This page is the methodology statement: what's measured, what's estimated, what's AI judgement, and the rules that keep the three from being confused. It exists so you can defend any number in a report to a client — and so you know which numbers to defend differently.
Where the numbers come from
Measured. Crawls read your actual pages, live. Rank positions, AI Overview presence, brand-mention counts, and search volumes come from live search-market data providers at run time. When the report says your page lacks a meta description, a fetch of that page found none.
Estimated. Traffic figures are modelled from rankings and volumes — industry-standard practice, honest label. Difficulty summarizes the strength of what currently ranks. Estimates are directionally reliable and exactly wrong — read trends, not decimals.
AI judgement. Article generation, audit recommendations, keyword proposals, competitor verification, and content scoring use AI models. The product constrains them with measured inputs: writers link only to real crawled pages, keyword proposals derive from your site's actual content, and the content score's checks are defined rules, not vibes. Where judgement could fail, a deterministic rule backstops it — link counts are enforced in code, and score gates use the same engine that scores your own pages.
The honesty rules
Each of these shapes what you see:
- Nothing is fabricated to fill a gap. A section without data says so.
- Gated features are labelled gated, and sample data is labelled sample — the per-assistant AI tiles being the standing example.
- A zero that's true stays a zero: no AI Overview for your query means exactly that.
- "Published" means the public URL was fetched and found — not that a platform accepted a write.
Limits and usage
- Search data reflects provider snapshots; brand-new trends can lag.
- Crawls sample very large sites rather than claiming exhaustiveness, and reports show the crawl count so coverage is visible.
- AI judgements vary slightly between runs by nature; the measured inputs and rules don't.
If something isn't working
Two tools disagree on a number. Check what kind each number is — a measured position and an estimated traffic figure answer different questions and move on different schedules.
You need the source for a specific claim. Every report section's help page names where its data comes from — start at the audit report guide →.