Methodology
How the Viscovery AI visibility score works
Updated · written from the code that computes the score
Viscovery shows two kinds of number: the 0–100 score the free checker returns, and the five health scores at the top of a paid report. This page states how each is produced, what data it uses, and what it does not claim.
The free score
The checker fetches /robots.txt, /llms.txt and the homepage over the public internet. No language model is involved in scoring. Each signal is a fact about a file or a tag. The score is points earned over the 115 points possible, scaled to 100.
| Signal | Points | Share | What is measured |
|---|---|---|---|
| AI assistant access | 35 | 30% | robots.txt parsed for the 10 crawlers listed below; credit per crawler allowed. No robots.txt is permissive and scores full marks. |
| Structured data | 20 | 17% | An Organization, WebSite or similar identity block in JSON-LD earns 12; up to two further types (Product, Service, FAQPage, BreadcrumbList and the like) earn 4 each. |
| Answerability | 20 | 17% | Homepage word count measured up to 600 (12 points) and H2 headings measured up to 4 (8 points). |
| How the homepage introduces itself | 15 | 13% | A title of at least 10 characters, a meta description of at least 50, and an H1. 5 points each. |
| Your keyword on the homepage | 15 | 13% | Full credit if the phrase typed appears in the title, description, H1 or body; half if only its individual words do; none otherwise. |
| llms.txt | 10 | 9% | Present or absent at /llms.txt. |
Structured data and answerability are measured and counted in the score; their detail is shown in the paid report under Current Position and What to Build, not in the free result. The score is therefore a score of the whole site. Buying the report does not change it.
Two consequences. A site that blocks all 10 crawlers cannot score above about 70, whatever else it does. And because there is no model, the same site scores the same number on every run until something on it changes, which is why we publish our own result at /audit/viscovery and re-run it daily.
Crawlers checked
- OAI-SearchBot — ChatGPT search results.
- ChatGPT-User — ChatGPT browsing your page on request.
- Claude-SearchBot — Claude's search results.
- Claude-User — Claude reading your page on request.
- GPTBot — OpenAI's model knowledge of your brand.
- ClaudeBot — Claude's model knowledge of your brand.
- PerplexityBot — Perplexity answers and citations.
- Google-Extended — Gemini's training and its Search-grounded answers.
- CCBot — Common Crawl's open web archive.
- Applebot-Extended — Apple's model knowledge of your brand.
Our crawler
The free checker identifies itself as ViscoveryPreviewBot/1.0. When someone checks a domain it makes one request each for /robots.txt, /llms.txt and the homepage. It follows redirects but not links, so it never crawls the rest of the site. It reads at most the first 1.5 MB of a page, stops waiting after 8 seconds, and only connects to public internet addresses. The same three requests produce our own result at /audit/viscovery and the data in our published studies. Questions about it: support@viscovery.ai.
The report’s five health scores
A paid report opens with five 0–100 scores, each flagged Strong, At risk or Critical:
- Search health
- AI visibility
- Page experience
- Security posture
- Crawl foundation
An audit agent produces them from: a crawl of your pages; live Google results and keyword data for the keywords you bought; the answers ChatGPT, Claude, Gemini and Perplexity give when asked your primary keywords (the top one, two or three depending on plan), with the sources they cite, plus a ChatGPT-only pass over the remaining keywords; a crawl of the competitor pages those results surface; lab and field page-speed measurements; and a screenshot capture of the site. Every tool call the agent makes is logged against the job, so a number in a report can be traced to the data that produced it.
The report uses fixed section headings: Decision Brief, Current Position, Evidence, What to Build, 90-Day Roadmap and Measurement. Growing Brands and Market Leaders add Keyword Portfolio and Competitive Landscape; Market Leaders adds Content Architecture and Execution Plan. The headings are checked by the delivery gate; a report missing one is not delivered.
The five signals the findings are written through
The homepage names Trust, Authority, Relevance, Consistency and Citations. They are not a scoring formula. They are the lens each finding is written through, in a fixed order: what was found, why it matters, the evidence, the action. A reader who disagrees with a conclusion can see the data under it.
What the numbers do not claim
- The free score measures readiness: can assistants read and understand the site. It does not measure how often they recommend it. Two sites at 90 can be recommended very differently.
- The report records what the assistants and the search results returned when it ran. AI answers vary between runs and between users. The report states what it observed and when.
- A finding the free engine did not measure carries zero points, not an invented weight.
Check it yourself
Run the free checker on any site, then open that site’s /robots.txt, /llms.txt and homepage source. Each unlocked finding names the file, rule or tag it came from. Our own result is at /audit/viscovery; three complete reports are at /sample.