We publish how we grade ourselves.

One number: the share of real buyer questions in which an engine names you. If we test 60 questions across the engines we query and you are named in 9 of them, your Recommendation Rate is 15%.

It is deliberately not “rankings,” not “visibility score,” and not “share of voice.” Those are constructions that can be redefined when they stop flattering the vendor. Named or not named is a fact.

At onboarding we build 60 questions from how patients actually ask — service plus city, service plus qualifier, comparison, price, and problem-first phrasing. They span your full service mix, not just your most profitable one.

The panel is locked and shared with you in writing on day one. We cannot add easy prompts later to make a trend line look better. Any change is logged, disclosed, and reported alongside the old set for a full cycle.

ChatGPT, Google Gemini, Perplexity, and Claude. Each is queried through its own provider’s API with web search enabled, and we print the answer word for word. We never ask one model what another would say — that is a guess with a citation attached, not a measurement.

We report engines separately and never blend them into one figure, because they demonstrably do not trust the same evidence. A source that carries weight with Perplexity may barely register with ChatGPT. An API answer is that engine’s answer to that question at that moment; it is not a screenshot of the consumer app, and given ~85% volatility it will not be identical if you ask again next month. That is why we run the panel on a schedule rather than once.

Rating floors behave as hard filters rather than ranking signals. Below an engine’s floor, no amount of other work makes you eligible — which is why rating repair is always the first move we make.

Derived from your actual market rather than a list you supply. We take the practices that engines and local sources genuinely surface for your services in your city. You may add competitors; you may not remove one because the comparison is uncomfortable.

Roughly 85% of businesses named in AI local results turn over as models re-evaluate fresh data. A practice named today may be absent in six weeks with nothing having changed on its end.

This is the single most important fact about this channel, and it has a commercial consequence we’ll state plainly: it is why a one-time “AI rebuild” decays, and it is also why we can charge a monthly fee in good conscience. Visibility here is a position you hold, not a project you complete.

On May 15, 2026 Google published documentation rejecting AEO and GEO as disciplines separate from search, and named the tactics that had been packaged as specialties:

“Prioritize effective SEO strategies over ‘AEO/GEO hacks’. For Google Search, you can ignore tactics like ‘chunking’ content, creating unnecessary AI text files (like llms.txt), or pursuing inauthentic mentions.”— Google Search Central, AI Optimization Guide

We agree, and we act on it. We will not sell you:

• An AEO or GEO line itemIt's search work. Charging separately for it is a markup, not a discipline.
• llms.txt files or content chunkingNamed by Google as ineffective.
• Purchased or inauthentic mentionsNamed by Google as ineffective, and a liability besides.
• Incentivised or fabricated reviewsIllegal under FTC rules, and the fastest way to destroy a practice we're paid to protect. If you ask us to, we'll end the engagement.

The scan on our homepage is real and it is honest about what it is. It measures your presence in the source corpus engines ground on, your listing footprint across the platforms they cross-reference, your website’s machine-readability, and your eligibility against each engine’s rating floor. It reports a source corroboration score — not Recommendation Rate, which requires querying the engines and is only produced by the Benchmark.

Two limits worth stating plainly, because you will notice them otherwise. A practice can rank in these results and still not be an engine’s pick. And a practice with an excellent reputation can be named by an engine while ranking nowhere here, because engines also read review aggregators like Healthgrades and Birdeye that don’t surface in ordinary search. Corroboration is a strong leading indicator of being named. It is not a substitute for asking the engine.

It also asks a real engine. The free scan puts three buyer questions to ChatGPT through the OpenAI API with web search enabled and prints the answers verbatim. Three questions, one engine — enough to prove the instrument is real, and we label it as ChatGPT-only on the page rather than letting you assume it speaks for all of them. The Benchmark runs the full sixty-question panel across Gemini, ChatGPT and Claude.