Why does AI still recommend a competitor that shut down?

Every AI answer comes from one of two places. Either the model answers from what it learned in training, with no web access. Or it runs a search, reads the pages it retrieves, and writes from those.

Both paths can recommend a company that no longer exists:

• Retrieval-off answers come from training data frozen at a knowledge cutoff. If your competitor shut down after that date, the model has no record of the closure.
• Retrieval-on answers come from whatever pages the engine fetched. If the top pages for "best tools for X" are listicles written while the competitor was trading, the answer inherits their picture of the market.

The fix differs by path. Our guide to search-on vs search-off answers explains the split in more depth. The short version: you wait out the first, and you work on the second.

A model's trained knowledge has a cutoff date. OpenAI's model documentation lists a February 16, 2026 knowledge cutoff for the GPT-5.6 family. The Instant path that ChatGPT uses (chat-latest) carries an August 31, 2025 cutoff, about five and a half months older. So "what ChatGPT knows" isn't one value. It depends on which model handled the question.

Say you sell project management software and a rival closed in May 2026. Every model trained before that date still knows the rival as an active product. Asked for alternatives without search, it may list them first.

Nothing a publisher does this quarter reaches a retrieval-off answer. The record changes only when the vendor ships a model trained on newer data. Our page on knowledge cutoffs and brand news covers how to date-check your own news against each model, and switching models explains why two models can disagree about the same company.

Turning search on doesn't guarantee a current answer. As of October 2026, retrieval is discretionary on most platforms. OpenAI says ChatGPT search typically rewrites your question into one or more targeted queries and sends them to search providers. Claude.ai search is a user toggle. Google AI Overviews don't trigger on every query.

When search does run, the engine works with what its index holds. Comparison posts, review roundups and directory listings often outlive the companies they describe. If those pages never mention the shutdown, the answer won't either.

Two further limits apply:

• Index eligibility. Google says a page must be indexed and eligible for a snippet to appear as a supporting link in AI Overviews or AI Mode. A page that isn't indexed can't correct the record there.
• Crawler access. OpenAI's crawler docs say sites opted out of OAI-SearchBot won't be shown in ChatGPT search answers. A blanket AI block can remove your current page from the candidate set, leaving only the old ones.

Not every wrong answer about a dead competitor has the same cause. Look for these:

• No citations shown. The answer probably came from trained memory, though you can't be certain from outside. Re-ask with search switched on and compare.
• Citations shown, all old. Retrieval worked, but the retrieved pages are stale. That's a third-party content problem, not a model problem.
• The model invents a successor or merger. That's closer to a fabrication than an outdated fact. Our guide to hallucination vs outdated answers helps you tell them apart.
• One engine is right, another wrong. Each engine draws on a different index and model. Measure them separately rather than blending them into one number.

You'll also see the reverse problem, where AI says a live company closed. The causes overlap.

You can't edit a model's memory. You can make the current facts easy to fetch and quote for every answer that does search.

1. Check crawler access per bot. Make sure robots.txt allows OAI-SearchBot, PerplexityBot and Claude-SearchBot. Then confirm your CDN or firewall doesn't return a 403 to them. The AI Crawler Checker requests a page as each crawler and shows which ones get turned away.
2. Put key facts in the raw HTML. Vercel found that ChatGPT and Claude crawlers fetch JavaScript files but don't execute them. Server-render your pricing, product names and positioning.
3. State specific facts plainly. A one-line sentence such as "We offer free data migration for former customers of discontinued tools" gives an engine something it can quote. Keep claims about the competitor factual and dated.
4. Get named where engines look. When answers cite third-party roundups, being named on those pages is associated with appearing in the answer. Ask editors of outdated lists to note the shutdown and review current options.
5. Re-test per engine. The free checker on this site shows whether engines name you for three buyer questions. Use our free competitor check to see who AI recommends instead.

The playbook for correcting outdated facts walks through each step. None of this guarantees a citation. It removes the blocks that are yours to remove.

Can I ask OpenAI or Google to remove a closed competitor from their models? There's no published process for editing what a model learned in training. Trained knowledge changes when the vendor ships a new model with a later cutoff. What you can influence is the search-on path: pages that engines can fetch, that state current facts, and that appear among the sources engines cite for your category.

Will blocking or allowing GPTBot change this? Not for current answers. OpenAI's crawler documentation says GPTBot governs training use only, while OAI-SearchBot controls whether a site appears in ChatGPT search. Each setting is independent. For search answers, the bot to check is OAI-SearchBot, plus whether your firewall lets it through.

How long until AI stops recommending the defunct competitor? It depends on the path. Search-on answers can change once the pages engines retrieve are updated and recrawled, though nobody controls the timing. Retrieval-off answers persist until a new model ships. Our page on [how long AI takes to update brand info](/questions/how-long-ai-update-brand-info) covers the timelines in more detail.

Should I write pages that tell AI the competitor is closed? A visible, accurate, dated statement on a fetchable page is fine and gives engines a sentence to quote. Hidden text, prompt injection or content shown only to AI crawlers is not. Research on search agents shows manipulated content can sway answers, which is exactly why engines treat it as abuse and why it puts your reputation at risk.