-friendly style. Google says you don't need to write in a specific way for generative AI search.
• Hidden text or cloaking for AI crawlers. Showing crawlers content users can't see is manipulation. It risks penalties and erodes trust. Don't do it.

Spend the time on the access steps above instead.