An AI search visibility audit shows where your brand appears, gets recommended, or gets ignored across ChatGPT, Perplexity, Gemini, Claude, and Google AI Overviews. This is the process we run at VisibleIQ, compressed so you can run it yourself.
- Test ~600 queries built from real buyer language: 100 AI prompts plus ~500 Google keywords, because Google's index feeds the engines. Roughly 85/15 unbranded to branded.
- Classify every AI appearance into 4 tiers: Recommended, Cited, Mentioned, Absent.
- Read branded and unbranded results separately. Competitors inside your branded searches is the alarm bell.
- Check the evidence layer engines cite: reviews, listicles, comparisons. 79% of product-query citations point to third-party sites.
- End with a revenue model and a 90-day plan, never a screenshot dump.
These are the fundamentals of the audit we run before an engagement: what the AI engines say about a company’s category, who they recommend, who they skip, and why. Our full version goes deeper on volume, scoring, and revenue modeling, but the bones below are the same, and a marketing lead can run this lean pass in an afternoon without buying anything.
What an AI search visibility audit measures
An AI search visibility audit shows where your brand appears, gets recommended, or gets misdescribed across ChatGPT, Perplexity, Gemini, Claude, and Google AI Overviews when buyers ask about your category. It measures 6 things: mentions, recommendations, citations, answer accuracy, sentiment, and share of voice against named competitors.
Notice what’s missing from that list. Traffic. Rankings sit in the inputs, and revenue sits in the outputs, but the audit itself measures influence: what the engines say about you when a buyer with budget asks a buying question. A brand can be mentioned warmly and never cited, cited once and recommended nowhere, or described with pricing that’s 2 years out of date. Each of those is a different problem with a different fix, which is why lumping them into one “visibility score” hides more than it shows.
Why your Google rankings don’t tell you what AI says
Unlike a traditional SEO audit, which scores how you rank for keywords, an AI search visibility audit scores how engines talk about you. The 2 can disagree completely. A site can hold page 1 positions on Google and be absent from every AI answer in its category, because the engines build answers from sources the rank tracker never looks at.
Our own citation study showed why. On Perplexity, Gemini, and Claude, 79.0% of product-query citations point to third-party websites: the reviews, listicles, and comparisons written about you. ChatGPT flipped the pattern and now pulls 74.6% of its product-query citations from vendors’ own sites, going directly to pricing and product pages with search operators. So there are 2 layers to audit, and they fail independently. Your site can be in great shape while the third-party layer says nothing about you, and the third-party layer can glow while your own pricing page is unreadable to the crawlers. It’s just how the game works now, and the audit has to cover both.
A 30-minute call where we map what AI says about you and build a tailored roadmap.
The audit, in 6 steps
The full version of this audit runs about 500 Google keywords and 100 AI prompts through every major engine, then classifies every appearance. The lean version follows the same 6 steps with 30 to 50 prompts, a spreadsheet, and an afternoon. The steps below follow our internal process in the same order, trimmed to what you can do yourself.
Step 1: Build the query sets: 100 AI prompts and ~500 keywords
The full audit tests around 600 queries: an AI prompt set of 100 and a Google keyword set of about 500. The prompts are the star, they’re what buyers type into the engines. The keyword set is in an AI audit on purpose: the engines retrieve from Google’s index, and Google’s own AI Overviews trigger on keywords, so your Google footprint sits upstream of your AI answers. Audit the prompts without the keywords and you can see that you’re absent without seeing why. Keep both sets roughly 85% unbranded and 15% branded, and weight the prompts toward the decision stage: the “best X for Y”, “X vs Y”, and “is X worth it” questions where vendors get picked. A buyer asking “what is desktop virtualization” is reading. A buyer asking “best VDI provider for a regulated company” is choosing.
We tend to pull the phrasing straight from sales calls and transcripts, because buyers’ words beat keyword-tool words: the category and problem in the buyer’s language, the constraints they name, the integrations they care about, the competitors they compare you against. The 50-prompt set behind our case study came from that method: transcripts first, keyword tools second.
Step 2: Run every query on every engine
Run each prompt on ChatGPT, Perplexity, Gemini, and Claude, and run the keyword set against Google with its AI Overviews. Same queries, same day, one pass per engine, logged in one sheet. Consistency is the whole game here: the audit only means something if the next run tests the same thing the same way.
Step 3: Classify every appearance into 4 tiers
For every answer that comes back, classify how your brand shows up:
| Tier | What it means |
|---|---|
| Recommended | The engine explicitly recommends you |
| Cited | Your URL appears in the citations |
| Mentioned | Your name appears in the text, without a citation |
| Absent | You’re not in the answer at all |
Log who else appears on every prompt, because the competitive picture matters as much as your own. Prompts where nobody dominates are open territory, and those are usually the cheapest wins on the whole map.
Step 4: Check the evidence layer
The engines lean on third-party sources, so audit the sources themselves. Which review platform does AI quote in your category? For the client in our case study, G2 appeared in the brand’s AI mentions more than any other review site, which told us exactly where the review work should go. Check whether the listicles and comparison pages in your category include you, who wrote them, and whether the best comparison of you was written by a competitor. Then check the plumbing on your side: domain strength, and whether AI crawlers can access your key pages at all.
A note on schema and llms.txt, since every audit checklist now includes them: we treat both as nice add-ons. Neither is the game. The engines read well-structured pages fine, and no markup rescues a page that has nothing quotable on it. Spend the hours on the evidence layer instead.
Step 5: Read branded and unbranded separately
Unbranded results answer “can buyers discover you?” Branded results answer “what do buyers find when they check you out?” These are different business problems and they get different fixes, so never blend them into one score. The alarm bell to watch for: competitors appearing inside your own branded searches. When a buyer asks an engine about you and the answer pitches somebody else, that prompt cluster goes to the top of the fix list.
Step 6: Model the revenue gap and build the 90-day plan
For each gap, estimate what it’s worth: monthly searches, times the click rate at the position you’d earn, times your visitor-to-lead rate, times your close rate, times your deal size. Present it as a range with a break-even, as in “you need 2 deals at your ACV for this to pay for itself”, because ranges survive scrutiny and single numbers don’t. Group the gaps by page type (comparisons, use cases, buyer research, education), then map the work: the AI answers are the goal, and SEO and content, review management, and digital PR are the levers that move them.
What the scores actually mean (and what they don’t)
A visibility score only means something if you can say what it measures. Ours blends brand visibility, citation rate, recommendation rate, share of voice, sentiment, and platform coverage, reported against named competitors on a defined prompt set. If a tool hands you a score and can’t tell you the denominator, you have a number, and a number isn’t a measurement.
There’s a version of this critique going around the industry, and we agree with it. AI search measurement is still young, so the discipline has to come from the method: a fixed prompt set, a defined classification, competitors named upfront, and the same test re-run the same way every month. We feel as though citations get treated as the whole scoreboard, and they shouldn’t be. Citations are the receipt. The KPIs that move pipeline are recommendation rate, answer accuracy, and share of voice, because those decide whether a buyer hears your name and hears it right.
What a real result looks like
The clearest proof of what this audit produces is the engagement it kicked off: a $10M B2B SaaS client added $1.1M in ARR from organic and AI search in 8 months. AI referrals were 4.2% of organic sessions and closed 23% of the revenue.
The chain ran exactly as the steps above suggest. Buyer questions became a 50-prompt set, the prompt set exposed the gaps, the gaps became decision-stage pages, a review campaign, and PR placements, and the answers moved. The case study publishes the funnel, the split, and the traffic decline we inherited along the way, because a result you can interrogate beats a result you have to take on faith.
Run it yourself or bring us in
Run it yourself if you have an afternoon for the audit, a spreadsheet, and a writer who can act on what you find. The lean version genuinely works, and running it internally teaches your team how the engines see your category, which is worth the afternoon on its own.
Bring in help when the findings span reviews, PR, and content at once, because at that point it’s a program, and programs need owners. It obviously comes down to budget, right? If you’re a lean team pre-Series A, run it yourself every quarter and fix the top 3 gaps. If pipeline depends on organic and the gaps span 3 service lines, that’s what an engagement is for.
How often to re-run it
Track your core prompts every month and re-run the full audit each quarter. AI answers move when models update, when competitors publish, and when review pages shift, so a set-and-forget audit is stale inside 90 days.
I’d say the tracking loop matters more than the full re-run. The audit is the snapshot; the tracking is the movie. Watching the same 50 prompts month over month is how you catch an engine changing its mind about you while there’s still time to do something about it.
Frequently asked questions
The questions founders and marketing leads ask us most about auditing AI search visibility, answered the way we answer them on calls: what it costs to run, how many prompts you need, how it differs from an SEO audit, and how long the whole thing takes.
How many prompts do you need for an AI visibility audit?
30 to 50 gets a lean audit that finds your biggest gaps, and around 100 covers a category properly. Volume matters less than sourcing: prompts phrased the way real buyers talk, weighted toward decision-stage questions, mostly unbranded.
Is there a free way to run an AI search visibility audit?
Yes. The engines themselves are free to query, and the 6 steps above need nothing more than a spreadsheet. Paid tools add scale and monthly monitoring, which matter once you’re acting on the findings, but that first audit doesn’t cost more than an afternoon.
What’s the difference between an SEO audit and an AI visibility audit?
An SEO audit scores how you rank: keywords, technical health, backlinks. An AI visibility audit scores how engines answer: mentions, recommendations, accuracy, sentiment, share of voice. They overlap on crawlability and authority, and the second one starts where the rank tracker stops.
How long does an AI search visibility audit take?
The lean version takes an afternoon. The full version, with 2 query sets, every engine, competitor classification, scoring, and a 90-day roadmap, takes 1 to 2 weeks including the write-up.
