The mental model

Learn these six ideas once and the rest of the tool reads like plain English. Everything on every screen is built on them.

1. The four engines plus Google AI surfaces, and “citation share”

The tool asks your questions to four AI answer engines: ChatGPT, Claude, Gemini, and Perplexity. It also tracks Google’s two AI search surfaces, Google AI Overviews and Google AI Mode, on the same panel. For each answer it records two separate things:

• Mention — did the answer name your brand at all?
• Citation — did the answer link your website as a source?

These are different, and the gap between them matters. Your headline number, citation share, is the percentage of tracked question-and-engine combinations where you were cited on a majority of tries. A related number, mention rate, is how often you were named. You can be mentioned a lot but rarely cited (people talk about you, but the engines pull their facts from someone else’s page), which points to a very specific fix.

2. Repeated passes and the majority rule

AI engines are not deterministic. Ask the same question twice and you can get different answers. The platform is built around that fact: every measurement can run each question multiple times per engine (the pass count is a platform setting), and whenever more than one pass runs, a result only “counts” if it shows up in the majority of those passes. Screens always show the real pass count for the run you’re looking at — the number on screen is the number that actually ran, never a claim.

3. OBSERVED vs INFERRED

Every finding in the tool is tagged one of two ways, and the distinction is enforced in the database, not just the wording:

• OBSERVED — something the tool directly measured, with a source reference you can trace back (a specific answer, a crawler visit, a Search Console row).
• INFERRED — a professional read of the evidence, clearly labeled as a judgment rather than a measurement.

When you are deciding how much to trust a line, check its tag. OBSERVED is fact; INFERRED is expert opinion grounded in fact. Reports keep the two in separate sections for exactly this reason.

4. Gates: some problems block everything else

Some issues are so fundamental that fixing anything else is pointless until they are solved. The tool calls these gates. The most common one: if your page’s content is not in the raw HTML (because it is rendered by JavaScript) or if AI crawlers are blocked at the door, then the engines effectively cannot see your page, so grading its content and schema is meaningless. When you see a gate-blocked state (labels like T1, T2, T3, S1, S5), the tool is telling you to fix that one thing first, then re-check, and the rest of the audit unlocks. It is not hiding your score; it is refusing to grade content nobody can read.

Why USCCA looks “gate-blocked”

The example brand, USCCA, sits behind a Cloudflare challenge that blocks automated visitors. The tool records that honestly as a gate rather than faking a score. Real verified crawlers may get through; the tool cannot from where it runs. This is the honest-empty-state promise in action, not a broken feature.

5. Honest empty states: “unmeasured, not zero”

The tool separates “we measured this and it is low” from “we have not connected the thing that measures this yet.” You will see phrases like:

• “not instrumented — no Cloudflare zone” — crawler analytics (which AI bots visited your site) need a Cloudflare connection. Until then this is blank, not zero.
• “not yet instrumented” — AI-influenced revenue needs a GA4 join that is a later phase. It is never faked.
• “insufficient window — 2 of 8 weeks” — trend and drift alerts need eight weeks of history to fire without crying wolf. Until the window fills, they stay quiet on purpose.

None of these are failures. Each one names precisely what to connect or how long to wait.

6. Progress is earned, not claimed: Verified vs. Confirmed

The tool recommends fixes in Actions, but it will not pretend a fix worked just because you marked it done. Each recommendation carries an honest status:

• Verified. The tool re-ran the measurement and directly proved the condition changed (for example, a crawler is now allowed, or citation share on a question moved). This is evidence, not a checkbox.
• Confirmed by you. For work the tool cannot measure directly, you can manually mark it complete. It is clearly labeled as user-confirmed rather than evidence-verified, so nobody mistakes your word for a measurement.

This is the same OBSERVED-vs-INFERRED honesty applied to your own progress. The retention loop closes only when movement is proven: measure, understand, act, then recheck to earn a Verified. Your headline number stays citation share; this idea is about how the tool grades the work you do to improve it.

That’s it

With those six ideas, open Your first 15 minutes and actually drive the tool. Every term you meet is also defined in the Glossary.