Five findings that can reveal how AI systems see, describe, and compare your business
An AI visibility audit should not give your business a pass-or-fail grade. It should show you what to investigate next: whether your business appears, whether it is described accurately, whether the answer holds across repeated runs, whom the system treats as alternatives, and which buyer questions remain unanswered.
TL;DR
Look for five findings: absence, inaccurate representation, variability, unexpected competitors, and unanswered buyer questions.
Each finding points to a different problem. Collapsing all five into “we need more AI visibility” can send the business toward the wrong work.
One prompt produces an observation, not a measurement. Repeat the question under consistent conditions and record what holds, what changes, and which sources support the answer.
The purpose of the audit is not to prove that an AI system is right. It is to expose how your business may be understood before a buyer ever contacts you.
Why the Audit Matters Now
AI-assisted research is already part of the buying environment. Gartner surveyed 645 B2B buyers in August and September 2025 and found that 45% had used generative AI during a recent purchase, primarily to gather information about vendors and products. The same research found that 51% believed they were more likely to encounter misleading information from generative AI, and 69% preferred to validate AI-generated insights with a sales representative.
That combination matters. Buyers are using AI to learn about businesses, but many do not assume the answer is complete or reliable. Your audit should therefore examine more than whether your company’s name appears. It should examine the representation the buyer may encounter and the evidence available to validate it.
Pew Research Center’s analysis of browsing behavior provides another reason to care about the answer itself. In its study of Google activity from 900 U.S. adults, 18% of observed searches produced an AI summary. Searches framed as questions produced an AI summary 60% of the time. When an AI summary appeared, users clicked a source cited inside it in only 1% of visits.
Pew studied Google search behavior rather than B2B purchasing specifically, so the findings should not be generalized beyond that setting. They nevertheless show why the generated representation deserves attention: a user may absorb the summary without examining the underlying source.
Finding One: Your Business Is Absent
Ask the question a serious prospective buyer would ask without naming your company. If relevant providers appear and your business does not, record the absence. Do not immediately conclude that the system is wrong or that your marketing has failed.
Absence can have several explanations. The system may not have found usable information connecting your business to the buyer’s question. Your company, services, people, or expertise may be described inconsistently. The wording, geography, platform, retrieval method, or timing may also affect the result.
The next step is not simply to publish more. First determine whether you have publicly answered that buyer question, whether the answer is specific enough to distinguish your expertise, and whether supporting evidence can be found and connected to your business.
Record: the exact question, system, date, names that appeared, sources cited, and whether your company appeared in later runs.
Finding Two: Your Business Appears, but the Description Is Wrong
Being named is not automatically a favorable result. An answer may attach the wrong specialty, audience, geography, history, relationship, credential, or service to the right company.
An inaccurate description matters because it can create the buyer’s first working assumption about you. If the buyer does not inspect the sources, the error may survive long enough to affect consideration. If the buyer does investigate, inconsistent information can weaken confidence.
Capture the exact wording before trying to correct anything. Then compare it with the information available on your website, leadership profiles, trusted directories, case studies, and external references. Clearer and more consistent evidence may improve what systems have available to retrieve, but no publisher can guarantee how every AI system will represent that evidence.
Record: the inaccurate statement, cited or apparent source, correct fact, and published evidence that supports the correction.
Finding Three: The Answer Changes
Run the same question again later and you may receive different names, ordering, descriptions, or sources. That does not make the exercise useless. It reveals why a single answer cannot carry the weight of a measurement.
Generative systems can vary because of model updates, retrieval results, prompt context, personalization, location, and the probabilistic nature of generation. The important question is not whether every word remains identical. It is whether the material representation holds: Do you continue to appear? Are you associated with the same problem? Is the description substantially accurate? Do credible sources recur?
A favorable screenshot is an observation. Repeated observations under defined conditions begin to show whether your standing is durable or incidental.
Record: what remained consistent, what moved, and which changes would materially affect a buyer’s understanding.
Finding Four: The Consideration Set Is Different from Yours
The answer may name companies you do not consider direct competitors: larger national firms, adjacent specialists, directories, software platforms, publishers, or businesses serving a different segment.
Treat that as a view of the consideration set created for that particular question, not as a definitive map of your market.
The result may reveal that the question is too broad. It may also show that other organizations have published clearer or more retrievable answers around the buyer’s problem.
The strategic question is not simply, “Why did that company appear?” Ask what evidence connected it to the question and whether your own public information makes the relevant distinction visible.
Record: every named alternative, why it may have been included, cited evidence, and whether it appears across repeated runs.
Finding Five: Nobody Answers the Buyer’s Real Question Well
Sometimes the most useful result is not a missing company. It is a missing answer. The system may return generic advice, weak evidence, irrelevant providers, or no credible provider at all when the buyer asks a specific decision-stage question.
That is not proof that the topic is unclaimed across the market. It is evidence that, under the conditions tested, the system did not produce a strong answer. The opportunity is to investigate whether your business can publish a more direct, useful, and well-supported response.
An unanswered buyer question is often more valuable than another broad category topic because it exposes the decision the buyer is trying to make. Answer it with specificity. Explain when the approach fits, when it does not, what evidence matters, and what the buyer should ask next.
Record: the buyer question, weakness in the answer, sources used, and the evidence your business could responsibly contribute.
What the Audit Can and Cannot Tell You
The audit can reveal a direction for investigation. It can show that your business did not appear, appeared inaccurately, moved between runs, faced an unfamiliar consideration set, or occupied a market where the buyer’s real question remains poorly answered.
It cannot establish market share, buyer preference, causal impact, or improvement from one observation. It cannot tell you how common a result is unless the method defines the question set, platforms, conditions, repetition, and comparison period.
The discipline is simple: preserve the prompt, platform, date, answer, citations, and material interpretation. Then repeat the test without changing everything at once. The audit tells you where to look. A controlled baseline and repeated verification tell you whether anything is changing.
Run the Audit as an Investigation
Ask a real buyer question without naming your business.
Save the complete answer and every cited source.
Classify what you found: absent, inaccurate, variable, unexpectedly compared, or unanswered.
Repeat the same question under defined conditions.
Separate what held from what changed, and decide what evidence or content should be improved next.
If you run the audit, I would like to know what you find, especially after more than one run. What held? What moved? Was the description true?
Share the observation in the comments or send it to me directly. I will treat individual reports as diagnostic evidence, not as a representative survey.
If you want a more structured view of how your business appears across the questions that matter, Right Angle combines Answer Engine Optimization with Sightline to examine visibility, representation, citations, and competitive presence.
Coming next: what a fixed question set is, why holding the questions constant matters, and how to build one for your business.
Frequently Asked Questions
What is an AI visibility audit?
An AI visibility audit is a structured review of how AI-assisted search and answer systems respond to questions relevant to your buyers. It records whether the business appears, how it is described, what sources are used, which alternatives are named, and how the answer changes across repeated runs.
How many times should I repeat an AI visibility question?
There is no universal number that turns an observation into reliable measurement. Begin with repeated runs across defined dates and platforms, preserve the conditions, and avoid changing the prompt while testing stability. The appropriate sample depends on the decision you intend to make from the results.
What should I do if an AI system describes my business inaccurately?
Save the answer and its sources, identify the precise error, and locate the strongest public evidence supporting the correct information. Improve clarity and consistency across authoritative properties, then monitor the same question over time. No single update guarantees a correction across every system.
Does appearing in one AI answer mean my business is visible in AI search?
No. It means the business appeared in one answer under one set of conditions. A credible visibility conclusion requires a defined question set, repeated observations, recorded platforms and dates, and a distinction between appearance, accurate representation, citation, and recommendation.
Why should I record citations during the audit?
Citations show which published sources supported the answer and help distinguish a discoverability problem from an evidence problem. They also give you a way to evaluate whether the representation rests on current, relevant, and credible information.
B Randall Willis
Founder, Right Angle
