How we measure

Every score is a weighted average of checks we can show you. Weights live in a configuration file and are listed here. Anything we couldn't measure is shown as "Not measured" and left out of the average, never guessed.

1. AI-Readability (Readable)

Six sub-scores, weighted: Access 25%, Rendering 20%, Structure 15%, Extractability 15%, Structured data 15%, Performance & accessibility 10%. Each check scores pass 1, warn 0.5, fail 0. The home page counts 40%, the other key pages share the rest.

Rendering compares the raw HTML a non-JavaScript crawler receives with the rendered page: text 50%, headings 20%, links 20%, prices/product data 10%.

Caps: AI search bots are blocked, so nothing else on the site can be read by them. (score capped at 40) Search engines are blocked or the page is noindex; most AI tools rely on search indexes. (score capped at 25)

Bot-access tests that send an AI crawler's user agent are indicative: CDNs verify real bots by IP address, reverse DNS or signatures, so a real bot may be treated differently. Confirm in your CDN dashboard or server logs.

2. AI Visibility (Visible)

For each AI engine we ask your buyer questions 3 times and combine: mention rate 35%, position when mentioned 20%, citation of your website 20%, share of voice vs competitors 15%, accuracy of what’s said about you 10%. Questions that name you count only toward accuracy. Competitors are scored with the same answers (without accuracy) to give an approximate rank.

Google and Bing use organic positions weighted by a click-through curve (First Page Sage 2026, checked 2026-11-08). Copilot has no public answer API and is shown as a proxy via Bing. Meta AI is a manual check.

The overall score weights engines by estimated US usage (low confidence): google-organic 50%, google-aio 13%, chatgpt 14%, gemini 10%, bing 5%, copilot 2%, claude 3%, perplexity 2%, grok 0.5%, meta-ai 0.5%.

Confidence is high with 15+ questions, 3+ repeats and run-to-run spread of 10 points or less; medium with 8+ questions and spread under 20; otherwise low.

API answers approximate, but are not identical to, what people see in the ChatGPT, Gemini or Claude apps. Results vary between runs and over time.

3. Agent-Readiness (Actionable), forward-looking

Can an agent find your offerings (30%), read product and service data (35%), and take the next step without JavaScript or unlabeled forms (30%)? Emerging protocols such as WebMCP and MCP can add at most 5 points; they’re pilots with no proven impact.

What we don't claim