When an AI answers a question about a business, it assembles the reply from web pages and reuses the facts those pages state. A page's own site cannot be the source for a fact it never writes down. So we use a simple, self-checkable diagnostic: read only the first ~100 words of a page and ask whether an AI could state the facts a buyer needs to decide.
This page explains the method and the data behind it. It is our own diagnostic, not a platform's ranking signal, and a score is a reading of your opening, not a prediction about any algorithm.
The method
Read the first ~100 words of a page. Score it 0 to 5 on whether it plainly states the decision-facts a buyer weighs. For the vertical we scored first (dental implants) those were price, financing, provider, candidacy, and options. The fact-set is business-agnostic: adapt it to what a buyer in your field actually decides on (for most service businesses, price, a named provider or credential, and who it is for).
Give one point for each decision-fact stated plainly in the opening. The 0 to 5 result is simply how many of the buyer's decision-facts are present up top. It is not a causal score of a page's "AI readiness," and it is not a citation-probability prediction. It measures one thing: what a machine could read from your intro.
What we found
Primary scan: dental-implant pages, dated 2026-09-05.
We scored each page in two passes and adjudicated the 7 disagreements (of 735 cells) against a fixed rule: a discount like "$500 OFF" is not a starting price, and a name that appears only in a testimonial is not a stated provider. That is why price sits at ~3% rather than ~5%; counting discounts as price is the only thing that lifts it. The two-pass agreement was 99.05%, a consistency measure between passes, not accuracy against ground truth.
Broader run: 7-vertical AI Visibility Index, dated 2026-09-06. The same pattern held across 852 scored page and location records across seven unrelated industries, where about 84% scored two or fewer of five. These are page/location records, not verified independent businesses (some domains carry several locations), each vertical used its own fact-set, and the dental scan above is part of this index rather than a separate replication. Most openings, in most fields, do not state enough for a machine to repeat.
How to read a low score
A low score is a diagnosis, not a verdict.
- The first-100-words window understates your full page. The fact may sit lower down; the test only asks what is up top.
- A blocked or JavaScript-heavy page is unconfirmed, not absent. It means a reader (human or machine) could not confirm the fact from the opening, not that your page failed.
- This is association, not causation. We are reporting what we measured in a page's opening, not a proven lever inside any model.
What a low score usually tells you is concrete: the facts a buyer wants are not within the opening we measured. That is fixable.
Run it on your own page
Read only your first 100 words and ask: could an AI state the facts a buyer needs to decide? Score one point per decision-fact stated plainly up top. That 0 to 5 number is your opening's readiness.
Prefer to have a model do it? Paste your opening into ChatGPT with a short scoring prompt and it returns the 0 to 5 plus the missing facts. A hosted one-click version needs a paid model API, so for now the manual check and the prompt are the free ways to run it.
The 100-Word AI Readiness Test
Fill in your service page and treatment. We build the check for you, then open it in ChatGPT.
Evaluate whether this page gives an AI search assistant enough specific, verifiable information to shortlist this clinic for [treatment] in [city]. Page: [paste your service page URL or text] What decision-making facts can you extract? What is missing or buried? Which essential answers are missing from the first 100 words? Do not assume facts that aren't stated on the page.
Tip: turn on web browsing in ChatGPT so it can read the page. If it can't, it will ask you to paste the page text, which works too.
Paste your opening and the check scores it 0–5 on the decision-facts a buyer weighs, then lists the ones it could not find. It reads only what you paste, so it scores your opening, not your whole page or your odds of being cited.
Method and limits
- Our own diagnostic, not Google's or OpenAI's. We never attribute it to a platform.
- Unit and definition: one record = one page or location URL. Score = count of the five decision-facts stated plainly in the first ~100 words (0 to 5). We adapt the five facts per vertical.
- Dental scan: 153 gathered, 147 scored, then 3 more set aside as extraction-unconfirmed, leaving 144 reported, 17 US metros, dated 2026-09-05. Two passes, 99.05% cell agreement, 7 disagreements adjudicated by fixed rule; reported figures (mean 1.64, 125/144 at ≤2, price 4/144) are on the 144-record adjudicated set.
- AI Visibility Index (dated 2026-09-06): 852 scored page/location records across 7 verticals, mean 1.75, about 84% at two of five or fewer. Records, not verified independent businesses; each vertical uses its own five-fact set, so results are per-vertical, not one universal scale.
| Vertical | Records | Mean | Fact set (examples) |
|---|---|---|---|
| Dental implants | 147 | ~1.6 | price, financing, provider, candidacy, options |
| Plastic surgery | 131 | 1.42 | price, surgeon, candidacy, options, financing |
| HVAC | 107 | 1.79 | price, license, service fit, options, financing |
| IVF / fertility | 103 | 1.57 | price, physician, candidacy, options, financing |
| Personal-injury law | 124 | 2.17 | fee, attorney, case fit, process, proof |
| Roofing | 124 | 1.91 | price, license, service fit, options, financing |
| Veterinary | 116 | 1.72 | price, vet/credential, fit, options, financing |
Adjudication log (the 7 cell disagreements)
All seven two-pass disagreements resolved the same way, toward the stricter reading. On the strict pass over all 147 records that gives mean 1.61 and 128 of 147 at two or fewer; after excluding the 3 extraction-unconfirmed records the reported result is mean 1.64, 125 of 144 at two or fewer (86.8%), price 4 of 144 (2.8%).
| # | Metro | Practice | pass1 → pass2 | Decision and basis |
|---|---|---|---|---|
| 42 | Atlanta | Artistic Dentistry | 2 → 1 | name appears only in a testimonial, not the service body → provider not credited |
| 44 | Atlanta | Dr Brett Langston | 1 → 0 | prosthodontics / CEREC is a specialty and crowns, not an implant-type option → not credited |
| 82 | New York | Herald Square Cosmetic Dental | 1 → 0 | "$500 OFF each implant" is a discount, not a stated implant price → price not credited |
| 104 | Portland | Revolution Dental | 3 → 2 | name only in a patient testimonial → provider not credited |
| 129 | Nashville | Solace Oral Surgery | 2 → 1 | "affordable cost" is not payment or financing info → financing not credited |
| 145 | Las Vegas | Providence Dental | 2 → 1 | "$1,200 off package" is a discount, not a stated price → not credited |
| 146 | Las Vegas | Sedona Ranch Dental | 2 → 1 | "$1,200 OFF package" is a discount, not a stated price → not credited |
Extraction-unconfirmed records (excluded, not re-scored): three records (Implant Center of Miami, Inspired Dental LV, Windy City Implants) had openings dominated by address or menu text, so their captured "opening" is not the page's real opening. Each currently scores 0, so a correct extraction could only raise them, not lower them. We mark them extraction-unconfirmed and exclude them from the reported figures, keeping the original 147-record numbers as a historical note. The core finding (most openings do not state enough) holds on the 144-record calculation. Association, not causation: this scores what is present in the opening, not how any model ranks or decides, and it is not a promise of a position.