How to Audit Your AI Search Visibility in Under an Hour

What is an AI search visibility audit?

An AI search visibility audit is a repeatable check of whether AI answer engines mention your brand, cite your pages, and describe your business correctly for relevant buyer questions. It turns a vague question such as whether AI recommends us into a small evidence set that can be reviewed, compared, and repeated.

The audit has two jobs. First, it shows answer presence: whether your brand appears across a defined set of relevant questions. Second, it shows citation performance: whether the answer links to your site or relies on another source when it mentions your category. Those measures are related, but they are not interchangeable. A brand can be named without being cited. A page can be cited without the brand being presented as a recommendation.

This is part of Generative Engine Optimization, often called GEO. The useful unit is not a generic claim of AI visibility. The useful unit is a prompt, an engine, a date, the full answer, the cited domains, and a pass or fail judgment against a rule you set before testing.

Google makes the distinction clear in its documentation for AI features. A page must be indexed and eligible for a Search snippet to be eligible as a supporting link in AI Overviews or AI Mode, but eligibility does not guarantee that Google will show it. That makes direct observation necessary. Search Console remains useful for traffic and technical diagnosis, while the audit records what buyers can actually see in the answer surface.

The scope should fit the time available. An under-an-hour baseline does not prove long-term performance. It gives you a clean starting point, exposes obvious gaps, and establishes the prompt set that future audits should reuse. Keep the first pass small enough to finish without changing questions midway through the work.

1. Set a narrow market, product category, or local service area before you test.
2. Use the same questions in every engine where the question can be run.
3. Capture the answer evidence before interpreting performance.
4. Separate a brand mention from a link or citation to the brand.
5. Mark uncertainty instead of forcing a pass result.

Which prompts should you test first?

Test prompts that mirror the questions a buyer asks before choosing a provider. Start with category, comparison, problem, and use-case questions. Branded queries can confirm basic recognition, but they are weak evidence of competitive visibility because they already name the answer candidate.

A B2B software company can begin with a category prompt, a comparison prompt, a jobs-to-be-done prompt, a pricing or implementation prompt, and a trusted-source prompt. A local business can substitute service-and-location prompts, urgent-problem prompts, provider-comparison prompts, and questions about local proof. Keep the prompt language natural. Do not add your brand name to a non-branded test.

Write the prompt set before opening any engine. That prevents a common audit failure: changing the wording after an answer does not contain the desired brand. A stable prompt set creates a baseline. It also lets a team distinguish a content gap from ordinary variation in a generative answer.

Use five prompts for a one-hour baseline. That yields 20 observations across four surfaces. Save each answer in the same worksheet with the prompt ID, engine, date, location or market setting where visible, mention status, citation status, accuracy status, cited domains, and notes. A screenshot or copied answer URL can add evidence where the surface supports it.

Prompt design should be specific enough to match buying intent without trying to steer the engine. Use terms your market would recognize. Include the location when the buyer would include it. Add a constraint such as best fit for a small team only when that constraint reflects a real purchase decision.

1. Category prompt: ask for the best [category] tools for [audience].
2. Comparison prompt: ask for [competitor] alternatives for [use case].
3. Problem prompt: ask how teams solve [specific problem].
4. Use-case prompt: ask for the best way to achieve [outcome] for [audience].
5. Local prompt: ask for the best [service] in [city or area].

How do you run the 20-point audit in under an hour?

Run the audit in four 15-minute blocks: prompt setup, engine checks, evidence review, and scoring. Work in a spreadsheet or document that preserves the raw answer before you summarize it. The audit is faster when the scoring rules are decided in advance.

Create five prompts and label them P1 through P5. Open ChatGPT, Perplexity, Gemini, and Google Search for AI Overviews. Run each prompt as written. Do not ask follow-up questions, rewrite the question because an answer is weak, or ask an engine to include more citations. Those actions may be useful research, but they make the baseline less comparable.

For each result, first record whether the answer addresses the question. Then record whether your brand is named, whether your owned domain is cited or linked, whether the description is accurate, and which other domains receive citations. If the engine has no visible answer or no AI Overview appears, mark the observation as unavailable rather than failing the brand.

Pew Research Center found that 18% of Google searches in its March 2025 study generated an AI summary. That means a Google audit must distinguish between a query that produced no AI Overview and a query that produced an AI Overview without your brand. The first is not evidence that the brand lost a citation opportunity. The second is.

Google also says AI Overviews and AI Mode may show different responses and links, and that AI Overviews often do not trigger. Treat the engine surface as part of every observation. A page's organic ranking, a Search Console click, and a supporting link in an AI Overview are different evidence types.

Once all 20 checks are complete, review the rows by prompt and by engine. Look for repeated patterns: a brand that is absent from category prompts, a product that is named but never cited, a stale description, or a recurring competitor source. Write the finding only after the raw entries are complete.

1. Minutes 0 to 10: Define five prompts and the pass or fail rules.
2. Minutes 10 to 40: Run the five prompts in four AI answer surfaces.
3. Minutes 40 to 50: Capture citations, mentions, accuracy, and unavailable results.
4. Minutes 50 to 60: Calculate the measures and identify the strongest repeated gap.

What does the printable 20-point AI visibility audit checklist include?

The printable checklist creates one pass or fail observation for every combination of five prompts and four major AI answer surfaces. Print a grid with P1 through P5 as the columns and ChatGPT, Perplexity, Gemini, and Google AI Overviews as the rows. Add the test date, market or location, and the exact prompt wording above the grid.

For ChatGPT, mark each of the five cells as a pass only when the answer visibly names the brand and cites or links an owned domain. Apply the same rule to each Perplexity cell and each Gemini cell. Record a fail when the brand is absent, the owned domain is absent, or the answer materially misstates the business. Record the cited domains in a notes field rather than relying on memory.

For Google AI Overviews, test P1 through P5 only when an overview appears in the search result. Mark a cell as a pass when the overview visibly links an owned domain. Mark it as a fail when the overview appears without an owned-domain supporting link. Mark it as unavailable when no overview appears, then exclude that cell from the Citation Share denominator.

This is deliberately strict. A mention alone can be useful, but it does not establish that the engine cited your content. The completed grid contains 20 observations, each with an explicit rule: 15 checks require a visible brand mention plus an owned-domain citation or link, while the five Google AI Overview checks require an overview and a visible owned-domain supporting link.

A failed cell does not tell you to manufacture pages or chase a single result. It tells you where to investigate. Check whether your site has a clear page that answers the prompt, whether the page is indexable, whether the claim is current, and whether the engine cites another source that has more direct evidence. Google states that there are no additional technical requirements or special markup for appearing in AI Overviews or AI Mode. Start with indexing, eligibility, and clear content rather than a supposed AI-only tag.

The checklist is a baseline instrument, not a promise that any content change will produce citations. Models and answer systems vary. Repeating the identical prompt set on a regular schedule lets you see direction without treating one answer as a final verdict.

How should you measure GEO performance after the audit?

Measure GEO performance with a small set of definitions that remain stable from one audit to the next. Start with answer presence, citation share, citation count per day when you monitor at scale, and share of voice against named competitors. Do not combine them into one score until the underlying observations are reliable.

Answer Presence is the percentage of tested answers that name the brand. Divide brand mentions by eligible answer observations. If 12 of 20 eligible answers name the brand, answer presence is 60%. Keep unavailable AI Overview results out of the denominator because no overview was available to evaluate.

Citation Share is the percentage of relevant AI answers in the defined prompt set that cite your brand or owned domain. Divide owned-domain citation observations by eligible answer observations. If 6 of 20 eligible answers visibly cite an owned domain, Citation Share is 30%. Use the same definition for every competitor when you compare share of voice.

Accuracy is a separate quality check. Mark an answer inaccurate when it makes a material mistake about the product, service, audience, location, or offer. An inaccurate mention should not be counted as a successful recommendation. Capture the exact wording and the correction needed, then investigate the pages and third-party sources that may support the wrong description.

A useful audit report has a compact result table, the prompt set, the observation date, the engine list, and the source evidence. It should not claim causal certainty from a single test. The point is to identify the next content or technical question with enough precision that someone can verify it.

Track results over time in a fixed market and language. If you change the prompts, engines, target region, or scoring rule, record the change beside the new audit. Otherwise a movement in Citation Share can reflect a measurement change rather than a visibility change.

1. Answer Presence: Divide brand mentions by eligible AI answers.
2. Citation Share: Divide owned-domain citations by eligible AI answers.
3. Citation Count per day: Record the total tracked citations observed each day.
4. Share of Voice: Compare your citation or presence result with named competitors.
5. Accuracy rate: Divide accurate brand representations by brand mentions.

What should you do when the audit finds a visibility gap?

Treat the gap as a research task. Start with the prompt and the engine where the failure repeats. Read the answer and its visible citations. Then compare the buyer question with the pages on your site that should answer it. A gap may reflect missing coverage, an unclear page, stale information, a technical eligibility issue, or an answer system that chose a different source.

Check basic site conditions before publishing anything new. Google says pages shown as supporting links in AI Overviews or AI Mode must be indexed and eligible to show a Search snippet. Check crawl access, index status, canonical handling, internal links, and whether the relevant information is available as readable text on the page.

If the content is thin, build a page that directly answers the buyer question with named scope, current facts, and sources. If the page exists but the description is wrong, correct the page and the authoritative sources you control. If a competitor earns the citations, inspect the cited source for its evidence and coverage rather than copying its wording.

Repeat the same audit after meaningful work has had time to be crawled and after a consistent reporting interval. Do not ask for guarantees. AI answers are variable, and visibility is earned through quality, coverage, freshness, and technical eligibility. The audit gives your team a disciplined way to see whether those inputs are producing citations.

1. Prioritize repeated failures over a single missing mention.
2. Fix technical eligibility before assuming the problem is content.
3. Create or improve the page that directly answers the tested question.
4. Keep claims current and supported by sources.
5. Re-run the unchanged prompt set and compare the evidence.

Sources

Source: Pew Research Center

18% of Google searches in Pew Research Center's March 2025 study generated an AI summary, and 88% of those summaries cited three or more sources. Pew Research Center, 2025-07-22

Source: Google Search Central

Google says there are no additional requirements or special optimizations to appear in AI Overviews or AI Mode, and eligible supporting links must be indexed and eligible to show a Search snippet. Google Search Central, 2025-12-10