What Reffed measures — and what it does not.
Reffed is an AI-search readiness audit. We separate evidence we can observe on your website from model-generated estimates and from live provider responses. A readiness score is not a promise that ChatGPT, Claude, Gemini, or any other engine currently recommends your business.
Three labels. No mixing evidence.
Every meaningful result should fit one of these evidence classes. The distinction matters because a website signal, an inference, and a live model response answer different questions.
The audit starts with your website.
Crawler instructions
Reffed reviews robots.txt and distinguishes search/user-fetch crawlers from training controls. A rule in robots.txt is evidence of an instruction, not proof that a crawler has successfully visited the page.
Entity clarity
We look for clear business identity: what the organization is, what it offers, where it operates, and whether structured data helps machines resolve that identity.
Page structure
Titles, descriptions, headings, canonical tags, structured data and answer-friendly formatting are inspected for clarity and machine readability.
Content evidence
Reffed evaluates whether the inspected page contains enough specific, useful information to support an answer. Thin or ambiguous content is reported as a readiness gap.
Engine-oriented views
The same observed signals are organized into six engine-oriented readiness views for ChatGPT, Claude, Perplexity, Google AI Overviews, Gemini and Copilot. They are heuristics, not six live ranking checks.
Prioritized actions
The report converts detected gaps into a short list of recommended next actions. Priority reflects the audit model's assessment of likely value and effort; it is not a guaranteed traffic or ranking outcome.
A score is a readiness benchmark.
The score is designed to help one site compare its own readiness over time. It is not a universal market ranking and should not be interpreted as the percentage chance that an AI engine will cite the business.
Useful interpretation: “Our site is technically clearer and more complete than it was last month.”
Not supported: “We have a 78% chance of being recommended by ChatGPT.”
Also not supported: “A 10-point Reffed increase caused a 10-point increase in AI visibility.” Repeated measurements can show an association after a change; causation requires stronger evidence.
Unknown stays unknown.
A trustworthy audit should fail visibly. If a site cannot be read, times out, blocks requests, or depends on rendering Reffed cannot complete, we record an incomplete state instead of silently converting missing evidence into a zero or a positive result.
Fetched and measured
Signals were available from the inspected page or supporting files and could be evaluated.
Estimated
The conclusion is an inference based on observed evidence rather than a direct provider response.
Incomplete or unknown
Evidence could not be obtained reliably. The report should say so rather than fill the gap with certainty.
The methodology is versioned.
AI products and crawler behavior change. Reffed therefore treats the scoring methodology as versioned product logic, not an eternal definition of “AI visibility.” Material scoring changes should update the methodology version and be reflected in future reports.