Asking Third-Party Sites to Fix Wrong Facts About Your Brand

AI answers about your brand come from two places. Some are generated from the model's training data with no web search at all. Others are grounded: the engine retrieves pages at answer time and writes from them. As of October 2026, retrieval is decided per query on every major platform, so the same question can produce either kind of answer.

When a grounded answer cites a directory, a review site or an old article that states the wrong fact, that page sits inside the engine's context. Fix the sentence and the next retrieval reads the corrected version. The evidence shows a strong association between what cited sites say about a brand and what the answer says. Nobody has published how often a correction flips the answer, so treat this as the best route you control, not a promise.

Two limits matter:

• Answers from memory. A correction never reaches an answer generated without search. That only changes when a future model is trained, which is why outdated facts past a knowledge cutoff need a different plan.
• Correcting the chatbot. Telling an assistant it is wrong inside a conversation is not the same as fixing the source. See whether correcting ChatGPT in a chat fixes it for everyone.

Fix your own site before you ask anyone else to fix theirs. If your pricing page and your about page disagree, a third-party editor has no clean fact to copy. The free check for contradicting facts on your own site is the place to start.

Work from evidence you can show an editor. Follow these steps for each wrong fact.

1. Record the error exactly. Copy the wrong sentence word for word. Note the platform, the date and the question you asked.
2. Capture the citations. Ask the same question on a surface that shows sources, such as ChatGPT search, Perplexity or Google AI Overviews. Save every cited URL.
3. Ask several phrasings. OpenAI says ChatGPT search rewrites a prompt into one or more targeted queries, so different wordings can pull different pages. Repeat step 2 with two or three versions of the question.
4. Open each cited page and find the sentence. A citation is not proof of influence. Research on how LLMs cite (arXiv 2606.28358) found citation behaviour could be changed without changing the sources. If no cited page contains the error, search the web for the wrong phrase verbatim.
5. Check which engines see the page. Engines draw on different indexes. In Otterly's 2,000 AI blogs experiment, Microsoft Copilot kept citing sites that Google had already removed from its index. The same wrong fact can persist on one engine after it disappears from another.
6. Rank your targets. Put pages cited across several questions and several engines first. A page cited once for an obscure phrasing can wait.

If an answer shows no citations at all, it may be coming from the model's memory. Do not chase a source that may not exist.

Editors act on requests that cost them little effort. Give them everything in one message.

The evidence pack

• The page URL and the exact wrong sentence, with a dated screenshot.
• The correct fact in one line, such as a price, a founding year or a current product name.
• A public proof URL: your pricing page, a press release, a registry filing. Make sure the fact appears in the page's HTML before JavaScript runs, because many AI crawlers do not execute JavaScript.
• Your name, role and why you can speak for the company.

Sending it

1. Use the owner's own route first. Claim the listing if the site offers a business portal. Otherwise use the corrections form or the editor's address. Community-edited sites have their own processes; follow them and disclose your conflict of interest.
2. Ask for one specific edit. Quote the old sentence and supply the replacement. "Please update our pricing" invites a guess.
3. Ask for removal of the stale fact. A page that states both the old and new price creates a contradiction.
4. Keep it short and polite. Three short paragraphs: what is wrong, what is right, where the proof is.
5. Follow up once after about two weeks if you hear nothing, then move on to the next target.

A hypothetical example: say you run a dental clinic in Austin and a directory lists your old address. Your message quotes the directory's address line, gives the new one, links your contact page and a local business registry entry, and asks them to replace the line.

Run through this list for every request.

• The wrong sentence is quoted exactly, with URL and screenshot date.
• The page was actually cited, or found verbatim, for a buyer question you care about.
• Your own site states the correct fact once, consistently, in plain HTML.
• The proof URL loads publicly with no login.
• The request asks for one replacement sentence, not a rewrite.
• The request asks for the stale sentence to be removed.
• You disclosed who you are and your relationship to the brand.
• You used the site's own correction or listing route.
• The request is logged with a date and a follow-up date.
• You have not asked for, offered or planned anything hidden: no hidden text, no instructions aimed at AI crawlers, no cloaked pages.

Good results

• The page shows your exact replacement sentence and the old one is gone.
• The corrected page stays live and indexed. 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, so a correction on a page that later drops out of the index helps less on those surfaces.
• Later grounded answers either cite the corrected page or stop stating the wrong fact.

Bad results

• The owner adds your fact but keeps the old one. Ask again for removal.
• The page is deleted but copies of the wrong fact survive on scraper sites. Trace and repeat for the copies that get cited.
• The page is fixed but the answer is unchanged with no citations shown. The answer is likely coming from memory, which no source edit reaches this quarter.
• The page is a rival's comparison page that refuses to change. That calls for a different approach, covered in what to do when a competitor's comparison page misrepresents you.

A result to avoid creating

Some people try to plant hidden text or instructions in pages so AI agents repeat them. Research on LLM search agents (arXiv 2606.16821) found manipulated web content can sway their answers, with results varying widely by model. That is a vulnerability, not a tactic. It risks penalties and reputational damage, and it is the kind of manipulation you may need to defend against when others do it to you.

A correction programme only works if you can see what changed. Keep one log with these fields.

Recheck the same questions on the same platforms a few weeks after a page is updated, because engines need to recrawl the page and Google says indexing and serving are not guaranteed. Run a full sweep monthly, and again after any price change, rebrand, product launch or move. Each event creates a fresh batch of stale pages.

Measure each engine separately. A fix can show up on one engine and not another because they read different indexes. For a quick baseline, the free checker on this site shows a snapshot of how AI answers three buyer questions about you. It is a snapshot, not continuous tracking, so keep your own log for trends. For the wider picture of on-site and off-site fixes together, see the fix plan for AI getting your brand wrong.

What if the site owner ignores my correction request? Follow up once after about two weeks through the same route, then stop. Log the page as ignored and check whether it keeps getting cited. If it does, make sure your own site and the other pages engines cite state the correct fact clearly, so retrieval has better sources available. You cannot force an edit, and pressure tactics tend to backfire with editors.

Should I ask for the page to be removed instead of corrected? Usually ask for a correction first. A corrected page that stays indexed gives grounded answers an accurate sentence to use. A removed page simply disappears from retrieval, and engines may fill the gap with another source, possibly a worse one. Ask for removal only when the whole page is wrong, such as a listing for a location you never had.

How long until the AI answer reflects the fix? No published study gives a timeline. The engine has to recrawl the page, keep it in its index, and retrieve it for the relevant question. Each step varies by engine and by query. Answers generated without web search will not change until a model is retrained. Recheck on a schedule rather than expecting a fixed date.

Does AI trust my website more than third-party sites? No engine publishes a rule that your own site outranks others for facts about you. Engines cite what they retrieve for a given query, and that is often a third-party page. Your site still matters as the proof editors link to. The detail is covered in [whether AI trusts your site over third-party sites](/questions/ai-trust-own-site-vs-third-party).

What if the wrong fact is my pricing? Pricing goes stale on third-party sites fast, so it is a common target. Publish one current price on a page AI crawlers can fetch without running JavaScript, then send requests to the directories and review sites that cite old prices. The specific steps are in [what to do when ChatGPT keeps getting your pricing wrong](/questions/chatgpt-gets-my-pricing-wrong).