AI Visibility Audit: A Step-by-Step Template

An AI visibility audit is a structured check of whether AI assistants like ChatGPT, Gemini, Perplexity and Google AI Overviews mention your brand when buyers ask about your category, where you appear, and which sources the answers rely on. It turns scattered spot checks into a baseline you can compare against and a ranked list of fixes.

Most teams start by typing a few questions into ChatGPT and getting a vague feeling. An audit replaces that feeling with a fixed prompt set, repeated runs, a scoring sheet and a gap list. This guide walks through the six steps, gives you the template columns, and explains how agencies can present the results to clients.

Key takeaways

• An AI visibility audit measures mentions, position, share of voice and cited sources across a fixed set of buyer prompts and assistants.
• The prompt set is the most important input: mostly unbranded, written in buyer language and tagged by intent.
• AI answers vary between runs and accounts, so run each prompt more than once and read results as samples.
• Cited sources and the searches assistants run are where most actionable fixes come from.
• Technical access for AI crawlers is a quick check that can explain a total absence from answers.
• An audit is a baseline, not a one-off: re-run the same prompts on a schedule to see what changed.

What an AI visibility audit answers (and what it can’t)

A good audit answers four questions. Are we mentioned for the questions buyers ask? Where do we appear relative to competitors? Which sources do assistants cite when they answer? What should we fix first?

It can’t tell you how many people asked those questions, how much traffic or revenue the answers drove, or exactly why an assistant chose one brand over another. Assistant vendors don’t publish ranking factors. Treat the audit as an observation of answers, and pair it with analytics if you need traffic numbers.

Step 1: Build the prompt set

Start with 15 to 30 prompts across the buying journey. Using a B2B CRM as the example:

• Category discovery: “What is the best CRM for a 20-person startup?”
• Use case and segment: “CRM with good email automation for a small B2B agency”
• Comparison and alternatives: “HubSpot vs Pipedrive for a small sales team”, “Alternatives to Salesforce for startups”
• Problem-led: “How do I stop leads falling through the cracks?”
• Branded checks: a few prompts like “Is [your brand] good for startups?” to see how you’re described

Pull wording from sales calls, support tickets and Search Console, not a brainstorm. Tag each prompt with intent, segment and priority, and freeze the wording. Our guide on choosing prompts to track covers this step in depth.

Step 2: Run prompts across assistants

Pick the assistants your buyers use. For most B2B audits that means ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews, plus DeepSeek or Grok if they matter in your market.

You can do this manually or with a tool. Manually, open each assistant, paste the prompt, and copy the full answer and its sources into your sheet. It works for a small set but gets slow: 20 prompts across 5 assistants, run three times, is 300 answers.

Whichever route you take, note the conditions. Answers from a logged-in account can be shaped by memory, location and settings. A logged-out session or a fresh account can give a different answer. Neither is “neutral”. Record which accounts you used so the next audit uses the same setup.

Step 3: Score visibility, position and share of voice

For each prompt and assistant, log three numbers:

1. Visibility rate. The share of runs where your brand is mentioned. Mentioned in 2 of 3 runs is 67%.
2. Average position. Where you appear when the answer is a list. First of five is very different from fifth.
3. Share of voice. Your mentions divided by all brand mentions across the prompt set. See how to measure AI share of voice for the calculation.

Then roll up by assistant and by intent. Patterns usually show up here: strong on branded prompts but absent from discovery prompts, or visible in Perplexity but not in ChatGPT. Report branded and unbranded prompts separately so branded wins don’t inflate the score.

Step 4: Map cited sources and fan-out searches into a gap list

This is where most fixes come from. For every answer, list the URLs the assistant cited. Group them by domain and count how often each appears.

You’ll typically find three kinds of sources:

• Your own pages. Which ones get cited, and for which prompts?
• Competitor pages. Comparison and “alternatives” pages often get cited for comparison prompts.
• Third-party pages. Review sites, listicles, forums and media. If a roundup cited for your top prompt doesn’t include you, that’s a gap.

Where an assistant shows the searches it ran, add them too. ChatGPT, for example, may turn one prompt into several web searches. These fan-out queries tell you what pages you need to rank for to be in the pool of sources at all. Google says AI Overviews and AI Mode may also use query fan-out, in its documentation on AI features and your website.

Step 5: Check technical access

If you’re absent everywhere, rule out access problems before anything else.

• Google AI Overviews. Google states that a page must be indexed and eligible to show with a snippet to appear as a supporting link, with no additional technical requirements.
• ChatGPT search. OpenAI’s crawler documentation says sites that opt out of OAI-SearchBot won’t be shown in ChatGPT search answers. Blocking GPTBot is a separate, training-related choice.
• Other assistants. Check robots.txt rules for each vendor’s crawlers, and whether a CDN or bot-protection layer blocks them. Our guide to AI crawlers and robots.txt lists the user agents.

Bing Webmaster Tools also has an AI Performance report, in public preview, showing citations of your pages in Microsoft Copilot and Bing AI answers. It’s a useful extra column for the audit.

Step 6: Prioritize fixes and set a re-audit cadence

Rank each gap by two factors: how much the prompt matters to revenue, and how much effort the fix takes. Typical fixes include getting listed on a frequently cited review site, publishing an honest comparison page, adding clear facts (who it’s for, price, integrations) to product pages, or unblocking a crawler.

Then set a cadence. A monthly re-run of the same prompts, with the same setup, shows whether fixes moved anything. Change one thing at a time where you can, and judge by trends across runs rather than one answer.

Audit template

Use one row per prompt, per assistant, per run:

For agencies: presenting the audit to clients

Lead with three numbers: visibility rate on unbranded prompts, share of voice against named competitors, and the top cited domains. Then show the gap list, ranked, with owners and dates. Be explicit about variance and account setup so clients don’t read one answer as a verdict, and promise a re-audit date rather than a ranking.

How to run steps 2 to 4 with Citegram

Steps 2 to 4 are the slow part, and they’re what Citegram automates:

1. Add your prompt set. The free plan covers 1 prompt; Pro covers unlimited prompts.
2. Run them in the real apps. The Chrome extension asks each prompt in ChatGPT, Gemini, Perplexity, Claude, DeepSeek, Grok and Google AI Overviews, in background tabs, using the accounts signed in to your browser. Runs count toward your own plan usage.
3. Read mentions, positions and sources. For each answer it records whether your brand was mentioned, its position and the cited links. For Google it records whether an AI Overview appeared.
4. Pull the fan-out searches. For ChatGPT it records the web searches the assistant ran.
5. Use the reports. Share of voice per brand and per assistant and the most cited domains feed your gap list, with trends over repeated runs. First results arrive in about 10 minutes.

Step 5 (technical access) and step 6 (prioritization) stay with you.

FAQ

How often should you run an AI visibility audit?

Run a full audit once to set a baseline, then re-run the same prompts monthly or after major changes. Answers shift as assistants update and the web changes, so a single audit ages quickly. Keep the setup identical between runs.

How many prompts do you need for a reliable audit?

For many brands, 15 to 30 well-chosen prompts are enough to see patterns, as long as each is run more than once per assistant. Coverage of intents matters more than raw volume. Larger companies with several products or regions may need more.

Can I do an AI visibility audit for free?

Yes. You can run prompts by hand in each assistant and log results in a spreadsheet using the template above. Tools save time on repeated runs; Citegram’s free plan tracks 1 prompt across 7 assistants. Our comparison of AI visibility tools covers other options.

Why do AI answers change between runs?

Assistants generate answers rather than retrieving a fixed list, and those that search the web may find different sources each time. Account memory, location and model updates also play a role. That’s why audits use repeated runs and trends, not single answers.

Run your first audit

Start with the prompt closest to revenue and see where you stand across assistants today. Citegram’s free plan tracks 1 prompt across all assistants, forever, and Pro adds unlimited prompts for $10/month. See pricing.