The five signals and their weight
The composite is a weighted blend. If a signal cannot be measured, it is left out and the rest are reweighted; nothing is invented.
| signal | weight |
|---|---|
| Google profile Does the listing exist and is it filled in: website, hours, at least two categories, and real photos. | 25% |
| Map results For the searches customers actually type, where do you place. The top three is the map pack; that is where the calls go. | 25% |
| AI readiness Can AI assistants read the site: structured data, honest title and description, crawlers allowed, and an llms.txt file. | 10% |
| AI mentions When a customer asks an AI assistant who to call, how often you get named. Sampled three asks per question, reported as a rate. | 20% |
| Reviews Rating strength and review depth together. Two businesses side by side, the fresher review wall wins the tap. | 20% |
The honesty rules
AI answers are sampled: every question is asked three times per assistant and reported as a rate, because assistants vary. If no AI signal can be collected at all, no composite is scored; a made-up number helps nobody. Site checks read the top 250 pages, not every page, and the report says exactly how many were read. Sampled and directional, not precise.
Site health deductions
Each issue deducts points scaled by the share of pages affected. Broken pages and dead canonicals are heavy, missing alt text is light.
| issue | weight |
|---|---|
| broken page | 30 points |
| dead canonical | 25 points |
| broken links | 15 points |
| title missing | 10 points |
| title duplicate | 10 points |
| meta missing | 8 points |
| meta duplicate | 8 points |
| schema missing | 6 points |
| h1 missing | 5 points |
| canonical missing | 4 points |
| h1 multiple | 3 points |
| images no alt | 3 points |
Why publish this
Because the score only matters if you can trust it. Any agency can hand you a mystery number; this page is the whole recipe. If you think a weight is wrong, tell me, and if you are right I will change it and say so here.