AI Visibility Audit

What is an AI visibility audit?

An AI visibility audit is a structured review of how AI search engines currently mention, cite and describe a brand, and why. It tests crawler access, runs real buyer prompts on 6 engines, records a citation baseline and ranks every fix by impact and effort.

AI engines now answer many of the questions buyers ask before they choose a supplier. Google AI Overviews and AI Mode, ChatGPT search, Perplexity, Gemini and Microsoft Copilot each write an answer, name a few brands and link a few sources. The audit finds out which brands and sources those answers use for your market today, and traces the reasons your site is or is not among them.

The “why” is the part that makes the audit useful. A low citation count can come from a blocked crawler, a page missing from the index, a question no page answers, an answer buried in a long introduction, or a brand described differently on every profile. Each cause has a different fix, and fixing the wrong one wastes months. The six steps below test each cause in order, from the pass/fail checks to the competitive ones, and every step ends in a written output.

The audit is also the reference point for everything after it. The baseline from step 3 is the number that later work is measured against, so it is recorded with the exact prompts, engines, dates and conditions used, and it can be repeated at any time.

Bikash Roy runs each audit personally — there is no sales team and no automated sequence. Bikash has worked in SEO and web development since 2019 and founded GrowNion in June 2025.

1. Check AI crawler access and eligibility

   Step 1 of 6

   Crawler access is tested for 16 AI user agents across robots.txt, firewall and CDN rules, and every key page is checked for index and snippet eligibility, including the Search Console setting that admits a site to Google's generative AI features.

   A user agent is the name a crawler sends when it requests a page. The audit tests all 16 that the site's crawler checker covers: GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-SearchBot, Claude-User, PerplexityBot, Perplexity-User, Google-Extended, Googlebot, Bingbot, Applebot-Extended, CCBot, Meta-ExternalAgent, Amazonbot and Amzn-SearchBot. Search crawlers and training crawlers are reported separately, because they are separate decisions: OpenAI states that sites opted out of OAI-SearchBot are not shown in ChatGPT search answers, while GPTBot concerns model training. Google-Extended controls Gemini training and grounding in Gemini Apps and Vertex AI, and Google states it has no effect on inclusion in Google Search — so blocking it does not remove a site from AI Overviews.

   Eligibility is checked page by page: indexed in Google and Bing, eligible to show with a snippet, and the site included in generative AI features through the Search Console setting. Pages whose main content is missing from the raw HTML are flagged, because crawlers differ in how they handle JavaScript.

   Example access matrix rows (illustration, not a client result)

   User agent	Path	Result
   OAI-SearchBot	/	Allowed
   GPTBot	/	Blocked by robots.txt rule Disallow: /
   PerplexityBot	/private/	Blocked by robots.txt rule Disallow: /private/
   Google-Extended	/	Allowed

   Output: access matrix — crawler × path × allowed/blocked, plus an eligibility list for every key page

2. Map the prompts your buyers ask

   Step 2 of 6

   Buyer prompts are collected, grouped by intent and matched to the page that answers each one, so the audit measures the questions that lead to revenue rather than vanity queries that contain the brand name.

   A prompt is the full question a person types into an AI engine. The set starts from the questions you send with the request and is completed from sales and support conversations, site search and the questions visible in your market. Each prompt is placed in one of four intent groups — definition, comparison, recommendation and how-to — and assigned to the page on your site that should answer it. Prompts with no matching page, and pages whose answer is hidden or incomplete, are marked as gaps.

   The set is sized to the business: a focused service company might need 20 prompts, a software product with several use cases closer to 50. Recommendation prompts — "best provider of X for Y" — get extra weight, because those answers name brands directly and shape shortlists. Once the audit starts, the wording of each prompt is frozen, so the same set can be re-run after fixes and compared like for like.

   Output: prompt-to-page map with gaps

3. Record current mentions and citations

   Step 3 of 6

   Every mapped prompt runs on 6 engines, 3 times each, under fixed conditions, and each answer is logged: brand named, brand linked, cited URL, competitors named and position. Search Console, Bing Webmaster Tools and GA4 data are added alongside.

   AI answers vary from one run to the next — a 2026 survey of 45 GEO studies describes substantial run-to-run variability — so a single check is not a measurement. Runs use a logged-out session with the country and language fixed. First-party data sits beside the tracked answers: the Search Console Generative AI performance report for impressions in AI Overviews and AI Mode, Bing Webmaster Tools AI Performance for Copilot citations and grounding queries, and GA4 referrals, including ChatGPT visits tagged utm_source=chatgpt.com.

   The competitor columns matter as much as your own. For every prompt the log shows which brands each engine names and which pages it links, so the baseline answers two questions at once: how visible you are, and who takes the answers you are missing. The cited competitor pages also show, in concrete terms, what a passage that wins looks like for your market. Where server logs are supplied, they show how often each AI crawler actually requests your pages, which confirms or contradicts what robots.txt allows.

   Output: baseline mention rate, citation rate and share of model per engine, with every answer logged

4. Audit passage-level answerability

   Step 4 of 6

   Each mapped page is read the way an AI engine reads it: passage by passage, checking whether the answer to its prompt can be lifted out and quoted on its own, with facts stated precisely.

   Three checks apply to every question heading: the answer appears in the first 40–60 words under it; numbers replace vague quantities; and claims point to sources a reader can verify. Where a passage fails, the audit supplies a rewrite.

   Before (example, a commercial cleaning company): We've helped lots of businesses over the years, and pricing really depends on the job, so get in touch to find out more.

   After: Office cleaning is priced per square metre, and the pricing page lists the rate for daily, weekly and monthly contracts. Every contract includes supplies and a named supervisor.

   The rewrite states the fact the prompt asks for in its first sentence, uses exact figures supplied by the business, and needs no surrounding paragraph to make sense.

   Output: rewrite list with before/after passages

5. Review entity and brand signals

   Step 5 of 6

   The brand's name, category, location and offer are compared across every source an AI engine reads, and each conflict is listed, because inconsistent descriptions weaken how engines identify and recommend a brand.

   An entity is a thing an engine can identify, such as a business or a product. Five sources are reviewed: structured data on the site, the About page, Wikidata presence where the business qualifies, review and directory profiles, and third-party mentions in publications and forums. The competitors named in step 3 are reviewed the same way, which shows where their descriptions are clearer or more widely repeated than yours.

   Typical conflicts are small but repeated: an old business name on a directory, a service area that differs between the site and the business profile, a founding year that changes between pages, or structured data that names a different category from the About page. Each is listed with the source, the current wording and the wording to use instead.

   Output: entity consistency report

6. Deliver the prioritised fix plan

   Step 6 of 6

   Every finding from steps 1 to 5 becomes a fix scored by impact and effort, placed in one of four groups, and walked through on a 60-minute call so your team knows what to do first and why.

   • Do first — high impact, low effort: for example, unblocking OAI-SearchBot.
   • Plan — high impact, high effort: for example, rewriting key service pages answer-first.
   • Batch — low impact, low effort: for example, correcting a business name on minor directory profiles.
   • Later — low impact, high effort: for example, new content for prompts buyers rarely ask.

   Each fix in the plan names the step that found it, the pages or profiles affected, the engines it applies to and the result to check after it ships. The plan also sets the re-test: the same prompts, engines and conditions as the baseline, so the change in mention rate and citation rate can be read directly.

   Output: fix plan + 60-minute walkthrough call

How does an AI visibility audit differ from an SEO audit?

An SEO audit checks rankings and technical health for a results page; an AI visibility audit checks what AI answers say about a brand and which sources they cite. It adds AI crawler tests, per-prompt citation baselines and passage-level extraction review.

The two overlap at the foundation — both check crawling and indexing — and the AI visibility audit reuses those checks rather than repeating a full SEO audit. It diverges at the unit being measured: an SEO audit follows a page’s position for a keyword, while this audit follows whether a passage is used, and credited, inside an answer to a prompt. A site with a healthy SEO audit can still fail here, most often because a search crawler for one engine is blocked or because answers sit too deep in the page to be lifted.

The audits also report to different people. An SEO audit usually lands with a developer or content team as a list of issues. The AI visibility audit is read by the person responsible for how the brand is presented, because its baseline shows, prompt by prompt, which competitors AI engines recommend instead, and its fix plan mixes technical, content and reputation work in one order.

No sales team · No automated sequences · Your data stays private. Scope and fee are confirmed after the first reply.

Which parts of the audit can you run yourself?

Two parts are self-serve: the crawler access check and the 5-metric baseline. The free checker tests robots.txt against 16 AI user agents in seconds, and the tracking sheet logs mentions and citations for a fixed prompt set.

Not ready to book? Start with the free AI crawler check, which covers the robots.txt part of step 1 for any path on your site. Then use the formulas, calculator and sheet behind the 5 AI visibility metrics to run step 3 by hand on a small prompt set.

What the self-serve route leaves out is the rest of the diagnosis: firewall and CDN rules, index and snippet eligibility, the passage review, the entity comparison and a fix plan that ranks everything together. If your own baseline shows engines citing competitors instead of you, those are the steps that explain why. Both tools apply the same method used throughout generative engine optimization (GEO) work on this site: measure first, change one thing, re-test.