Every number in your report traces back to a real website crawl, a real Google lookup, or a real Wikidata/Wikipedia record — never an AI model's opinion of how well you're doing. Here's exactly how, check by check, with links to verify every claim yourself.
Most "AI visibility" tools ask an LLM to judge your brand — "rate this business's accuracy," "score the sentiment of this mention" — and report that judgment as your score. That's a plausible-sounding number with no way to reproduce it: ask the same question twice and you can get two different answers.
We do the opposite. Every one of the 32 checks across Entity Auditor, Schema Builder, and Review Manager is computed from hard, independently-verifiable evidence: does this exact schema.org field exist on your homepage, does this exact Wikidata entity exist, does Google's own Places API report this exact review count. The check either finds the evidence or it doesn't. Nothing in between is a guess.
The one agent that genuinely uses AI in its evidence-gathering is AI Platform Scanner — and even there, the AI's job is to generate realistic customer search queries and fire them at real AI platforms. Whether your business gets mentioned in the real response that comes back is then checked with plain text matching, not another AI's opinion of how it did.
A binary pass/fail hides an important distinction: there's a real difference between "we checked and it's genuinely missing" and "we couldn't check this at all." Conflating the two either unfairly penalises a business for our own tooling limits, or quietly hides a real gap behind a free pass. Every check in this report resolves to exactly one of six states:
| Status | What it means | Credit |
|---|---|---|
| ✓ Pass | We checked, and the evidence is there — a real schema field, a real Wikidata entity, a real Google rating above the threshold. | Full |
| ~ Partial | The evidence is there but incomplete — e.g. 2 of 4 required schema fields present. | Half |
| ! Warning | The data exists but needs a fix (a schema syntax error) or verification (a possible entity-name mismatch) — not a full failure. | Partial |
| ✗ Fail | We checked, and it's genuinely absent. A real, fixable gap. | None |
| — N/A | We could not evaluate this at all — no API key configured for that data source, an optional upgrade you haven't connected, or a site that temporarily blocked our crawler. Not a fact about your business. | Excluded from your score entirely (see below) |
| ◷ Manual | This needs a human judgement call we won't fake with an algorithm (e.g. Wikipedia notability). Shown for context, never scored. | Not scored |
An earlier version of this scoring gave full credit for N/A checks within your 100-point score — mathematically fair, but it meant a business with several unverifiable checks (no Google data source configured yet, an optional connection skipped) could show a score that looked stronger than what was actually verified, with no visible way to tell the difference from a score backed by real evidence.
We now exclude N/A checks from both the numerator and denominator of your score entirely, then rescale back onto a consistent 0–100 scale — so your score reflects only what was genuinely checked, and a separate "Needs More Information" section tells you exactly what wasn't, and why.
We found this the hard way. A real, dominant local business — #1 on Google for its category, directly recommended by Claude when we tested it — came back with a near-zero Entity Auditor score. The cause wasn't the business at all: our own crawler had temporarily failed to load their homepage, so every schema check saw empty data and scored it as "no schema found," indistinguishable from a business that genuinely had none.
We fixed it so a failed crawl now reports as not_applicable — "we couldn't check this run" — rather than silently scoring as a hard fail. The business's real score, once the crawl succeeded, reflected genuine, fixable gaps (an incomplete address schema) rather than a false "you have nothing" verdict. That fix — and several others like it — is why the status contract exists as a hard rule, not a guideline.
Can AI confidently identify who you are? Pulls from your site's own schema.org markup, plus independent off-site lookups against Wikidata, Wikipedia, and Google Knowledge Graph.
Checks for a parseable name field in your homepage's JSON-LD — not whether your name is visible on the page, whether AI crawlers can read it as structured data.
A real lookup against Wikidata's API, with a homonym safeguard — a business sharing its name with an unrelated place or landmark is flagged as a mismatch, not silently credited.
A direct API check — found or not found, nothing inferred.
A connected OAuth link is proof of ownership by construction — optional, never a penalty for skipping it.
Does your site publish the structured data formats AI systems parse directly — FAQ schema, breadcrumb navigation, and review markup?
Checks for a real FAQPage block with at least 3 question/answer pairs — not just that a FAQ section exists visually on the page.
This check defers to Review Manager's own review-schema check when neither is present, specifically so one missing block never gets penalised twice across two agents.
Real Google review volume and rating — via a public Places API lookup that works for any business, plus (optionally) your connected Google Business Profile for richer data — compared against both a fixed threshold and your actual local competitors.
A real count from Google, not your on-site schema's self-reported number.
Compared against the ~10 businesses Google's own relevance ranking surfaces for your category in your city — a fixed "10 reviews" bar means nothing if everyone nearby has 500. The average is review-count-weighted, so one competitor's 1,800 reviews properly outweighs another's 6.
Real prompts, fired at real AI platforms, checked for whether your business is actually named — unprompted, the way a real customer's question would surface you or not.
We don't ask you to take our word for any individual check. Every row in your report carries a direct link to the same independent tool or public record we used to compute it:
Structured-data checks get three independent validators rather than one — if one can't reach your site, the other two still confirm the result. If you ever disagree with a check, click its verify link and see the same evidence we saw.
Being evidence-based cuts both ways: we won't dress up a correlation as a guarantee just because it would sound better in a sales page.
Run the free audit and every check above links straight to the evidence behind your actual score.
Run your free AI visibility audit →Questions about how a specific check works:
audit@aeogoogle.com