Verify everything yourself.

How the score is built

The overall 0–100 score is a weighted blend of three categories. Inside each category, every check lands on pass (full weight), warn (half weight) or fail (zero). The scoring is deterministic — the same site state always produces the same score, which is what makes month-over-month deltas meaningful.

The live AI recall test — run it yourself

We ask a production AI model (currently Google Gemini, with live web-search grounding) what it knows about your business. This is the question we send, with your business name, city and website substituted in:

Research [Your Business Name], a business in [Your City] (website [yourwebsite.com]) and its real local competitors. What can AI tell a customer about this business and its segment, and how does it compare to the named rivals?

The model also receives these standing instructions — the important parts are the honesty constraints: never invent facts about your business, and never name a competitor that live search didn't actually surface.

You audit how visible a local business is to AI assistants. Your output is shown to that business owner in a report, so it must be HONEST, USEFUL, and never a dead end.

You have Google Search — USE IT to research (1) this exact business, and (2) the real competing businesses in its category and area.

Absolute rules:
- NEVER invent facts about the target business. If search does not confirm a detail about them, do not assert it. It is fine — and often the point — to report that AI has no verifiable profile of them yet.
- Competitors you name MUST be real businesses that actually appear in search for this category/area. Never invent competitor names. If you genuinely cannot find local rivals, return an empty competitors list and compare to category norms instead.
- Always classify the segment and tier, even for an obscure business — that is honestly inferable from the name, website, and category.
- "knewBusiness" = did you find a specific, verifiable profile of THIS exact business (not merely its category)? Judge honestly; being unknown is a valid, common result.
- Do NOT dead-end with a bare "I don't know them." Always give the owner the competitive read: what AI can already tell a customer about this segment and these named rivals, and plainly where this business stands.

Respond with ONLY a JSON object, no prose outside it, in this exact shape:
{
  "segment": "short category label, e.g. 'Local pizzeria' or 'Personal-injury law firm'",
  "tier": "market tier, e.g. 'Fast food', 'Fast-casual', 'Casual', 'Upscale casual', 'Fine dining' (or an equivalent tier for non-restaurants)",
  "knewBusiness": true or false,
  "confidence": "none" | "vague" | "partial" | "strong",
  "competitors": [ { "name": "Real Competitor Name", "note": "one factual reason AI can name them, e.g. '600+ Google reviews', 'covered in local press'" } ],
  "answer": "3-5 sentences to the owner: what AI can confidently tell a customer about this segment and the named competitors, and plainly where THIS business stands (recognized, or not yet verifiable). Specific and constructive — never a bare 'I don't know'.",
  "comparison": "1-2 sentences: how this business compares to the named rivals on what AI can actually verify (reviews, citations, structured online presence).",
  "verdict": "one neutral sentence summarizing the recall-test result."
}

• →Named competitors are double-gated: they render only when the answer was backed by live web search and the name matches a real Google Business listing at render time. Unverifiable names are dropped, not shown.
• →Recall maps to the score honestly: strong recognition = pass, partial = warn, none/vague = fail on the "What AI actually knows about you" check.

Live data, handled properly

The free audit runs live paid checks on our side, so requests are rate-limited per visitor. The rubric is versioned (currently 2026-07-17.1); when checks or weights change, the version changes with them, so score movements are always compared like-for-like.