“AI visibility audit” can mean anything from a 40-page report to three screenshots of ChatGPT. Here’s what ours actually covers, in the order we run it, and how to do a lighter version yourself in an afternoon.
The order matters. There’s no point polishing content an AI crawler can’t reach, and no point measuring before you know what “normal” looks like.
When an audit is worth doing
| Trigger | Why it matters |
|---|---|
| AI tools describe your business wrong | Something in your sources is wrong or inconsistent, and the audit finds which one |
| Competitors appear in AI answers and you don’t | Usually access, readability or a lack of quotable specifics |
| A redesign, migration or new CDN | These quietly break crawler access and rendering more often than anything else |
| Classic SEO is fine but AI referrals are flat | Ranking and being cited are related, not the same |
| You’re about to hire an agency | A baseline lets you judge their work later |
Part 1: Can AI engines reach your site?
Everything else depends on this. We check:
- robots.txt. Are search crawlers like OAI-SearchBot, Claude-SearchBot and PerplexityBot allowed? Many sites block them by accident while trying to block training bots. The difference is explained in our robots.txt guide for AI crawlers.
- Firewall and CDN. Cloudflare and similar services can block AI crawlers regardless of robots.txt. We check the actual settings, not just the file.
- Server logs. Which bots really visit, how often, and whether they get errors.
- Status codes. Key pages return 200, aren’t noindexed, and aren’t behind a login.
Part 2: Can they read what they reach?
- Rendering. Is the important text in the server-rendered HTML, or only filled in by JavaScript? We view source on your top pages and search for the key sentences.
- Prices and key facts as text. Not inside images, PDFs or interactive widgets.
- Structure. Clear headings, answers near the top of sections, and FAQ blocks that are visible on the page.
- Schema. Organization or LocalBusiness, plus Product, Offer or FAQPage where they fit, all matching the visible text.
Part 3: Do your facts agree everywhere?
AI engines cross-check. If your website, Google Business Profile, Bing Places, review profiles and directories give different addresses, phone numbers, descriptions or prices, the engine has to pick one, and it might not be yours.
We compile every listing we can find into one sheet with name, address, phone, description and price, and flag every mismatch. This is usually where the most surprising problems hide. The repair process is covered in fixing an AI that describes your business wrong.
Part 4: What do the engines say about you now?
We build a prompt panel: 15 to 30 questions real customers ask, split into three groups.
| Group | Example | What it shows |
|---|---|---|
| Brand | “What does [brand] do and how much does it cost?” | Whether the facts about you are right |
| Category | “Best [service] for small businesses in [city]” | Whether you make the shortlist at all |
| Comparison | “[Brand] vs [competitor]” | How you’re positioned when buyers compare |
Good prompts sound like real customers, not like keywords. A few templates to adapt:
- Local service: “Who’s a reliable [service] in [city] who can come this week, and roughly what does it cost?”
- SaaS: “What’s a good [category] tool for a team of 10, and how does [your product] compare with [competitor]?”
- Ecommerce: “Where can I buy [product type] with free returns, and which brand is best for [use]?”
- Reputation: “Is [brand] legit? What do customers say about them?”
We run every prompt in ChatGPT, Perplexity, Gemini and Google’s AI Mode, in fresh sessions, and log three things each time: whether you appear, what’s said, and which sources are cited. The cited sources are the most useful output of the whole audit. They tell you exactly which pages the engines trust for your topic.
Part 5: Is your content worth quoting?
For the pages that should win your category prompts, we compare them against the pages that actually get cited. Usually the cited pages are more specific: real numbers, named examples, clear direct answers, current dates. We list the gaps page by page. Our guide to writing FAQs that AI engines cite covers the formatting side.
Part 6: Can you measure it from now on?
An audit without a baseline is a one-off opinion. We set up:
- Search Console’s Generative AI performance report, with notes on known data bugs
- GA4 channels for AI referrers, including ChatGPT’s utm_source=chatgpt.com
- ChatGPT Business Hub, if you’re eligible, for first-party mention and click data (see our Business Hub guide)
- The prompt panel, saved so it can be rerun monthly with the same questions
How we score each part
Each of the six parts gets a simple traffic-light rating, so the summary fits on one page:
| Rating | Means | Example |
|---|---|---|
| Red | Blocking. AI engines can’t use you properly until it’s fixed | OAI-SearchBot blocked at the firewall; prices only in images |
| Amber | Working, but costing you citations or accuracy | Old address on two directories; FAQs hidden behind JavaScript |
| Green | No action needed now | Clean robots.txt, matching schema, consistent listings |
Every finding is written the same way, so it can go straight into someone’s to-do list:
Finding: Perplexity and ChatGPT both quote an old phone number.
Evidence: both cite a directory listing from 2023 (screenshot attached).
Fix: update the directory listing; request removal of the duplicate.
Owner: office manager. Priority: amber.
One prioritized list, not a 40-page PDF. Each item has the problem, the evidence (a screenshot, log line or prompt result), the fix, and who should do it. Access issues go first, then consistency, then content. Most sites have five to ten items that matter and a long tail that doesn’t.
The afternoon version you can do yourself
- Run your homepage through the free AI visibility checker.
- Open yoursite.com/robots.txt and check you’re not blocking OAI-SearchBot, Claude-SearchBot or PerplexityBot.
- View source on your main service page and search for your price and your main claim.
- Ask ChatGPT, Perplexity and Gemini “What does [your business] do, where is it, and how much does it cost?” Write down anything wrong and click every cited source.
- Search your business name and check your top five listings match your website exactly.
- Write down this month’s numbers from Search Console’s Generative AI report as your baseline.
For a small business, that catches most of the problems that matter. The rest is depth: logs, the full prompt panel, third-party outreach, and content gaps across many pages.
Free tools that cover most of it
| Part | Free tool |
|---|---|
| Access | Your robots.txt, server logs from your host, Cloudflare’s bot settings if you use it |
| Readability | View Source, Google’s Rich Results Test, the Schema.org validator |
| Consistency | A spreadsheet and a search for your business name and phone number |
| Prompt panel | ChatGPT, Perplexity, Gemini and Google AI Mode in private or logged-out sessions |
| Measurement | Search Console, GA4, ChatGPT Business Hub (if eligible), and server logs (see our guide to identifying Google traffic in analytics and logs) |
Paid tracking tools can automate the prompt panel and add competitor comparisons. They’re useful at scale, but not required to start.
After the audit: a 90-day plan
- Weeks 1-2: fix every red item. Access first, then readability.
- Weeks 3-6: clean up listings and inconsistent facts. Some directories take weeks to update, so start early.
- Weeks 6-12: improve or create the pages that should win your category prompts.
- Every month: rerun the prompt panel and record the numbers from Part 6.
- Day 90: compare against the baseline. Live search answers usually reflect fixes within weeks; answers from a model’s training memory can take until the next model update.
Google’s own guidance on AI features and your website is worth reading alongside this. It confirms there’s no special trick: the same fundamentals that make a page useful and accessible are what get it used.
FAQ
What is an AI visibility audit?
It’s a check of whether AI engines like ChatGPT, Perplexity, Gemini and Google’s AI Mode can reach your site, read it, trust your facts, and cite you. It ends with a prioritized list of fixes and a baseline for measuring progress.
What does an AI visibility audit include?
Crawler access, page readability, schema, consistency of your business details across listings, a prompt panel across major AI engines, content gaps against cited competitors, and a measurement setup.
How long does an AI visibility audit take?
A basic self-check takes an afternoon. A full audit with server logs, a 15 to 30 prompt panel across four engines and a listings review takes longer, depending on how many pages and listings you have.
Can I do an AI visibility audit myself?
Yes, a lighter version. Check robots.txt, view source on key pages, ask the main AI engines about your business, compare your listings, and record a baseline from Search Console’s Generative AI report.
Which AI engines should an audit cover?
At minimum ChatGPT, Perplexity, Gemini and Google’s AI Mode, since they draw on different sources. Check in fresh sessions so personalization doesn’t skew results.
What is a prompt panel?
A fixed set of real customer questions you run across AI engines on a schedule, logging whether you appear, what’s said and which sources are cited. Running the same panel monthly shows real change.
How often should I repeat an AI visibility audit?
Rerun the prompt panel and check your numbers monthly. A full audit makes sense every six to twelve months, or after a big site change like a redesign or migration.
Does schema guarantee AI visibility?
No. Schema helps engines parse your pages, but access, consistent facts and genuinely useful content matter more. It’s one item on the audit, not the whole audit.
Disclosure: this is the audit structure we use at AxoVox, shared in full so you can run it yourself or check what an agency is selling you. See our full disclaimer.