Methodology
How we verify what AI says.
GEOlens is diagnosis-first. We capture what engines actually return, verify citation behavior, then turn those receipts into prioritized gap cards with first fixes. Scores stay visible, but evidence is the primary artifact.
How receipts are captured
For each scan, GEOlens probes ChatGPT, Claude, Perplexity, and Gemini using brand recall, category placement, and citation-focused prompts. Every diagnosis card is grounded in a query, answer excerpt, citation outcome, and capture timestamp.
- Engine receipts are sampled live per scan run (not replayed from archives).
- Citation checks distinguish discovery gaps from trust/source gaps.
- Findings are normalized into plain categories: missed, misread, or invisible.
What we score (secondary layer)
GEOlens still reports two top-line scores for orientation:
- SEO from hosted Lighthouse via Google PageSpeed Insights.
- AEO from Engine Visibility, AEO Hygiene, and Citability signals.
We present gaps first because they are easier to act on than aggregate numbers.
Scoring details (open-book)
Full weights and formulas remain transparent for teams that need the math.
Expand exact weights and multiplier formula
- SEO = Performance 25% · Accessibility 25% · Best Practices 20% · SEO 30%
- AEO = Engine Visibility 60% · AEO Hygiene 25% · Citability 15%
probe_score = base × position × sentiment × accuracy position: primary 1.0 · secondary 0.7 · tertiary 0.4 · none 0.0 sentiment: positive 1.2 · neutral 1.0 · negative 0.5 accuracy: accurate 1.0 · partial 0.7 · misattrib 0.3
What we don't do
- Live Google AI Overviews probes (deferred; requires SerpAPI-class data)
- 200M-prompt search-backed databases (deferred; that's incumbent moat)
- Continuous monitoring (one-shot reviewer for v1)
- Persist raw HTML beyond the scan (only computed signals; PRD §13)