35 technical tests and 3 live tests in ChatGPT, Perplexity and Gemini add up to one AI Readiness Score from 0 to 100. This page covers what we test and why, and how we keep the method stable from one run to the next.
Every automated test is pass/fail or scored against a threshold, so the same website gets the same result on every run. Server response time is the one value that moves. We measure it again each time and score it against a threshold.
Each test cites official crawler documentation from OpenAI, Anthropic, Google, Perplexity or Bing, or an open web standard (RFC 9309, schema.org). Where the documentation says nothing, we say so and test it ourselves.
A technically clean site is only the means. What counts is whether AI recommends you, so we ask real questions in ChatGPT, Perplexity and Gemini and screenshot every answer.
AI crawlers and AI search change quickly, so the methodology has a version number and a changelog. Every report states its version. When we compare scores from different versions, we say so.
Three checks give partial credit in proportion, and two more are scored against thresholds. We document the normalisation and every partial-credit rule, so the numbers in a report always add up.
Check maximums add up to 103 points (including the 1-point llms.txt bonus); the score is normalised: round(100 × earned / 103).
The full list of 35 tests, with scoring, thresholds and sources for each, comes with the paid audit report, along with your own result on every one.
Category G is the only part we measure by hand, and we do that on purpose. Each business gets three fixed questions, asked in the same engines in the same order: ChatGPT, then Perplexity, then Gemini. We use the free tiers, because that's what your customers use. Every answer is screenshotted and attached to the report. In monthly monitoring we ask the same questions again, word for word, so you can see how the answers change over time.
On its own, a score doesn't tell you much. Every report puts yours next to the median and the leader in your industry, from our own benchmark:
Benchmark: wave 3 (28 September 2026, 419 websites) computed with methodology v0.4.1; wave 2 (10 September, 449 businesses) with v0.4 and wave 1 (July, 447) with v0.2. Reports always label comparisons across versions. For the full distribution, category statistics and the checks businesses fail most, see the benchmark report →
In one year, AI crawlers went from a handful to a dozen, each with its own job: training, search, or fetching a page for a user. So every month we read the bot documentation from OpenAI, Anthropic, Google, Perplexity and Bing. Every quarter we check all tests, thresholds and weights against where the market is. Any change to scoring ships as a new version with a fresh benchmark run, so medians and client scores stay comparable. When something big happens, like a new crawler or an engine reading robots.txt differently, we don't wait for the next cycle. We review it within days.
No points or thresholds changed. This version fixes how the entity checks read a page: a copyright year ("© 2025") no longer passes as a postal address, a date no longer passes as a phone number, an "Org.nr: NO … MVA" line now counts as an organisation number, and a name like "Løvås Rør" now matches its domain lovasror.no. The firewall test also retries once after a network error. Like every change that affects results, it shipped with a new benchmark wave.
The firewall test now covers nine crawlers instead of four: the search bots of Apple, Amazon, Mistral, Meta and DuckDuckGo joined OpenAI, Anthropic and Perplexity. The reason is Cloudflare. From September 2026 it blocks training and agent bots by default but lets search bots through, so "blocked for AI" now depends on which bot you ask about. A new check confirms that both the bare domain and the www version respond. We added it after an audit found a company whose bare domain had no DNS record at all.
The firewall test went from one bot to four, Claude-SearchBot was added, and Gemini replaced Copilot in the live-test protocol.
Current version: v0.4.1
Enter your domain and get a first AI Readiness Score right away.