AI Is Getting My Brand Wrong: A Fix Plan

An AI answer about your company comes from one of four stages. Each one fails differently.

• Parametric (trained) knowledge. With no web search, the model answers from what it learned before its cutoff. OpenAI's model docs list a February 16, 2026 cutoff for the GPT-5.6 family and August 31, 2025 for the Instant path behind chat-latest. Anything you changed after that date is invisible to an ungrounded answer.
• Access. A crawler has to be allowed in, get past your firewall, and read the content in the HTML it receives. If any of those fail, the engine never stores your correct facts.
• Retrieval. At answer time, the engine decides whether to search and which pages to pull. It often rewrites the user's question into its own queries first.
• Generation. The model writes the answer from whatever it pulled. If it pulled a stale directory listing or a competitor's comparison page, it repeats that.

So "AI is wrong about us" is usually one of three things: an outdated trained answer, a site the crawler couldn't read, or a third-party source that is wrong. The fix depends on which. Our guide to hallucinations versus outdated answers covers how to tell them apart.

Be honest with your team about this split before anyone promises a fix.

In your control

• Which crawlers your robots.txt allows, bot by bot.
• Whether your CDN or firewall lets those crawlers through.
• Whether your key facts are in the server-rendered HTML.
• Whether your own pages agree with each other.
• Snippet and index controls such as noindex, nosnippet, max-snippet and data-nosnippet.
• Requests to update third-party listings you own or can edit.

Not in your control

• What a model learned in training. That changes only when the vendor ships a new model.
• Whether an engine searches for a given question. ChatGPT search runs at the model's discretion, and Claude.ai search is a user toggle.
• Which retrieved pages the model cites, and in what order.
• Corrections typed into a chat. Those don't update the answer for other users, as explained in what happens when you correct ChatGPT in a chat.

The July 2026 critical survey of 45 GEO studies (arXiv:2607.14035) found no technique with a stable, cross-platform causal effect on whether content gets discovered. Your job is to remove the blocks and make the correct facts available. You can't force the answer.

Work through these in order. Each step depends on the one before it.

1. Record the errors per engine. Write down the exact wrong claim, the engine, the date, and whether the answer showed sources.
2. Check crawler access. Confirm robots.txt and your firewall let each retrieval crawler in.
3. State the facts plainly in fetchable HTML. Put prices, names and key numbers in the page source, not only after JavaScript runs.
4. Resolve contradictions on your own site. One fact, one value, everywhere.
5. Check index and snippet status. Look at Google Search Console and Bing Webmaster Tools separately.
6. Update third-party listings. Fix the pages the engines actually cite.
7. Re-check each engine. Compare against your first record, engine by engine.

Steps 2 to 4 are where the evidence is strongest. Step 6 matters most when an engine cites someone else's page instead of yours.

Don't start fixing from a screenshot someone sent in Slack. Build a short log first.

For each error, capture the question asked, the engine, the date, the exact wrong claim, the correct fact, and any cited URLs. Test logged out with memory off where you can. ChatGPT may use saved memories when it rewrites a search query, so a logged-in test isn't reproducible.

The cited URLs are the most useful field. They tell you which stage failed.

Keep engines separate. One engine can cite a page thousands of times while another barely touches it, so a blended number hides where the problem is. For engine-specific steps, see fixing ChatGPT errors, correcting Perplexity and wrong facts in AI Overviews. The free checker on this site gives a quick snapshot for up to three buyer questions if you want a starting point.

An engine can't quote a correction it can't fetch. Access has three gates, and all three must pass.

robots.txt, per bot

Rules apply to each crawler separately. OpenAI's crawler documentation says GPTBot covers training, OAI-SearchBot covers ChatGPT search, and "each setting is independent of the others." Blocking GPTBot doesn't remove you from ChatGPT search. Allowing it doesn't put you there. The common mistake is a blanket AI block that also shuts out the search bots.

User-agent: OAI-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Claude-SearchBot
Allow: /

Firewall and CDN

A firewall or CDN can return a 403 to a crawler that robots.txt allows. OpenAI's help centre says eligibility needs both: allow OAI-SearchBot and confirm your host lets its published IP addresses through. As of September 15, 2026, Cloudflare blocks Training and Agent crawlers by default on ad-displaying pages for new domains and existing free-tier customers, so a site can lose access without anyone editing it. The AI Crawler Checker requests a page as each AI crawler and shows which ones your firewall turns away.

Google

There is no robots.txt token for AI Overviews or AI Mode. Google says a page must be indexed and eligible to show with a snippet to appear as a supporting link.

Put the fact in the HTML

Vercel's crawler study found that ChatGPT and Claude crawlers fetch JavaScript files but don't execute them, so they can't read client-side rendered content. If your pricing table, plan names or product list only appear after JavaScript runs, those crawlers see an empty space. Googlebot does render JavaScript, so this mainly affects non-Google engines.

To check, view the raw page source (not the inspected page) and search for the fact. If it isn't there, ask your developers to server-render that content.

Say it in one sentence

Write the fact as a plain, specific statement on a page anyone can fetch: "The Starter plan costs $X per month." "We were founded in 2019 in Austin." A specific fact gives an engine a sentence it can quote. That is the only positive experimental signal for on-page content in the evidence. Rewriting whole pages in a special "AI style" is not: Google says you don't need to write in a specific way for generative AI search.

One fact, one value

Check your pricing page, about page, footer, help centre, press kit and any structured data. Say your pricing page shows the new price but an old help article still lists last year's. An engine that retrieves the help article will repeat the old number. Delete, redirect or update the stale version. For pricing specifically, follow the pricing error playbook.

Also check for accidental suppression. A template-level nosnippet or data-nosnippet keeps content out of Google and Bing AI answers.

When engines cite other sites instead of yours, those sites are the route into the answer. The evidence shows a strong association here, not a guaranteed effect.

Start from the cited URLs in your error log. Typical culprits are directory listings, review profiles, marketplace pages, old press coverage and comparison articles.

1. Update every listing you control: directories, app marketplaces, review profiles, social bios.
2. Ask publishers to correct factual errors in articles. Send the correct fact and a link to the page that states it.
3. For a competitor's comparison page, the options are narrower. See what to do when a competitor's comparison misrepresents you.

Then re-test each engine and compare against your original log. Search-on answers can change once the source changes and gets re-crawled. Search-off answers won't change until the vendor ships a new model.

• Adding schema to fix the answer. Google says structured data isn't required for generative AI search. Ahrefs' matched test of 1,885 pages found AI Overview citations fell 4.6% after adding schema, though every page studied was already heavily cited. Existing markup does no harm. It just isn't a lever for AI answers.
• Publishing an llms.txt. Google says Search ignores it, and no other engine documents reading it.
• Allowing GPTBot to fix ChatGPT search. GPTBot governs training only. Search uses OAI-SearchBot.
• Blocking every AI bot at once. That removes you from search answers that might carry the correct fact.
• Mass-publishing pages. In OtterlyAI's 2026 test, Google deindexed two new sites of about 1,000 AI-generated posts each, with no manual action shown in Search Console.
• Hidden text or cloaking for AI crawlers. Injected instructions and content shown only to bots are manipulation. They risk spam action and erode trust in your real facts.
• Bumping update dates. A new dateModified with no real change proves nothing and fixes nothing.
• Trusting one blended score. Measure each engine on its own, with retrieval state noted.

How long does it take for a correction to show up in AI answers? It depends on the stage. Answers that use live search can change after the engine re-crawls the corrected page or source, and nobody publishes how long that takes. Answers from trained knowledge won't change until the vendor releases a model with a later cutoff. Log the date you made each fix, then re-test each engine regularly and compare against your original record.

Can I ask OpenAI or Google to correct wrong information about my company? Vendors offer feedback buttons inside their products, but none of the evidence here documents a process that edits a model's answer about a specific brand on request. The dependable route is fixing what the engines read: your own fetchable pages and the third-party sources they cite. Our page on [asking AI companies to correct brand info](/questions/ask-ai-company-correct-brand-info) goes into the options.

Is the AI hallucinating or just using old information? Check the sources. If the answer shows no citations and matches a fact that was true before the model's cutoff, it's probably outdated trained knowledge. If it cites a page that states the wrong fact, the source is the problem. If it states something with no source and no past basis, treat it as a fabrication and make the real fact easier to find.

Does blocking GPTBot remove wrong answers from ChatGPT? No. OpenAI says GPTBot controls whether content may be used for training, and OAI-SearchBot controls ChatGPT search. The settings are independent. Blocking GPTBot doesn't delete anything already learned, and it doesn't affect search answers. Blocking OAI-SearchBot would stop ChatGPT search from showing your pages, which removes your correct facts rather than the wrong ones.

Should I add FAQ schema so AI picks up my correct facts? The evidence doesn't support it. Google says structured data isn't required for generative AI search and deprecated FAQ rich results in May 2026. Ahrefs' matched test found AI Overview citations fell 4.6% after pages added schema. Keep existing markup if you have it, but put your effort into plain, consistent facts in server-rendered HTML.