--- title: What an AI visibility audit actually checks (and finds) description: How an AI visibility audit works and what the score really means, plus the findings that come up again and again for professional service firms. author: Danny Ford date: 2026-06-29 canonical: https://scalewithquail.com/blog/ai-visibility-audit-explained/ --- Before you read another word of this, open ChatGPT and ask it the question your best client would have typed before they ever heard your name. Not "tell me about my firm," which the model answers politely because you handed it the answer. Ask the buyer question: "who's a good estate planning attorney in Scottsdale?" or "I need a fee-only financial advisor near Columbus, who should I actually call?" Then read the response slowly and check whether you're in it. Most owners who try this find out they aren't. That slightly queasy moment, where the machine confidently recommends four competitors and never says your name, is the whole reason AI visibility audits exist. I run these every week for law firms and advisory practices, so here's what a real one checks and what firms do with the results. Everything below is enough to DIY a rough version, and that's fine by me. ## The questions matter more than the tools An audit starts with a list of 15 to 25 questions phrased the way buyers phrase them, not the way marketers wish they did. Real people don't type "top-rated legal services provider." They type "do I need a trust or just a will" and "personal injury lawyer near me who actually answers the phone." Some of the list is early research and some is ready-to-hire intent with a city attached, because those two moments produce different answers and you want to know where you stand in both. That list then runs across more than one system, since the answers genuinely differ. ChatGPT, Gemini, Perplexity and Google's AI Overviews all pull from different sources and refresh on different rhythms, and it's completely normal for a firm to look strong in one and be invisible in another. One detail people miss when they DIY this: use fresh sessions. If you've spent three weeks chatting with ChatGPT about your own practice, it remembers, and your results will flatter you. An audit uses clean sessions with no memory, which is far closer to what a stranger sees. ## Who gets named, and why The output isn't a yes or no on your firm. For every question, you record which firms get named and what reason the assistant gives for naming them, because the reasons are the diagnosis. "Known for flat-fee probate" points at positioning, while "over 400 Google reviews" points at your review pipeline. When the assistant cites sources, log those too; they're usually less glamorous than people expect: directory pages, local news mentions, a bar association listing, somebody's well-structured FAQ page. Watch for one pattern in particular. If the same competitor shows up across ten different questions, that isn't luck. Something in the public record keeps feeding them to the models, and working out what is half the value of the whole exercise. ## Then a machine reads your website The second half of the audit is your own site, seen the way a crawler sees it. This part is unglamorous and it decides a lot. 1. Access. Can GPTBot, ClaudeBot, Google-Extended and PerplexityBot actually fetch your pages, or does a security plugin or an old robots.txt line quietly turn them away? More firms block AI crawlers by accident than anyone would guess. 2. Structured data. Schema markup that states plainly what you do and where, and who the practitioners actually are. Machines prefer being told over inferring, and most professional sites tell them nothing. 3. Readable answers. Whether the answers to those buyer questions exist on your pages as plain text, or whether they're trapped in a homepage slider or a PDF nothing can parse. 4. Consistency. Whether your address and phone number read the same on your site as they do on your Google Business Profile and the directories that matter in your field. ## What the score actually means Most audits, ours included, compress all of this into a number: roughly the share of buyer questions where you get named, weighted toward the high-intent ones, plus a site-readiness component. Useful, but only if you read it the right way. The absolute number matters less than two comparisons: you against the competitors who keep getting named, and you now against you in 90 days. A 40 sounds grim until you learn the strongest firm in your market scores a 55, and a 40 that was a 15 last quarter is a genuinely good story, because these systems respond to changes in the public record faster than most people assume. Why bother moving it at all? Because the traffic is small but absurdly warm. Semrush measured AI search visitors converting at about 4.4 times the rate of traditional organic visitors, and in Seer Interactive's client data ChatGPT referrals converted at 15.9 percent against 1.76 percent for Google organic. Ten visitors who arrived from a ChatGPT recommendation are not ten visitors who arrived from a banner ad. ## The findings that keep repeating for professional firms After enough of these reports, the same handful of findings does most of the work. Firms that are perfectly visible on their own name and invisible one step earlier, on the questions that actually start a buyer's search. A single competitor dominating the recommendations on nothing more mysterious than review volume and recency (we looked at how many reviews it actually takes in [how many reviews you actually need](/blog/how-many-google-reviews/)). Websites full of adjectives and empty of facts, with no prices and no towns served, not even a plain statement of who does what. Crawlers blocked by accident. And the ugliest one, an assistant confidently repeating something false, like an old address or a partner who left in 2022. ## What people do with the results The report tends to order itself. Wrong facts get fixed first, because misinformation about a licensed professional isn't cosmetic and it compounds every month it sits there. Site readability and schema come next, usually a week of work rather than a rebuild. After that it's a steady review cadence and, on a longer horizon, content that answers the specific questions you lost, which is the game we walked through in [how firms get recommended by ChatGPT](/blog/get-recommended-by-chatgpt/). You can run every step of this yourself with nothing but the outline above and a spreadsheet, and if you do it honestly you'll learn more about your market in an afternoon than most firms learn in a year. If you'd rather have it done for you, we run exactly this audit free at [scalewithquail.com/ai-visibility](https://scalewithquail.com/ai-visibility/), and we send the full report whether or not you ever hire us. Either way, run the queries. The worst outcome isn't a bad score; it's not knowing that the answer buyers read every day has four names in it and none of them is yours.