A research-backed look at why ChatGPT can show different versions of the same company, which sources matter, and what marketing and PR teams can actually control.Ask ChatGPT five questions about the same company and you may meet five slightly different versions of it.One answer explains what the company sells. Another compares it with competitors. A third brings up customer complaints. A fourth cites the company's own website. A fifth relies on a directory, publication, review page, or source the marketing team has barely thought about.This is what makes AI reputation different from the reputation problem brands already know.A company can edit its homepage. It can update its product pages and newsroom. It cannot edit an independent review, an old article, a Reddit discussion, or a directory it does not own. More importantly, it cannot decide which combination of those sources ChatGPT will use for the next question.So the question is not really, "How do we control what ChatGPT says about our brand?"A better question is:What creates the version of our brand that ChatGPT shows to a customer?The answer is more complicated than "your website" or "digital PR." Current research suggests that the source mix changes with the industry, the question, the search behavior, and even the way ChatGPT reasons through a task.For marketers, that makes AI reputation worth treating as its own problem.Your Brand Does Not Have One ChatGPT ReputationStart with a simple example."What does Company X do?""Is Company X reliable?""Company X or Company Y for a small business?""What are the biggest complaints about Company X?"Those are all questions about the same company. They do not require the same evidence.A product page may be useful for explaining features. It is less convincing when the question is about customer satisfaction. Reviews become more relevant there. A comparison question may send ChatGPT looking at several company websites, documentation, reviews, and independent comparison pages. A question about a past controversy may require news coverage and historical sources.OpenAI explains that ChatGPT search can rewrite a user's question into one or more targeted web searches. It may then refine those searches after seeing the initial results.That means the prompt itself can change the information ChatGPT goes looking for.AI reputation is not one answer. It is the pattern that appears across the questions people ask about your brand.Research from Semrush illustrates the problem. Its 2026 ChatGPT study covered more than 50,000 brands, 1,094 topic categories, and over 600,000 citations. Brand visibility changed considerably across five related prompts within the same topic. Only 15.2% of the categories in the study had what the researchers defined as a clear "owner."Looking good in one prompt can therefore give a false sense of security.One Screenshot Is Not an AI Reputation AuditThis matters because AI monitoring is often done in the simplest possible way.Someone opens ChatGPT.They type: "Best agencies for X."They take a screenshot.The brand appears, or it does not.Then the screenshot becomes evidence that the company's AI visibility is "good" or "bad."Recent research shows why that is risky.In September 2026, Glen Allsopp of Detailed tracked more than 70,000 ChatGPT and Google AI Overview responses over 28 days. For ChatGPT, the brand that appeared most often for a prompt still appeared on 82% of days. So there was usually a relatively stable leader.But everything around that leader moved much more.Allsopp separated repeatedly appearing brands into a "core" and less consistent brands into a "tail." On ChatGPT, core brands changed by 13% from one day to the next. The tail changed by 78%. Cited domains showed a similar pattern, with 14% day-to-day change in the core and 74% in the tail. The strictest test was even more striking. For a typical prompt, two randomly selected days produced exactly the same set of brands only 0.3% of the time. The exact same set of cited domains appeared on 2.4% of day pairs. The study notes that this is a deliberately strict measure, since one changed brand makes the whole set count as different. There was also substantial page-level turnover. Only about a quarter of the individual pages ChatGPT cited for a prompt were cited again the following day. This does not mean ChatGPT's answers are random.It means marketers should separate stable patterns from temporary appearances.Core reputationThe brands, facts, associations, and sources that keep returning.Reputation tailThe brands, sources, comparisons, examples, and claims that appear sometimes but not consistently.A one-off answer can show you something interesting. It cannot tell you which of those two you are looking at.The Question Can Change the Reputation You SeeThere is another problem with one-prompt monitoring.Different questions encourage different retrieval behavior.A 2026 analysis from Discovered Labs examined ChatGPT retrieval for B2B software questions. The researchers found that documentation and vendor pages became a major source category for factual software questions. Reddit retrieval became more concentrated around explicitly opinion-shaped prompts, such as requests for honest experiences or what people think about a product.That distinction makes intuitive sense."Does this software integrate with Salesforce?""What do users dislike about this software?"The company's documentation may be one of the best places to verify an integration. It is not enough to answer a question about customer dissatisfaction.The information need changed.Customer questionExampleEvidence ChatGPT may needIdentityWhat does Brand X do?Company website, About page, reference sourcesFactsHow much does Brand X cost?Pricing, documentation, official pagesComparisonBrand X or Brand Y?Company pages, documentation, specialist comparisons, reviewsExperienceWhat do customers think about Brand X?Reviews, communities, user discussionsReputationHas Brand X had major problems?Reporting, official statements, historical sourcesRecommendationIs Brand X a good choice for me?A mix of factual and evaluative sourcesThis is not a published ChatGPT ranking formula. It is a practical way to think about the different evidence required by different customer questions.The important part is that "What does ChatGPT say about us?" is too broad a question to be useful.There Is No Universal Source HierarchyThe next popular assumption is that ChatGPT always prefers independent third-party websites to company-owned information.The evidence is much messier.A Visibly study published in 2026 analyzed 13,216 citations across 2,636 ChatGPT answers to commercial questions in four UK industries.Company-owned websites represented:60.2% of citations in accountancy56.1% in fitness74.1% in recruitmentCommercial law looked completely different. Brand-owned sites accounted for only 22.1% of citations. Directories and rankings accounted for 76.3%. Chambers alone supplied 56.8% and Legal 500 another 18.9%.In fitness, publishers also mattered far more than in the other categories. Editorial sites supplied 30.3% of citations, with Runner's World alone responsible for 10.5% in the study.That is a much more useful result than saying "get third-party mentions."It suggests that the relevant source ecosystem depends heavily on the market.A commercial lawyer may operate in a category where recognized directories carry unusual weight.A recruitment firm may be in a category where company websites supply far more of the cited material.A consumer product may be surrounded by publishers, retailers, reviewers, and specialist testing sites.Before deciding what a brand needs for ChatGPT, first ask: Where does credible information normally live in this category?But Other Research Finds Third-Party Sources DominatingThis is where the story gets more interesting.Another 2026 study reached a very different headline.Research published on arXiv analyzed 167,551 URL-grounded citations for 128 brands across 12 markets and 13 languages. Across that dataset, 85.7% of citations pointed to domains the brands did not own, while 14.3% went to brand-owned domains. Wikipedia was the most-cited domain in 11 of the 12 languages studied.At first glance, that seems to contradict the Visibly numbers.It does not necessarily mean one study is wrong.They measured different datasets, markets, languages, brands, prompt types, and AI environments. The arXiv paper covered a wider multilingual set of LLM brand answers. Visibly specifically examined ChatGPT commercial buying questions in four UK industries.That difference is exactly why universal AI optimization rules should make marketers suspicious.There may not be one ideal ratio of first-party to third-party information.The more useful question is whether the sources that matter for your customers' questions contain enough accurate information about your company.Being Mentioned Is Not the Same as Owning the StoryAnother ChatGPT-only study makes this distinction clearer.Viziquo tested 131 B2B SaaS brands across 5,055 buyer-style ChatGPT conversations and analyzed 18,036 answer citations.When the brand was explicitly named in the prompt, it appeared in 99.9% of conversations.When users asked organic buyer questions without naming it, the tested brand appeared in only 12.2% of conversations.Recognition and discovery were clearly different.In organic conversations where the tested brand did appear, only 17.5% of supporting citations came from that brand's own site. Competitor-owned content supplied 43.9%, while other third-party sources contributed 38.6%.Think about what that means.ChatGPT may mention your company while another company is helping explain it.Suppose someone asks about a type of software and your brand appears. If the answer uses a competitor's comparison page to explain your features or limitations, you have visibility without full narrative ownership.Are we mentioned?Are we cited?Are we described accurately?What sources are doing the describing?Is the framing positive, negative, neutral, or simply outdated?Those should not be collapsed into one "AI visibility score."ChatGPT Does Not Simply Repackage Google's First PageTraditional reputation monitoring often begins with Google.Search the brand. Look at page one. Review the articles, company profiles, ratings, and results.That still matters. But ChatGPT's citation set is not simply Google's first page rewritten into prose.A September 2026 Suff Digital study compared 20,275 pages cited across 5,836 ChatGPT answers with Google's top 10 results for the same questions on the same day.It found that 87.7% of pages cited by ChatGPT were not in Google's top 10 for the equivalent question. At the domain level, 66.2% of cited websites had no page in Google's top 10. In 41.9% of ChatGPT answers, none of the cited websites appeared on Google's first page.The overlap also changed dramatically by topic. Only 20% of websites cited in software answers were missing from Google's top 10, compared with 88.5% for recipes and 82.9% for beauty. Again, the category matters.The sensible conclusion is not that Google rankings have stopped mattering.It is that a clean Google page one does not give you a complete picture of your ChatGPT reputation.AI reputation monitoring should look at ChatGPT directly.Even the ChatGPT Mode Can Change the SourcesThere is one more variable marketers rarely consider.The same question can produce different source sets depending on how much reasoning the system performs.Semrush and Kevin Indig tested 100 prompts across 20 buyer journeys in B2B SaaS, finance, consumer technology, health, and lifestyle. Search Engine Land reported that only 25.6% of cited domains overlapped between minimal and high reasoning for the same prompts.High reasoning also ran 1,130 web searches across the test set versus 245 in minimal reasoning. Cited responses averaged 4.5 sources in high reasoning compared with 2.6 in minimal reasoning. The source mix changed too. Government and academic sources gained share, as did official documentation and support pages. Reddit and other user-generated or review sources lost share in that test.A better question is: What sources does ChatGPT use for this question, in this context, with this search process?That is less satisfying than a simple optimization checklist.It is also more accurate.The Citations You See Are Not Necessarily the Whole SearchVisible citations are extremely useful for reputation analysis. They show which pages ChatGPT is using to support specific parts of an answer.They should not be treated as a perfect map of everything the system considered.OpenAI's web search documentation makes this distinction explicit for its API. Inline citations contain the most relevant references shown with the answer. A separate sources field can return the complete list of URLs consulted during the web search, and OpenAI notes that this list is often larger than the visible citation set. That matters when someone sees three citations and concludes: "These three websites determine what ChatGPT thinks about us."They do not necessarily tell the whole story.Visible citations are evidence worth auditing.They are not a complete explanation of the model's internal decision process.What Can a Brand Actually Control?Very little of the final answer.Quite a lot of the information environment.Those are different things.A practical framework is to separate the brand's online reputation into three areas.ControlThe company can directly manage its website, service descriptions, About page, pricing pages, executive bios, contact information, newsroom, documentation, official profiles, and other first-party information.Accuracy here should be basic hygiene.If the company itself publishes three different versions of what it does, when it was founded, who leads it, or which markets it serves, there is no reason to expect an outside system to resolve those contradictions perfectly.Technical access is part of this layer too.OpenAI says any public website can potentially appear in ChatGPT search. Publishers that want their content to be discovered, surfaced in summaries, cited, and linked should make sure OAI-SearchBot is not blocked. OpenAI also separates OAI-SearchBot from GPTBot, so search crawling and model-training preferences can be managed separately. Allowing access does not guarantee visibility. OpenAI says search placement is determined using multiple factors designed to surface relevant and reliable information, and placement is not guaranteed.Crawler access is best understood as eligibility, not optimization magic.InfluenceA company cannot edit an independent newspaper article or force a reviewer to change an opinion.It can still influence the quality of information available about itself in legitimate ways.Good media relations make accurate company information available to journalists. Useful original research gives publications something worth referencing. Strong customer service affects real customer experiences. Clear expert commentary can help a company become part of an industry conversation. Accurate directory submissions reduce basic factual errors.None of this guarantees a ChatGPT citation.It creates a stronger public information record.ObserveSome information needs to be monitored rather than controlled.Customer conversations. Critical reviews. Historical reporting. Independent opinions. Reddit threads. Old comparison pages. Public complaints.Trying to make every independent source positive would defeat the point of independent sources.But recurring patterns deserve attention.If several sources repeat the same outdated fact, that is a reputation problem.If customers repeatedly complain about the same service issue, that may be a business problem.If ChatGPT repeatedly associates your company with a category you left years ago, that may be an information freshness problem.AI reputation monitoring is useful partly because it can expose those gaps.Expert View: Trust Is Becoming Part of the Search ProblemSEO consultant and Growth Memo founder Kevin Indig has argued that AI search is shifting attention from pure relevance toward trust.In a 2026 presentation drawing on large-scale AI search research, he put it simply: "one thing trumps rank, and that is trust."His broader trust stack includes evidence, experts, validation, and reputation.That idea helps explain why AI reputation cannot be reduced to rankings.A company may rank well and still have weak independent validation.It may be frequently mentioned but described through a competitor's content.It may have excellent owned content but a serious review problem.It may appear in ChatGPT often, yet be associated with an old product or outdated criticism.Visibility tells you whether the brand entered the answer.Reputation tells you what happened once it got there.A Better Way to Audit Your ChatGPT ReputationA useful audit should measure more than whether the company appears in one answer.MediaPeachy can think about the problem across five dimensions:IdentityDoes ChatGPT understand what the company actually is? Check the business category, products, services, audience, locations, leadership, and basic company facts.AssociationWhat does ChatGPT connect the company with? Look for industries, competitors, customer types, price positioning, features, strengths, weaknesses, and recurring themes.SentimentWhat positive and negative information appears? Do not reduce this to a simplistic positive or negative score. Record the actual issues being discussed and whether they come from credible evidence.EvidenceWhat sources support the answer? Separate first-party pages, publishers, directories, review platforms, communities, government sources, and competitors. Open the citations. Check dates. Confirm that the pages actually support the claims.StabilityWhich findings survive repeated testing? Ask several related questions. Repeat important prompts over time. Separate persistent patterns from things that appeared once.A ChatGPT answer is a sample of your AI reputation. It is not your AI reputation itself.AI Reputation Checklist for Marketing and PR TeamsDefine the questions that matter. Start with real customer questions, not only your company name.Test several types of intent. Include identity, factual, comparison, customer-experience, reputation, and recommendation questions.Record the exact prompt and date. Without that information, later comparisons are much less useful.Review what ChatGPT says, not only whether the brand appears. Track positioning, competitors, strengths, criticisms, and inaccuracies.Open every important citation. Check whether the source genuinely supports the claim and whether the information is current.Separate owned and independent evidence. A company's own pricing page and an independent review play different roles.Look for repeated patterns. A recurring claim across prompts and dates matters more than a single unusual answer.Audit first-party information. Remove contradictions and outdated facts across company-controlled pages.Check OAI-SearchBot access. Do this if appearing in ChatGPT search matters to the business.Investigate the underlying cause before trying to optimize the answer. An inaccurate webpage needs a content fix. Repeated poor reviews may need a customer-experience fix. Missing independent evidence may be a PR problem.What Marketers Should Stop Assuming"Third-party websites always matter more than our own website." Sometimes they do. Sometimes company-owned pages dominate the citation set."Reddit is the secret to ChatGPT visibility." Its importance changes by query intent and dataset. In some recent commercial studies, its visible citation share was small."If we rank first in Google, ChatGPT will use us." There is overlap, but the two source sets can differ considerably."One ChatGPT test tells us where we stand." Response and citation volatility make that unreliable."Getting mentioned means we control the narrative." A brand can be mentioned while competitors or independent sources provide much of the supporting evidence."There must be a checklist that guarantees inclusion." OpenAI itself says placement in ChatGPT search is not guaranteed.Your Public Record Is Becoming Part of the ProductFor years, online reputation management largely meant watching search results, reviews, news coverage, and social discussion.ChatGPT adds a new interface to the same underlying information.The difference is that the customer does not have to inspect those sources individually. ChatGPT can retrieve several of them and produce a single explanation.That answer may change tomorrow.The source set may change.The question may trigger a different search path.A deeper reasoning mode may find different evidence.And another customer may ask a completely different question.This makes "controlling ChatGPT" an unrealistic objective.A better objective is to make sure the company has a public record that holds up from several angles.Clear first-party information.Accurate documentation.Independent reporting when there is something worth reporting.Real customer feedback.Credible expert commentary.Current directory information.Useful evidence.A willingness to investigate criticism rather than trying to bury it.Brands cannot write every answer ChatGPT gives about them.They can do much more about the information those answers have to work with.
