§ I — What we measure

Score My Business measures how often AI assistants — ChatGPT and Gemini, each on both its free-tier and its paid-tier model, plus Perplexity — describe and recommend your restaurant or hotel when real customers search for places to eat or stay. The output is a single 0–100 visibility score, broken down by eight customer intents and compared against your 8 closest AI-perceived competitors.

The score is the citation rate. We ask 80 questions on each of 5 AI models — 400 answers — count how many of those answers name your business, and divide. 155 of 400 is a score of 39. Nothing is weighted, curved or adjusted.

We print the count and the sample size next to the score so you can check the arithmetic yourself, and a 95% confidence interval so you know how much of a change is real rather than noise. At 400 answers that interval is about ±5 points, so a score moving from 39 to 42 means nothing and a move from 39 to 55 means something. As far as we know we are the only people in this category who publish either number.

Until 2026-10-15 the score also weighted each mention by where it appeared in the answer. We removed that: measured against our own data it worked out to a near-constant multiplier of 1.32, so it inflated every score by about a third without distinguishing one business from another — and it made the score disagree with the count printed beside it.

The eight intents we cover for restaurants:

  • Comparison — broad "best [type] in [city]" queries
  • Cuisine — specific cuisine queries (e.g., "where to eat pasta")
  • Location — neighborhood and landmark proximity queries
  • Occasion — first-date, anniversary, business-lunch queries
  • Dietary — vegan, halal, gluten-free, kosher options
  • Accessibility — wheelchair, kid-friendly, pet-friendly, parking
  • Atmosphere — rooftop, intimate, design-forward, historic
  • Recognition — Michelin, Bib Gourmand, James Beard, awards

For hotels, two intents change: cuisine becomes Amenities (pool, spa, gym, restaurant on-site) and dietary becomes Services (24h reception, airport transfer, concierge, room service). The other six intents are shared with restaurants. The scoring formula is identical.

§ II — How we measure

Each audit runs your business through a structured query library designed to test the breadth of how a real customer might ask AI for a recommendation. The library is fixed across all our audits — we don't customize it per business. That's the point: we measure how AI describes you when the user has no prior knowledge, asking the kinds of questions real customers ask.

Scan volume

One product, one volume — every buyer gets the same scan. There are no tiers to compare, and the question set does not shrink for your kind of business.

  • AI Visibility Audit ($99) — 80 distinct customer questions, one run each, across 5 AI models · up to 8 competitors · site readiness audit · 14 action items

AI assistants covered

Every audit covers five models. ChatGPT and Gemini matter most for restaurant and hotel discovery, and we scan each of them twice: the model served to users without a subscription, and the model served to paying subscribers. Those are different models and they do not always give the same answer, so your report tells you what each audience is shown. Perplexity is the fifth, included for everyone. Web search grounding is enabled on every call; without grounding, AI assistants answer from stale training data, which is not what your future customers actually see.

Variance handling, and what we publish about it

AI is non-deterministic, so we report a citation rate — the share of questions that name you — rather than a binary cited/not-cited. But we also measured how non-deterministic it actually is, because that determines what the number is worth.

Across 710 question-cells asked three times each, 79.6% returned an identical verdict every time. Repetition, in other words, buys very little: asking the same question a second time adds about 16% to the effective sample while doubling the cost, where asking a different question adds 100%. So we spend the budget on breadth — many distinct questions — rather than on repeats.

That means the real uncertainty comes from which questions we asked, not from AI's randomness. Every score therefore carries a 95% confidence interval computed from the sample size, and we show the sample size next to it. A narrow band means we asked enough questions to be confident; a wide one means we did not, and you should read the score accordingly. We would rather publish an honest band than a precise-looking number we cannot stand behind.

Citation detection

Every AI response is passed through an independent detection pass that extracts citation data: was your business named? In what position? With what context? We never trust the source AI to detect itself — that would be like having the witness grade their own testimony.

How geography enters the measurement

Geography is carried in the query itself — your city, and where we can establish it reliably, your neighbourhood. So we ask the assistants the same geographically-specific questions a customer would ask, and the grounded web search resolves them against live local results.

We want to be precise about what that does and doesn't mean: we do not currently route requests through a customer's home-country IP address. Query-level geography is what we measure and what we report.

Wall-time and delivery

Most audits complete well inside an hour. Because every report is reviewed before it goes out, we quote under an hour as the typical scan-to-draft window, with a hard SLA of 24 hours end to end. If we breach the 24-hour SLA, the refund is automatic — and you still get the report.

§ III — What we don't measure

For honesty's sake, here's what's outside our methodology:

  • Google AI Mode and AI Overviews — the AI answer that now sits at the top of a Google search results page, and the conversational mode behind it. This is almost certainly the largest AI surface for local intent, because it reaches people who never opened an assistant at all. There is no API for either: what they show is assembled per-session from the live index, the signed-in user's history and their precise location, and the only way to read one is to scrape a logged-in browser session — which produces a number we could not reproduce, stand behind, or compare between two businesses. We measure Gemini, which is Google's assistant and shares much of the same grounding, and we say plainly that it is not the same thing as the box on the search page.
  • Claude — not scanned on any tier. Anthropic's assistant is a real surface with a real audience, and the honest reason it is absent is budget rather than principle: five engines is what the $99 audit pays for. It is the next one we would add, and the first thing we will do before adding it is verify that it actually searches the web on our prompts rather than answering from training data.
  • Grok — xAI's API isn't yet stable enough for production scanning at our quality bar. We'll add it when it stabilizes.
  • Meta AI — no public API, despite ~1B monthly active users. We monitor the situation and will add coverage if access opens.
  • Individual user personalization — we measure aggregate AI behavior (what AI assistants tend to say) rather than what any one signed-in user sees with their personalization profile applied. Customer-specific personalization is unobservable from outside the platform.
  • Search ads — paid Google / Bing ads are out of scope. This is an organic-recommendation audit.
  • Direct booking conversion — we measure visibility, not bookings. Visibility is upstream of conversion. The audit's actions help your visibility; tracking conversion is your CRM's job.

§ IV — Limits and caveats

AI is non-deterministic. Re-running the same audit on a different day will produce a slightly different score — typically within a small confidence interval. We mitigate via multi-run averaging across platforms.

AI assistants update their training data and grounding behavior continuously. Citation patterns can shift week-to-week, especially when platforms release new model versions. We re-validate the query library quarterly and publish library changelogs (available at /blog).

We can't tell you exactly why AI cites you (or doesn't) — that's inside model weights we can't inspect. We can tell you reliably whether AI cites you, in what context, and which signals correlate with citation across our customer base. Your action items are derived from those correlations.

§ V — Frequently asked questions

How is this different from SEO?

SEO measures how Google ranks your website on a results page — a 10-blue-link world. Score My Business measures something different and newer: how AI assistants describe and recommend your business in conversational answers. SEO is keyword-driven. AI recommendation is sentiment + reputation + grounding-data driven. Different game, different scoreboard. We measure the new one.

Can I actually improve my AI visibility score?

Your audit's action list explains how, and the site-readiness part of your score is the fastest-moving piece — it is a set of deterministic checks on your own website, so you can fix it and verify it yourself within days. The levers our audits most often surface: (1) a complete English-language Google Business Profile description, (2) machine-readable pages — schema markup, an llms.txt, menus as text rather than PDFs (we generate these files for you), (3) review-language coverage in the markets you want visibility in, (4) press mentions in publications AI assistants are likely to cite, (5) consistent name/address/phone across the web.

We do not yet publish a figure for how much a given action moves a score, because we have not measured one. A one-time audit measures you once, so we have no before-and-after data of our own; any number we quoted would be invented. We would rather say that than make one up.

How accurate is the free score?

The free score uses a narrower scan set than the paid audit — enough to give you a directionally-correct visibility number, but without the depth of intent breakdown, verbatim AI quotes, or competitor cluster analysis. Use the free score to learn whether you appear at all. Buy a paid audit when you want to know exactly where you rank, what AI says about you, and what to do.

What if my score is low?

That's the most useful version of the report. A low score means there's signal in the gap between where you are and where competitors with similar fundamentals already are. Every audit ships with 14 specific actions, ordered by our judgement of impact against the effort involved, and each one is tied to the scan data that prompted it — the questions you lost and who was recommended instead. We do not tag actions with an expected point gain, because we have not measured one.

How long does delivery take?

Under 24 hours is our SLA. The scan and draft typically take under an hour; every report is then reviewed before it is sent. Refund window: if delivery exceeds the 24-hour SLA, we refund automatically and still send the report.

What if my data isn't on Google?

We try Google Places first; if your business isn't listed, we fall back to manual entry. You provide the business name, address, and category. The audit still works — Google Places gives us slot data (cuisine type, neighborhood, landmark proximity) faster, but isn't load-bearing for the scan itself.

Are the AI quotes you show me real?

Every verbatim AI quote in your report is captured live during the scan and stored unmodified to Cloudflare R2 (every raw response, every audit, retained indefinitely). If you want to see the source response, ask support — we'll send the raw scan data. We don't paraphrase, condense, or invent quotes.

What's a "competitor" in this audit?

Two layers. (1) AI-detected competitors: the businesses ChatGPT/Gemini consistently mention alongside yours when answering "best restaurants in [your city]" type queries. We extract these automatically from your scan. (2) User-picked competitors: you can swap in any of the 8 slots yourself before the audit runs.

Both kinds are measured the same way, and it is worth being precise about what that means: we do not run a separate scan for each competitor. Every competitor's citations are read out of the same answers we asked about your category — the same 80 questions, the same 5 models, the same moment. That is why 8 competitors cost no more than 3, and it is also a stronger comparison than scanning them separately would be, because you and they are measured in the same responses.

Do you scan in different languages?

Not yet. Every audit currently runs its queries in English. Scanning in your customers' other languages is in development, and we won't sell it until it measures something genuinely different.

What about Yext, BrightLocal, or other GEO tools?

Yext is enterprise-priced ($10k+/year) and built around managing structured listings. BrightLocal is SEO-focused. Score My Business is purpose-built for AI-recommendation visibility specifically, designed to be affordable for independent restaurants and hotels (one-time $99 once), and methodology-transparent (we publish exactly what we scan and how we score).

Does ordering an audit affect my AI visibility?

No. We're observers, not participants. The scan reads how AI describes you; it doesn't modify your business profile, reviews, listings, or any underlying signal. Your audit is a measurement, not an intervention. The intervention is what you do based on the action list.

Privacy — what data do you store?

We store your business name, Google Place ID, and every raw AI response from your scan, indefinitely (audit trail / re-run capability). We do NOT store payment data (Stripe-handled). Your email is used only for delivery and lifecycle communication (you can opt out anytime). We never sell or share data with third parties. GDPR + US privacy compliant. Details on the privacy page.

Can my agency white-label this?

Agency pricing is handled case by case for now — email us. Bundles are not sold self-serve, because we have not built redemption for them. Small agencies running quarterly audits across a portfolio. We don't currently offer white-label co-branded reports — your client sees "Score My Business" as the audit provider. If you need agency-tier features (custom branding, API access, multi-client dashboard), reach out about Citeabl SaaS instead.

What's Citeabl SaaS — should I buy that instead?

Score My Business is one-time. Citeabl is a recurring subscription that will monitor your AI visibility weekly with a live dashboard, weekly delta alerts, peer-cluster tracking, and continuous action checklist. It is not open yet; audit buyers get the price of their audit credited toward Citeabl. Most owners start with one SMB audit, see what their score looks like, then subscribe to Citeabl when they want continuous tracking.

§ VI — Open the data

We intend to publish summary statistics from anonymised audits — no business names, no identifying data — covering citation rates, intent distribution, and how platform behaviour shifts as models change. We have not published any yet, and we will not until the sample is large enough to support the claims. Saying so is part of the methodology.

Full anonymized scan dataset available on request for academic research. Email support@scoremybusiness.ai with your institution and project description.

§ VII — Contact

Methodology questions, press inquiries, research access: support@scoremybusiness.ai

Response time: 1 business day.