Practical GEO guide

AI visibility audit: a practical checklist

Measure where your business appears, which sources get cited, what the answers get wrong, and what to fix first.

By Tin Computer · Published September 2, 2026 · Updated September 2, 2026

Free five-step scan · No card · See it run on your site

An AI visibility audit checks whether search and answer engines can access your site, understand your business, cite your pages, and state your facts correctly. Start with a fixed set of buyer questions. Save repeated answers from ChatGPT, Claude, Google AI, and Microsoft Copilot. Then compare answer frequency, cited sources, factual accuracy, and qualified visits.

Do not turn one answer into a ranking claim. AI answers can change by surface, location, date, and phrasing. Record the conditions, repeat the same prompts, and report an appearance rate instead of a permanent position.

A useful audit produces a short evidence file and a prioritized repair list. It does not stop at a single visibility score. The score shows where to look. Saved answers, crawl checks, cited pages, and analytics show what to change.

How it works

  1. 01

    Freeze the questions

    Choose five to ten buyer questions. Include unbranded discovery, comparison, problem, local, and branded fact checks.

  2. 02

    Check access

    Test robots rules, indexability, canonical URLs, sitemaps, internal links, response codes, and rendered page text.

  3. 03

    Save repeated answers

    Run each question more than once on each chosen surface. Save the date, settings, answer, citations, and errors.

  4. 04

    Score what happened

    Count brand appearances, owned-page citations, useful third-party citations, and incorrect or missing business facts.

  5. 05

    Fix the first break

    Repair blocked access and wrong facts first. Then improve weak answers, source evidence, and measurement.

1. Define the audit before you search

Write the question set once. Do not change prompts after seeing an unfavorable answer. Use questions that match real decisions: which provider fits, what a service costs, how two options differ, who serves a location, and whether a product supports a required feature.

  • Discovery: Which companies solve this problem for this buyer?
  • Comparison: What are the best alternatives to a known option?
  • Fit: Which provider works for this size, location, or constraint?
  • Branded fact check: What does the company offer, charge, or support?
  • Trust check: What evidence supports the recommendation?

Use a reproducible baseline

For each run, save the exact prompt, answer surface, model or mode when shown, web-search setting, location, date, answer, and cited URLs.

2. Check whether each engine can reach you

A page cannot become a reliable source when the engine cannot fetch or index it. Review robots.txt, page-level robots directives, response codes, canonical tags, XML sitemaps, crawlable internal links, and the rendered text. Test important pages on mobile and without a logged-in session.

Crawler controls are not interchangeable. OpenAI documents OAI-SearchBot for ChatGPT search. Anthropic documents Claude-SearchBot for search discovery and Claude-User for user-requested retrieval. Google and Bing tie AI visibility to their search indexes. Allow or block each crawler according to your policy, then record the decision.

3. Score answers without inventing a rank

Run every valid prompt the same number of times. Divide answers that name your business by completed answers. Report that appearance rate by question and engine. Keep failed runs separate. A citation is another measure, not proof that the answer recommends you.

  • Appearance rate: answers that name the business divided by completed answers.
  • Owned citation rate: answers citing your domain divided by completed answers.
  • Source coverage: unique relevant domains cited across the saved answers.
  • Fact accuracy: checked claims that match the current primary source.
  • Referral outcome: visits and useful actions from identifiable AI sources.

4. Trace every important fact to a source

Create a fact ledger for your name, category, audience, offer, location, price, availability, and proof. Name the primary page that owns each fact. Then check whether key third-party profiles repeat it accurately. Fix contradictions at their source instead of adding more copy everywhere.

Review the pages that engines cite for your questions, even when they cite competitors. Look for evidence patterns you genuinely lack: first-hand tests, complete comparisons, current pricing, expert detail, local facts, or independent reviews. Build only evidence you can support.

5. Inspect pages for answer readiness

The best repair page answers one complete buyer task. Put the direct answer in visible text. Use descriptive headings, short sections, accurate dates, and source links beside the claims they support. Add structured data only when it describes visible content. No schema type or AI-only file guarantees a citation.

For a recurring audit and repair program built around local discovery, reviews, directories, search, and answer engines, see Tin GEO for small businesses.

6. Use first-party reports where they exist

Google Search Console reports generative AI visibility for Google Search. Bing Webmaster Tools reports citations, cited pages, and grounding queries across supported AI experiences. OpenAI adds utm_source=chatgpt.com to ChatGPT referral URLs. Compare these records with your saved answer tests and useful on-site actions.

Keep observation and cause separate. A visibility change may follow your edit, a demand change, a crawl, or a model update. Record the edit date and measurement window. Do not claim that one change caused movement without a controlled test or stronger evidence.

AI visibility audit checklist

Complete this once for the baseline, then repeat the answer and measurement checks on a fixed schedule.

  • Freeze five to ten buyer questions and their exact wording.
  • Record every engine, surface, setting, location, and test date.
  • Check robots.txt for Googlebot, Bingbot, OAI-SearchBot, Claude-SearchBot, and Claude-User.
  • Verify response codes, indexability, canonicals, sitemaps, internal links, and rendered text.
  • Repeat every prompt equally and keep failed runs separate.
  • Save brand mentions, cited URLs, recommendation context, and factual errors.
  • Trace your core business facts to one current primary page each.
  • Compare cited competitor sources with evidence your site genuinely lacks.
  • Review Google, Bing, referral, and on-site action data where available.
  • Prioritize blocked access, wrong facts, missing evidence, weak pages, then measurement gaps.

What not to call an audit

  • One answer from one engine with no saved prompt or date.
  • A single score with no cited pages, factual checks, or repair list.
  • A ranking claim that ignores answer variation and failed runs.
  • A crawler checklist that never tests the important page itself.
  • A long list of generic content ideas with no buyer question attached.

Primary sources

Frequently asked questions

What is an AI visibility audit?
It is a repeatable review of crawler access, answer appearances, citations, factual accuracy, source coverage, referrals, and useful on-site actions across selected AI search surfaces.
Which AI engines should I audit?
Choose the engines your buyers use. A practical baseline often includes ChatGPT, Claude, Google AI features, and Microsoft Copilot. Record the exact surface because results can differ.
How many prompts should an audit use?
Start with five to ten high-value buyer questions. Cover discovery, comparison, fit, branded facts, and trust. Repeat each question equally before comparing appearance rates.
Is AI visibility the same as an AI ranking?
No. Generated answers can vary, and a citation does not have a stable numbered position. Report answer frequency, citation frequency, accuracy, and outcomes instead.
Does llms.txt improve AI visibility?
No major answer engine in this guide guarantees visibility from an llms.txt file. Prioritize supported crawler controls, indexable pages, useful content, and accurate evidence.
Does structured data guarantee an AI citation?
No. Structured data can describe visible page content and support search eligibility. Google states that valid markup does not guarantee a search appearance.
How do I measure ChatGPT traffic?
OpenAI says ChatGPT referral URLs include utm_source=chatgpt.com. Segment that source in analytics, then measure qualified actions instead of visits alone.
How often should I repeat the audit?
Keep the baseline fixed and rerun it after material site or offer changes. A monthly check is practical for many teams. Use the same prompts and settings each time.