Most owners have no idea. In about ten minutes you'll know — and you'll know whether it's worth doing anything about.
$29.90 once — not a subscription. Includes two free re-measures of the same domain within 180 days: measure, fix, verify — three data points for one price. The check below is free.
Takes a minute or two — we read your site while you wait. No account, no card.
We read your site and show you the exact questions we will ask. If we cannot analyse your site, you find out here.
One time. Then we run the analysis — 75 answers recorded, from 150 AI calls: every answer is asked once and then graded by a second pass.
A private web page that stays up for 12 months, plus a PDF that is yours to keep.
To be clear about the order: the free check tells you what we found about your business and which questions we will ask. It does not include the answers — those are the report. We charge before running it because every report costs us real money in AI queries.
Ahrefs, Semrush and HubSpot all have a free AI-visibility checker, and they're genuinely useful. If a free tool answers your question, you've saved $29.90 and we'd rather you did. They all stop in the same place, though — and that gap is the entire thing we sell:
23 questions across every surface, plus repeat sampling
3 assistants asked directly, 0 read off live Google results
Out of 23. The only way to test unprompted recall
Every answer independently graded by a second model pass
Every number here is derived from the constants published on our method page. Check our arithmetic.
We asked, without naming them:
“Best botox clinic Nashville”
This clinic ranks fine on Google and has 415 reviews at five stars. Across the 18 questions we asked without naming them, AI offered the customer 169 different businesses — and named this one once, on a question about chemical peels. On this question, the one they most want to win, it did not come up at all.
Read the whole report they got →Ten stages, about ten minutes, roughly 150 paid API calls. These are the real stage names from our code, not a simplified diagram — you can match each one against the method page.
Sitemap, robots.txt, your llms.txt if you have one, and every page we can reach — the same way a crawler would. Nothing here is guesswork about your business; it is your own text.
From the page text, not your meta description. Business shape, who you serve, which market, whether you sell to consumers or businesses. You get to correct all of it before we run.
Not a template list. 23 questions built for your category, spread across four intent tiers — from someone who has not worked out the category yet, to someone ready to buy. 18 of them never contain your name.
Each question goes to every model surface in its own clean session, so nothing we asked earlier can contaminate a later answer. That is 69 answers. Your 3 starred questions are each asked 3× in total — a further 6 — because a single run of a drifting system is a coin flip, not a measurement.
Built and disclosed, not running yet. Reading the Google results page needs a paid search-data vendor we have not switched on, so today this stage is skipped and your report says so on its own line. Nothing is estimated in its place, and you are not charged differently for it — what you get is the 3 assistant surfaces above.
Structured data errors, pages that cannot be rendered, and snippet controls. That last one matters more than it sounds: Google requires a page to be eligible for a snippet before it can appear in AI features, so a stray nosnippet tag is a hard blocker, not a nitpick.
Google Business Profile category, review volume, and whether your name, address and phone agree everywhere they appear. Skipped entirely if you have no physical location — we do not pad the report with sections that cannot apply to you.
When AI cites a source about your category, it is usually not your website — it is a directory, a review site or a local publication. We find which ones hold that ground in your category and whether you are on them.
This is the part most tools skip. We do not search the text for your brand name — string matching cannot tell the difference between being recommended, being mentioned in passing, being cited as a source while someone else gets recommended, and a model simply echoing a name we put in the question ourselves. So all 75 answers go through a separate graded read against a written rubric, which also extracts every competitor named ahead of you.
Page one is a verdict — one of 4, chosen by a published rule rather than an opinion, and it is allowed to say your money is better kept. Behind it: scores, the competitor grid, 8–12 ranked actions with the winnable ones marked and split by who can actually carry them out. Everything we could not measure goes in an appendix rather than being quietly dropped — and so does the full text of all 75 answers, so every claim we make can be checked against the words it came from.
Sites we cannot read are stopped at the free check, before any card is involved. It is the one stage that runs before you pay, and that is on purpose — it is the gate, not a step.
Google results, local signals and off-site footprint depend on other people's systems. If one is unavailable your report says so on the page where that section should have been. We would rather hand you a report with a hole clearly marked in it than quietly fill the hole.
An answer on page one, then the evidence for it — written to be read by the owner and not by an SEO agency. Two appendices close it: what we could not measure, and the complete text of every one of the 75 answers behind everything above.
One of 4 answers — including “leave this alone, you are already being recommended” — picked by a rule we publish at /method, not written by hand. It comes with the numbers it was derived from and how confident we are in them, and if any part of the method failed to run that confidence is downgraded automatically.
Before the evidence, the limits. What a sample this size can and cannot support, which questions are outside what anyone can measure today, and — when your numbers are too thin to diagnose from — a plain statement that they are directional only.
The core numbers on one page, and the one thing worth knowing first. If you read nothing else after the verdict, read this.
Every factual claim 3 engines make about you, checked against your own site. Wrong hours, wrong services, confusion with a similarly-named business — this is where those surface.
A grid of 18 unbranded questions × 3 engines, broken out by how close the asker is to buying — because being named to someone comparing vendors is worth more than being named to someone still learning the category. Share of voice, whether the answer holds when you ask again, and the pages AI reads instead of yours.
Structured data errors, pages AI cannot read, snippet settings that quietly block you from AI answers. If you serve an area: your Google Business Profile, reviews, and where else on the internet you exist.
Between 8 and 12 specific actions, ordered. Each says what to change, where, and what we expect it to move. No 60-item checklist.
What still works for your shape of business and what has stopped being worth the effort — filtered to your profile, not a generic best-practice list.
Four free places you can watch AI traffic yourself, with the exact settings. This is the section that makes buying a second report optional.
Your action list split three ways — what you can do today for nothing, what whoever edits your site can do, and what genuinely needs a developer — next to what the market charges for each.
We'd like to use Google Analytics to count how many people read this page and how many go on to buy a report — nothing beyond that. It writes a cookie to your browser and sends your IP address to Google, so we have to ask first. Saying no changes nothing about how this site works for you. What it collects.