Published
Reading time: 20 min
Updated

Published
Reading time: 20 min
Updated
Ask ChatGPT the same question tomorrow and it may give you a different answer. Your business might appear where it was absent today, and a competitor appearing first tells you very little about whether customers are choosing them. A reliable AI visibility audit tests the questions customers ask before choosing a product or service. You repeat those questions across several AI systems and use the results to make a decision.
This guide explains how to run those tests and when to act on the results. It draws on research into how AI answers vary, documentation from the platforms and our own visibility reports.
It includes a sample spreadsheet layout and a way to decide whether to act now or schedule a follow-up test.
An AI visibility audit measures whether systems such as ChatGPT, Gemini and Perplexity mention, cite or recommend a brand when answering customer questions. You ask the same questions under consistent conditions, record each answer in full and identify where the business is missing. Use the findings to decide what to do next.
An audit measures visibility in a sample of answers. A missing brand may point to an opportunity, while any effect on revenue would need separate evidence. For the wider context, see our guide to AI search optimisation.
A useful audit gives you enough evidence to make a decision about budget and next steps. Start with four questions.
Decide how much effort to put into AI visibility before you start measuring, based on how your business makes sales. The audit gives you evidence to confirm that choice or revisit it. See our guide to AI visibility and SEO.
Agencies and tools measure different things. Our visibility reports cover seven areas. The metrics show where the brand is missing, and the other findings help you investigate possible causes and decide what to do.
| Area | What to check | Example question or signal |
|---|---|---|
| Metrics | Whether the answer names the brand (mention), uses its website as a source (citation) or suggests choosing it (recommendation) | The answer names the business and omits a link to its website |
| Topics and platforms | Which groups of questions and which platforms bring up the brand | The brand appears in ChatGPT and is absent from Gemini’s answers about pricing |
| Competitors | Which businesses appear alongside the brand or in its place | The same competitor comes up in most answers |
| Cited sources | Whether the answer draws on the brand’s website, directories, publications or competitors’ sites | An answer about the company’s services links to a third-party site |
| Brand understanding | Whether the answer accurately describes the business, its services and location, and identifies the right company | The assistant asks which company you mean |
| Technical accessibility | Whether crawlers can access the site and its pages are indexed | The robots.txt file blocks ChatGPT’s search crawler |
| Findings and decisions | Which gaps you found, what you decided, who will act and by when | See the section on recording your findings |
We explain the distinction between mentions, citations and recommendations in our AI search optimisation guide. For technical checks, use the platforms’ own documentation. OpenAI uses separate identifiers for OAI-SearchBot and GPTBot, so you can allow the search crawler and block the crawler that collects content for model training. OpenAI crawler documentation.
PerplexityBot helps surface and link to websites in search results. Perplexity-User may visit a page in response to a user’s question. Perplexity crawler documentation.
Google’s AI Overviews and AI Mode follow standard SEO requirements, including indexing. Google’s AI search documentation.
Fix issues with crawler access and indexing as soon as you identify them.
Start by choosing what to test, then repeat the questions under consistent conditions and record the answers. A shorter version is available in our guide to checking and monitoring AI visibility. For a full audit, follow these eight steps.
Use Search Console and Bing Webmaster Tools to supplement your tests. To see how ChatGPT, Gemini and Perplexity answer your customers’ questions, test those platforms directly.
Use one row per answer. Give mentions, citations and recommendations separate columns, and record enough detail to reproduce the test conditions. Summarise results by question and platform so the differences that inform your decision stay visible.
| Column | What to record | Example (fictional) |
|---|---|---|
| Date and round | The test date and round number | 12 November 2026, round 2 |
| Platform and mode | The AI system and session settings | ChatGPT, web search enabled, signed out |
| Language and location | The language of the question and the market you’re testing | Polish, Poland |
| ID, category and buying stage | The question number, type and buying stage | P07, supplier comparison, comparison stage |
| Question | The exact wording entered into the AI system | Which accountancy firms work with online retailers in Poland? |
| Answer | The full answer or a link to the saved copy | Link to the answer for P07 from round 2 |
| Mention | Whether the answer named the brand | No |
| Recommendation | Whether the answer recommended the brand and in what context | No |
| Citation of the brand’s website | Whether the brand’s website appeared among the sources | No |
| Cited URLs | Pages listed as sources | accountants-directory.example |
| Accuracy | Whether the description of the business was accurate | Not applicable (no mention) |
| Competitors | Who appeared in the answer | competitor-alpha.example, competitor-beta.example |
| Notes | Follow-up questions from the assistant, errors or unusual behaviour | None |
For each question, count the runs in which the brand appeared. Then group questions with the same intent and display the results for each platform side by side.
A consistent spreadsheet helps you tell whether visibility has changed across rounds or whether the answers are varying as they usually do.
For each platform, look at the share of answers that mention the brand across questions with the same intent. If it never appears in the first round, record zero mentions for that sample. When the brand is still absent in later rounds on different days under the same conditions, you have evidence of a gap on that platform.
In a January 2026 study, SparkToro and Gumshoe found that even brands that appeared in nearly every answer rarely held the same position. The order of brands almost never repeated. The share of answers that feature the brand is therefore more useful than its position in the list. The study was published without peer review, and Gumshoe sells an AI visibility monitoring tool. SparkToro and Gumshoe study.
In a 2026 preprint, Ronald Sielinski found that the precision of a domain’s share of citations depended on the platform and sample size. Across three consumer topics, the confidence interval reached a width of five percentage points at roughly 30–50 queries on Gemini and 90–100 on Perplexity. These figures apply to citation share in that study. Choose the number of questions for tracking mentions and recommendations based on the scope of your audit. Sielinski’s study.
Language and location matter too. In August 2026, Dmitrij Żatuchin ran 234 ChatGPT tests across four countries and six languages, using the signed-out interface and the API. In four of six questions, the top recommendation changed across six identical runs. In that small sample, the language of the question determined whether local suppliers appeared, and changing the IP address shifted the market that the recommended brands came from. Run tests in your customer’s language and from the market you sell to. Żatuchin’s study.
Treat the result as an estimate from a sample. Interpret it alongside competitor results and your own baseline.
An audit leads to one of three decisions. Fix critical issues, particularly technical ones, straight away. Act on recurring gaps in answers to questions that matter to sales if there is a budget for organic marketing. Where visibility is patchy, assess each gap before deciding whether to act or schedule another test.
| Situation | How to recognise it | Decision |
|---|---|---|
| Critical issues or inaccurate information | Blocked crawlers, missing pages in the index, outdated service information in answers, or confusion with another company | Fix straight away, regardless of scale |
| No visibility | Across several rounds on a given platform, the brand is absent from answers to commercially important questions that leave out its name, such as when customers compare suppliers before getting in touch | Act if there is a budget for organic marketing |
| Partial visibility | The brand appears in some answers | Use the table below to decide where to act and where to measure again |
In our experience, when visibility is patchy, the decision depends on where the brand is absent and what it would be worth to appear there. For each gap, assess the service’s contribution to revenue or margin and where the question falls in the customer’s buying process. Give more weight to questions about comparing or choosing suppliers than to general research questions.
| Service’s contribution to revenue or margin | Buying and comparison questions | General research questions |
|---|---|---|
| High | Act | Consider the cost of taking action and how often the gap recurs |
| Low | Consider the cost of taking action and how often the gap recurs | Schedule another test |
Fictional example. An accountancy firm offers three services. It comes up regularly in answers about bookkeeping for small service businesses, so the firm focuses its efforts elsewhere. It rarely appears for personal tax returns. That service contributes little revenue and customers rarely compare offers, so the firm schedules another test in three months. It is also missing from answers about accounting for online retailers. That service brings in the most revenue, and people ask about it when comparing firms, so the firm acts on that gap. The audit also reveals that AI confuses the firm’s abbreviated name with another business. The firm addresses that information error straight away.
Use the audit to confirm or change how much effort you put into AI visibility. That can range from monitoring to SEO shaped by customer questions or dedicated work on AI visibility. Choose your starting point based on how your customers buy. If you decide to test again, record the date.
In our work, gaps most often come from unclear information on the website, weak connections between related information or technical issues such as indexing and robots.txt rules. Use the audit to narrow down the likely causes, then check them before planning changes.
| Possible cause | What to check |
|---|---|
| Unclear information on the website | Whether the company name, abbreviations, services and location are easy to identify |
| Weak connections between related information | Whether the site clearly connects the business with its services, experts and locations |
| Technical issues | Indexing, robots.txt and whether a firewall blocks crawlers |
| Sources outside your website | Whether answers draw on directories, publications or competitors’ websites |
| Gaps in your content | Whether the website answers the questions on your list |
| Problems with the test | Whether your choice of questions or test conditions distorted the results |
Our AI search optimisation guide explains what to do in each area.
We often see AI assistants ask which business or town the user means, or confuse two companies when an abbreviation is unexplained on the website. A follow-up question of this kind is a strong sign that the company is hard to identify. We also regularly see answers describe businesses using someone else’s website as the source. The information customers receive then depends on what others say about the business.
Record which sources each answer cites and whether the brand’s site is among them. If the assistant regularly cites other sites, check what information it takes from them and whether your own site explains it as clearly. If your site lacks an answer to a customer question, see our guide to creating content AI can cite.
Test again after making changes to see whether your diagnosis holds up.
Record enough detail for someone to act on each finding and measure what changes. The audit should include the scope, the conditions you tested under, a baseline, and a decision, owner and deadline for each gap or action.
Who carries out the work depends on the resources available to your business. See our guide to AI search optimisation in-house or with an agency.
An initial test across several platforms can uncover inaccurate descriptions and give you a starting point for technical checks. If your brand never appears, record zero mentions for that sample and repeat the test on the same platform to see whether the gap persists. If it appears for some topics, repeat the test where it was missing, then assess how much those gaps matter to sales. You can start with our free AI visibility report. To run a fuller audit with our team and agree a plan, talk to us about AI search optimisation.
The most common mistakes start with how you run the test. Too few questions, missing buying questions, inconsistent test conditions and incomplete records can all skew your findings. For mistakes in interpreting the results, see what conclusions to avoid.
Each answer is one observation. Look for patterns across the full sample.
Start with 10–20 customer questions that matter to your services and ask each one several times on different days. If the share of answers mentioning your brand varies widely between the first few rounds, run more tests under the same conditions.
Run the tests signed out or in private browsing mode where possible. If you use a ChatGPT account, choose “Unpersonalized” in Temporary Chat before your first message. The personalised version can use existing memories. OpenAI’s Temporary Chat guidance.
ChatGPT can tell you what it says about your brand in that conversation. To run an audit, ask your customers’ questions across several separate sessions and record the answers so you can compare results over time.
A tool can collect test answers automatically for regular monitoring. You’ll still need to choose the questions that matter to sales and assess the commercial value of any gaps. Base your conclusions on the sampled answers, since actual user conversations are private.
An SEO audit checks whether search engines can access and understand your website. An AI visibility audit checks whether AI systems mention, cite and recommend your brand in answers. The two overlap because Google’s AI Overviews and AI Mode follow the same SEO requirements as the rest of Google Search. Measure the outcomes for each independently.
Set the next test date when you write up the audit. Repeat it regularly and after major website changes. The right schedule and metrics for ongoing reporting are a separate topic.
A gap can point to a missed opportunity. The test measures whether your brand appears in sampled answers. To decide how much the gap matters to sales, look at the value of the service and whether customers ask those questions when choosing a supplier.
Use the findings to decide whether to fix an issue straight away, plan further work or measure again on a set date. AI answers vary between runs, so base that decision on how often your brand appears across a series of answers and how important the gaps are to your business.
Start with a list of the questions customers ask before choosing a product or service. Record the first round in your spreadsheet as a baseline.