A prospect asks ChatGPT for software recommendations and sees your competitors, but not your company. That can be a useful signal to investigate, but one answer cannot tell you why your brand was absent. Prompt fit, public evidence, retrieved sources and variation between answers are all possible explanations. The practical next step is to test the same buyer questions consistently, then compare what the answers say with evidence a reader can verify.
Why does ChatGPT mention competitors but not your brand?
A competitor may fit the wording of the prompt more closely, have stronger public associations with the requested category, appear in sources retrieved for that question, or simply be selected in a variable response. Those are plausible explanations, not established causes for your company without controlled observations. A brand mention is also different from a recommendation or a cited source.
A screenshot captures one response under one set of conditions. It does not show whether the same competitor appears for other prompts, on another date, or with a different model configuration. Research in the repeated-query preprint documents meaningful variability in the names returned. Record the prompt, configuration and run context rather than trying to infer them after the fact.
The absence itself is not proof of an algorithmic penalty, weak product-market fit, or a universal technical problem. It is a starting point for diagnosis. First ask whether the prompt describes a buyer need your product actually serves. Then inspect which sources support the claims made about competitors and whether comparable, specific evidence about your product is publicly accessible.
What does existing research actually say about AI brand visibility?
Observational studies associate branded web mentions with greater AI brand visibility in their sampled datasets, but they do not show that adding mentions causes a model to recommend a company. The later cross-platform analysis also used a filtered sample, so its results cannot diagnose every B2B SaaS brand. A correlation means two measured variables tend to vary together; it does not establish that one caused the other.
Brand associations versus causal levers
The available findings can help frame questions, but they do not establish a prescription for an individual company. The numbers below describe associations in particular datasets, not expected results from a marketing action.
Finding | Population or setup | Permissible inference | Not established |
|---|---|---|---|
Ahrefs' earlier AI Overviews study reported a correlation of 0.664 between web mentions and brand visibility, compared with 0.218 for backlinks. | The study began with 75,000 brands selected using Domain Rating above 40 and a highest-volume keyword with at least 800 monthly searches; roughly 26% had no AI Overview mentions, and the analysis studied the remaining approximately 74%. | In that dataset, web mentions had a stronger measured association with visibility than backlinks did. | That mentions caused visibility, that the same relationship applies to ChatGPT, or that earning mentions will displace a named competitor. |
Ahrefs' later cross-platform analysis reported branded-web-mention correlations around 0.66 to 0.71 across ChatGPT, AI Mode and AI Overviews. | The cross-platform analysis filtered for domains with Domain Rating above 40 and a highest-volume keyword with at least 800 monthly searches. | Branded web mentions were associated with visibility in the selected sample and systems studied. | That the pattern generalizes to smaller or less-established SaaS firms, or identifies the cause of one brand's absence. |
A 2026 repeated-query preprint examined 4,500 responses to 50 buying questions across six configurations, with 15 calls per cell. | A repeated-query audit using the study's questions and configurations. | Repeated runs and explicit run context can reveal variation that a single answer hides. | That its results predict every company's prompt-level visibility or capture every possible answer. |
The preprint reports median expected ten-call recovery of 92.6% to 95.2% of the names observed in the study's 15-call set using its historical dictionary. After candidate names were re-adjudicated, the reported range was 89.5% to 94.7%. Its name extraction was AI-based, with no human reference study. These are sample-name recovery rates relative to that study's observed set. They are not a guarantee of complete recall in your audit, a measure of factual accuracy, or a recommended number of runs.
Taken together, these studies support careful measurement and hypothesis testing. They do not show that purchasing links or mentions, publishing more pages, or changing a single technical setting will cause ChatGPT to select your brand. Treat each proposed mechanism as a question to investigate against your own prompt-level evidence.
How can you test whether the gap is real?
Use a fixed set of unbranded buyer-intent prompts, repeat them across declared engines or configurations and dates, and record which brands and evidence appear in each response. This is a sampling exercise, not a census of all possible answers. Keep the wording and inclusion rules stable enough to compare runs, while documenting any change in model, retrieval availability or geography.
Build a prompt and response log
Use a simple protocol that another team member could repeat:
Choose representative buyer jobs. Write questions around real evaluation tasks, such as finding a platform for a particular workflow or comparing solutions under a stated constraint. Include category language buyers use, not only your product's preferred terminology.
Freeze the prompt wording and inclusion rules. Save each exact prompt. Decide in advance what counts as a brand mention, a recommendation and a citation so the team applies the same definitions across runs.
Repeat the tests and record context. Log the date, engine, model or configuration, geography when known, and whether retrieval or browsing was available. There is no magic number of runs that guarantees a complete picture; choose a repeat plan suited to the decision and report its limits.
Record the answer, not just the brand count. Capture named companies, recommendation language, rationale, cited URLs, and factual errors. Preserve enough of the response to review how a name was used.
Compare variation by prompt cluster. Separate questions by buyer job or category. A change across runs may be concentrated in one type of prompt rather than reflecting a general visibility shift.
Prompt | Run context (engine/configuration, date, geography) | Named competitor | Your brand mentioned? | Recommendation? | Cited URL | Accuracy and notes |
|---|---|---|---|---|---|---|
Exact saved prompt | Record each run | Record each name | Yes / no | Yes / no, with rationale | Copy the observed URL or record none | Note unsupported, outdated or incorrect claims |
The repeated-query preprint illustrates why the log matters: its authors examined 4,500 responses across 50 buying questions and six configurations, with 15 calls per cell. Its recovery calculations compare the expected number of names in a sampled ten-call subset with names observed in the corresponding 15-call set. They describe the study's sample and method, not the completeness you should expect from another audit.
A single answer can still be worth saving as an observation. Label it as one run, though, and do not turn it into a diagnosis or claim of a lasting competitive gap.
Which public evidence gaps should a SaaS team inspect?
Compare the claims for which a competitor is recommended with your own publicly accessible, verifiable product facts, third-party coverage and documentation. Check whether the evidence is specific to the buyer's question and can be accessed without relying on an internal sales conversation. A source appearing beside a rival does not prove that copying its format or seeking the same kind of mention will change future answers.
Compare like-for-like claims, not just brand counts
A name count cannot tell you whether the answer described the product accurately or matched the buyer's requirements. Build a comparison around a claim and its supporting evidence:
Buyer question | Rival claim and cited source | Your verifiable equivalent | Missing or inaccurate fact | Owner and next action |
|---|---|---|---|---|
What does the buyer need to know? | Record the exact claim and the URL shown in the response. | Link the claim to current public documentation or a product page. | State what is absent, unclear, outdated or incorrect. | Assign a product, security, compliance or marketing owner and a concrete check. |
One technical check is whether the relevant pages appear in Bing as well as Google. OpenAI says ChatGPT Search for Enterprise and Edu may share disassociated search queries with Bing. This does not establish that every ChatGPT Search response uses Bing, or that a page missing from Bing caused a brand's omission. OpenAI's Enterprise and Edu search guidance describes the Bing relationship and its scope.
Check Bing index status alongside OpenAI's guidance on OAI-SearchBot access. Treat any discrepancy as a diagnostic lead to test against the same prompts and cited sources, not as a diagnosis on its own.
For a B2B software product, the comparison might examine integration support, security documentation, pricing scope, deployment constraints or customer evidence. These are examples of evidence categories to inspect, not claims that any particular competitor has an advantage. Have a product or compliance owner verify sensitive and operational claims before publication. Do not fill a gap with an unverified feature or customer result.
Are citations, mentions and recommendations the same thing?
No. A mention is a brand name appearing in an answer; a recommendation explicitly selects or favors a product for a buyer need; and a citation is a linked source presented in support of text. Factual accuracy is a separate check, because a visible mention or citation does not by itself establish that the answer describes the product correctly.
Record these as separate fields. A company can be mentioned without being recommended, recommended without a visible citation, or cited in a source that does not support the answer's wording. Keeping the categories distinct prevents a visibility count from being mistaken for endorsement, accuracy or qualified demand.
What should you change first, and how will you know?
Correct inaccurate or missing public product evidence first, then retest the same prompts over repeated runs. Prioritize buyer-critical factual gaps over broad, unmeasured generative engine optimization (GEO) activity. A before-and-after shift is directional evidence, not proof of causality when models, retrieval, prompts or other conditions may have changed.
Prioritization and reporting
For each proposed change, assess five practical factors:
Claim accuracy: Is the current public information wrong, incomplete or ambiguous?
Buyer relevance: Does the gap concern a decision buyers need to make?
Source accessibility: Can an external reader and the relevant systems access the evidence?
Accountable owner: Who can verify and maintain the information?
Effort: What is required to correct it, and what else would the work displace?
Report mention rate, recommendation rate, source inclusion and factual error rate separately. Break results out by prompt cluster and engine, and include the observation period, sample size and run conditions. These measures describe what the audit observed; they do not, on their own, show whether visibility produced qualified leads or revenue.
If the comparison points to a factual-description problem, use audit AI answers about your company to assess the claims. If the issue appears technical, check crawler access before concluding that the product is irrelevant. For broader measurement, track brand share of voice in AI search with a defined panel and denominator.
Conclusion
A missing brand in one ChatGPT answer is a signal to investigate, not a diagnosis. Repeat fixed buyer prompts, record the conditions and distinguish mentions, recommendations, citations and accuracy. Then compare competitor claims with public product evidence and correct buyer-critical gaps first. This gives the team a defensible basis for a focused AI visibility audit without promising that any change will secure a recommendation.
Frequently Asked Questions
Why is my brand not showing up in ChatGPT?
One response cannot establish why a brand is absent. Check whether the prompt matches a buyer need your product serves, repeat the same prompts under recorded conditions, and compare the answer's sources with your own accessible, verifiable product evidence.
How Nortav Can Help
If competitors appear in ChatGPT answers and your brand does not, Nortav can help structure a careful diagnosis of the prompts, sources, and public evidence without treating one response as a cause.
Our approach starts with understanding your current search visibility, identifying gaps against competitors, and developing a strategy focused on driving qualified traffic, generating leads, and delivering measurable business value.
Book a Discovery Call →
