How to Run an AI Visibility Audit Step by Step

Marketers working through an ai visibility audit checklist

An AI visibility audit is a one-off review of how AI assistants and AI search features represent your brand today. You run a fixed set of buyer prompts across the main engines, record whether you are mentioned, cited, how high you appear and in what tone, compare that with competitors, then check whether AI crawlers can actually reach and read your site. The output is a prioritized fix list, not a single score.

This guide walks through the audit in eight short steps, each linked to a deeper article, and ends with a scorecard that turns raw findings into a ranked to-do list.

What an AI visibility audit covers and what it leaves out

An audit is a snapshot. It answers three questions: where do we stand in AI answers, why, and what should we fix first. Ongoing tracking answers a different question, whether things are moving week to week, and it needs a larger and more stable setup.

The audit has two halves. The outside-in half looks at the answers themselves: mentions, citations, position, sentiment and competitor share. The inside-out half looks at your site and brand footprint: crawler access, rendering, entity facts and the sources AI engines lean on. Findings from both halves meet in one scorecard, because a weak answer often has a technical cause.

How to run an AI visibility audit in eight steps

Step 1. Set the scope, market and competitors

Write down the products or services in scope, the country and language you sell in, and three to five competitors. Pick competitors from what buyers actually compare you with, and add any brand that keeps appearing in AI answers even if you never thought of it as a rival. Location matters because several engines tailor answers to the user’s country, so an audit for the US says little about Germany.

Step 2. Build a baseline prompt set

Draft 20 to 50 prompts that a buyer would type at different stages: category questions (“best payroll software for small agencies”), comparisons (“X vs Y”), problem questions and branded checks (“is X reliable”). Keep the wording natural and leave your brand out of non-branded prompts. For a full method on sourcing and grouping prompts, see how to build a prompt set for AI visibility tracking. Freeze the list before you start, because changing prompts mid-audit makes the numbers incomparable.

Step 3. Run every prompt across the engines, more than once

Cover the engines your buyers use. As of October 2026 that typically means ChatGPT, Gemini, Claude, Perplexity, Microsoft Copilot, Google AI Overviews and Google AI Mode, with Grok, DeepSeek and Mistral added where relevant. Use a logged-out or clean session so memory and past chats don’t skew results.

Run each prompt at least three times per engine. AI answers are not stable: in a SparkToro and Gumshoe study of 12 prompts run 60 to 100 times each, ChatGPT and Google’s AI returned the same brand list less than 1% of the time, and the authors concluded that visibility percentage across many runs is more reliable than any single rank (Search Engine Journal, January 2026). One run per prompt tells you what happened once, not where you stand.

Step 4. Record mentions, citations, position and sentiment

For each answer, log four fields in a spreadsheet: was the brand named (mention), was one of your URLs linked as a source (citation), where in the answer you appeared relative to other brands (position), and whether the description was positive, neutral or negative (sentiment). Also note which domains were cited. The difference between the first two fields matters more than it looks, and it is unpacked in AI mentions vs citations. Turn the log into rates with the formulas in the AI visibility metrics that matter.

Step 5. Compare against competitors

Use the same answers to count competitor mentions and citations. Two views are useful. Share of mentions shows who the engines recommend most often. A prompt-level gap list shows the prompts where a competitor appears and you don’t, plus the domains cited in those answers. That list is usually the richest source of content and PR actions.

Step 6. Check technical access for AI crawlers

Now switch to the inside-out half. Four checks catch most technical blockers.

  • robots.txt by bot. Training and search bots are separate. OpenAI’s crawler documentation says sites that opt out of OAI-SearchBot will not be shown in ChatGPT search answers, while GPTBot only covers training. Check every host and subdomain, then compare the rules with the bot list in AI crawlers and how to control each bot.
  • Firewall and CDN rules. A bot management setting can return 403s to bots your robots.txt allows. Filter server logs for AI user agents and check the status codes they receive.
  • Rendering. View the raw HTML of your key pages with JavaScript off. Vercel’s analysis of its own network in December 2024 found that none of the major AI crawlers rendered JavaScript, with Gemini (through Google’s infrastructure) and AppleBot the exceptions. Prices, specs or answers that load only client-side may be invisible.
  • Indexability for Google’s AI features. Google states that AI Overviews and AI Mode have no additional requirements beyond being indexed and eligible for a snippet, and that no special AI text files are needed. So a stray noindex or nosnippet is a real blocker, and a missing llms.txt is not. Treat llms.txt as optional housekeeping; what llms.txt does and doesn’t do covers when it is worth adding.

Step 7. Check entity consistency

Engines describe brands using whatever facts they find most often. Compare your name, one-line description, category, founding year, location and key people across your About page, Organization markup, LinkedIn, Wikidata or Wikipedia if you have entries, and the main review or directory sites in your niche. Compare that with how the engines described you in Step 4. Outdated pricing, a discontinued product or the wrong category in AI answers usually traces back to one of those sources.

Step 8. Score and prioritize the findings

List every issue from Steps 4 to 7 and score it with the scorecard below. The goal is a short, ordered list that someone can act on next week.

AI visibility audit scorecard with severity and priority

Score each finding on three scales, then compute a priority number: Priority = Severity × Reach ÷ Effort. Severity runs from 1 (cosmetic) to 4 (blocks visibility). Reach is 1 if the issue touches a few prompts or one engine, 2 for a prompt cluster or several engines, 3 for most prompts or all engines. Effort is 1 for under a day, 2 for under a month, 3 for longer. Fix anything with severity 4 first regardless of score.

Magnifying glass over documents, reviewing brand mentions in ai answers
How to run an ai visibility audit step by step 4
Check What a failure looks like Default severity Typical fix
Search bots allowed in robots.txt OAI-SearchBot, Claude-SearchBot or PerplexityBot disallowed 4 Critical Edit robots.txt per host
Bots not blocked at CDN or firewall 403 or 503 responses to AI user agents in logs 4 Critical Adjust bot management rules
Key pages indexable and snippet-eligible noindex or nosnippet on pages that should be cited 4 Critical Remove the directive
Critical content in raw HTML Prices, specs or answers only appear after JavaScript runs 3 High Server-side or static rendering
Brand facts accurate in answers Wrong category, old pricing or retired product described 3 High Correct the source pages and profiles
Mention rate on non-branded prompts Competitors named, you absent 3 High Content and third-party coverage for the gap prompts
Citation rate when mentioned Named but your pages never linked 2 Medium Create citable pages that answer the prompt directly
Sentiment Recurring negative or hedged descriptions 2 Medium Address the underlying complaint and the sources repeating it
Entity facts consistent across profiles Different descriptions or founding facts on different sites 2 Medium Align profiles and markup with the About page
llms.txt present No file 1 Low Optional; add if cheap

Worked example with illustrative numbers

The figures below are invented to show the arithmetic, not taken from a real brand. A software company audits 30 prompts across 5 engines with 3 runs each, which gives 30 × 5 × 3 = 450 answers. It is mentioned in 81 of them, an 18% mention rate, while its main competitor appears in 171, or 38%. Its pages are cited in 27 answers, 6% of the total.

Three findings come out of it. First, Claude-SearchBot is disallowed by a robots.txt rule copied from an old template: severity 4, reach 2, effort 1, priority 8, and fixed first anyway as a blocker. Second, the pricing page loads plans with JavaScript: severity 3, reach 2, effort 2, priority 3. Third, the brand is missing from 14 comparison prompts where the competitor is cited from two review sites: severity 3, reach 3, effort 3, priority 3. The ties are broken by effort, so the rendering fix goes ahead of the longer outreach work.

Free AI visibility checkers and when a full audit is needed

A free AI visibility checker is a fast way to cover parts of Step 6. Asheard’s free AI bot access checker reads robots.txt rules per AI crawler, and its “Can AI read this page?” check shows what a page looks like without JavaScript. Others run a handful of prompts and return a quick score.

Team sorting audit findings by priority on a whiteboard
How to run an ai visibility audit step by step 5

What a quick AI search visibility checker can’t do is replace the outside-in half. A few prompts run once won’t survive the run-to-run variation described in Step 3, and a single score hides which prompts and engines drive it. When the audit turns into ongoing measurement, Asheard’s AI visibility tracking runs a saved prompt set across ten engines and records mentions, citations, position, sentiment and competitor share of voice over time.

Common mistakes when auditing AI visibility

  • Testing in a logged-in account. Memory and history shape the answers you see, so your results won’t match a buyer’s.
  • Running each prompt once. A single run is an anecdote. Use at least three runs and report rates, not ranks.
  • Only checking branded prompts. Engines usually know your name. The gap sits in category and comparison prompts.
  • Blocking the wrong bot. Disallowing a search bot while meaning to block training costs you citations.
  • Ending with a score instead of a list. Every finding needs an owner, a fix and a date for a re-check.

Frequently asked questions

How often should I run an AI visibility audit?

Run a full audit once to set a baseline, then repeat it after major changes such as a site migration, a rebrand or a new product line, or roughly every six months. Between audits, track the same prompt set on a regular schedule so you notice shifts early. Re-run the technical checks whenever robots.txt, your CDN or your rendering setup changes.

Is there a free AI visibility checker?

Yes. Several free tools check whether AI crawlers can access your site, whether a page reads correctly without JavaScript, or how a few prompts answer today. They are useful for the technical half of an audit. For the answer half you still need a frozen prompt set, several engines and repeated runs, which a quick free check rarely covers.

How many prompts should the audit include?

Between 20 and 50 prompts is a workable range for one brand in one market, covering category, comparison, problem and branded questions. Fewer prompts make each result swing the totals too much. More prompts add cost without much new insight for a one-off audit. Run each prompt at least three times per engine and report rates.

What is the difference between an AI visibility audit and AI visibility tracking?

An audit is a one-off diagnosis: where you stand, why, and what to fix first, including technical checks on crawlers, rendering and entity facts. Tracking repeats the same prompts on a schedule to measure change over time. Most teams audit first, fix the blockers, and then use tracking to see whether the fixes moved mentions and citations.