What does AI do when sources disagree about my company?

When two sources disagree about your company, the AI engine picks one, blends them, or hedges. Which of those happens is not documented by any engine. OpenAI's help page says ChatGPT "ranks search results using multiple factors intended to help users find relevant, reliable information" and that "placement is not guaranteed." That is the whole public disclosure.

It helps to separate where a wrong fact can come from:

• Training data (parametric). With web search off, the model answers from what it learned before its knowledge cutoff. Whatever conflicting versions were in that data stay frozen until the next model.
• Access. A crawler has to be allowed to fetch your page, and it has to be able to read the fact once it does.
• Retrieval. With search on, the engine runs its own queries and pulls a handful of pages into context. Your page may or may not be one of them.
• Generation. The model writes the answer from what it retrieved. If two retrieved pages disagree, its choice is internal.

You cannot pick the winner at the last two stages. You can reduce how many wrong versions exist, and make sure the right one is readable.

A conflict can show up in one engine and not another, or on Tuesday and not Wednesday. Three conditions drive that.

First, retrieval is discretionary. ChatGPT decides per query whether to search. When it does not, you get the training-data answer, which may hold an older or contradictory fact.

Second, engines draw on different indexes. In Otterly's 2026 experiment with two AI-generated sites, Google removed both from its index while Bing kept them. Microsoft Copilot, which grounds on Bing, accounted for 2,718 of the 3,934 citations Otterly tracked for one site, while Google's AI surfaces almost never cited it. The same logic applies to facts: an outdated directory listing indexed by one engine can surface there and nowhere else.

Third, the engine writes its own queries. OpenAI says ChatGPT search "typically rewrites your query into one or more targeted queries." You never see which pages those queries return.

What this means: check each engine separately, and note whether the answer cited sources. A blended score hides which engine is wrong and why.

The best evidence on how models choose between sources comes from lab studies, not live engines. A 252,000-trial factorial study (arXiv:2605.25517) found topical relevance and list position were the biggest drivers of being cited first. C-SEO Bench reached a similar view on position. Both tested pages already placed inside the model's context, with no live search index involved.

A 2026 critical survey of 45 GEO studies (arXiv:2607.14035) summed it up: "no reviewed technique shows a stable, longitudinal, cross-platform causal effect on organic discoverability." No publisher can move their page to position one in a model's context.

Two more cautions apply. Well-known brands and sources dominate citations in correlational studies, and nothing controls for that. And a citation is not proof of use: mechanistic research has shown that citation output can change without the sources changing.

What this means: there is no trick that makes your version of a fact beat a competing one. The practical goal is fewer competing versions.

Some errors are not source disagreements at all. Diagnose before you fix.

• The fact is older than the model. OpenAI lists a February 16, 2026 knowledge cutoff for its GPT-5.6 models and August 31, 2025 for the Instant path ChatGPT also uses. A price you changed in March 2026 can be wrong in a search-off answer and right in a search-on one. See how hallucinations differ from outdated answers and correcting facts past a knowledge cutoff.
• Your own site disagrees with itself. The pricing page, about page, footer, markup block and a 2023 blog post can each state a different number. Read more on whether mistakes on your own website cause AI errors.
• The right fact is invisible. Vercel found ChatGPT and Claude crawlers fetch JavaScript files but do not execute them. If your current price only appears after JavaScript runs, the crawler sees the old cached version elsewhere, or nothing.
• A third-party page is manipulated. Research on LLM search agents shows planted content can sway answers once retrieved. That is a reputational risk to watch for. Answering it with hidden text or cloaking is illegitimate and risks your own standing.

1. See what each engine says. Ask real buyer questions on each platform and record the answer, the cited sources, and whether search ran. Our guide to finding out what AI says about your company covers the method. The free checker on this site gives a snapshot for three buyer questions.
2. Remove contradictions on your own site. List your core facts (pricing, founding year, headquarters, product names, ownership) and find every page that states them. Run the free brand fact consistency check, then follow fixing contradictions on your site.
3. State each fact plainly in server-rendered HTML. A specific sentence ("Plans start at $29 a month") on a fetchable page gives an engine something it can quote.
4. Check access. Confirm robots.txt allows the search bots (OAI-SearchBot, PerplexityBot, Claude-SearchBot) and that your firewall or CDN does not return 403 to them.
5. Ask third-party sites to correct wrong facts. Directories, review sites, old press coverage and partner pages all count. Being named correctly on the sites engines cite is associated with appearing correctly in answers, though no effect is guaranteed. See how AI weighs your site against third-party sites.
6. Re-check per engine. Search-on answers can change once pages are recrawled. Search-off answers change only with a new model.