What we check
How We Test
Everything we sell rests on a measurement. So here is exactly what the measurement is, what it can support, and what it cannot.
The test
The free Patient Shortlist Check runs 10 patient-intent prompts on every engine it has a key for — 50 observed outcomes at full coverage.
The Shortlist Fix Pack runs a deeper pass first: 46 prompts across 5 engines — 230 observed recommendation outcomes — before anything is written. Every one is a different question on a different platform; nothing is asked twice.
The 46 prompts are 16 universal patient questions plus 24 treatment-specific ones: four priority treatments × six patient intents — discovery, recommendation, desired outcome, expertise, trust, and value/decision.
Every prompt is stored exactly as it was sent, tagged with the prompt set version
(2026-09-14.v1). When we change the questions we publish a new version rather
than editing the old one, so a rerun a year later is comparable to the original.
Why we never ask the same question twice
A generated answer is one sample, not a ranking. Until 2026-09-15 we answered that by asking the decisive questions again in the same sitting, which was the wrong fix: two runs of one question are correlated, so the second one mostly re-measures the first while costing the same as a new one.
Breadth is what separates "consistently absent" from "absent once", and breadth means more questions and more platforms — both of which are independent samples. So every observation in a report is a distinct question on a distinct engine. Measuring the same market again over time is a different question, and re-running your check later is how you ask it.
What we record
For every observation: the engine, the model where the platform reports it, the mode and grounding state, the exact prompt, the geography it was pinned to, the full answer text, the businesses named and in what order, whether the practice appeared, where in the answer it appeared, the sources cited where the engine shows them, and the timestamp.
Raw answer text is never discarded. Every conclusion in a report traces back to observations you can read in the appendix.
When an engine doesn’t answer
Engines time out, decline, and occasionally name nobody. Those outcomes are recorded as what they are and excluded from the figures. They are never counted as "your practice was not mentioned", and a missing observation is never filled in.
The Patient Shortlist Index
A 0–100 diagnostic index built from five components:
| Component | Weight | What it measures |
|---|---|---|
| Shortlist Presence | 35% | How often the practice was named at all, against the most-named practice in the same tests. |
| Recommendation Prominence | 20% | Where in the answer it appeared — primary, core shortlist, or also-consider. |
| Treatment Coverage | 20% | Visibility for the four treatments the practice most wants patients for. |
| Cross-Engine Consistency | 15% | Visible across engines, or strong on one and invisible on the rest. |
| Information Accuracy | 10% | Whether what is said about the practice matches its verified facts. |
Score configuration psi-1.0. Every component is
measured against the strongest practice observed in the same market under the same prompts, not
against an absolute target — in a market where AI names almost nobody, appearing as often as the
leader is a good score, not a failing one.
What this is not
AI-generated recommendations are dynamic. Results vary by platform, wording, timing, location context, source availability, personalisation and model behaviour.
The Patient Shortlist Index is The Aesthetic Shortlist’s own diagnostic measure under documented test conditions. It is not an official OpenAI, Google, Anthropic or Perplexity ranking, and it is not a guarantee of future recommendation placement.
We do not claim a causal link between an implementation and a change in AI behaviour where only a correlation has been observed.
What we promise
We show you how major AI systems currently represent and recommend your practice under documented patient-intent tests, then identify the controllable public-information gaps worth fixing.
We never promise guaranteed placement on any platform, a "#1 in AI" position, a causal increase in patients or revenue, or proprietary access to any AI recommendation system. No legitimate provider has those things.
What we fix, and why
There is no magic switch. The controllable work is crawlability, factual clarity, useful first-party content, location and entity consistency, provider clarity, treatment clarity, structured data that matches what the page visibly says, internal linking, and consistent public business data.
We do not use fake citations, fake reviews, spam directories, unsupported superiority claims, invented credentials, or mass AI-targeted content. Those are the things that get a practice in trouble, and they are also the things platform guidance explicitly warns against.