Score composition
The overall score is a weighted sum of six dimensions. Each dimension is scored from 0 to 100 from evidence captured during that scan.
- Offer understanding, 20%: category, audience, use cases, and differentiation are explicit.
- Recommendation fit, 20%: the offer can be matched to relevant buyer questions without guessing.
- Evidence and trust, 20%: important claims have specific, accessible support.
- Answer-ready content, 15%: useful pages directly answer comparison and purchase questions.
- Technical access, 15%: public pages, metadata, schema, robots rules, and sitemap signals can be read.
- Distinctiveness, 10%: the site provides concrete reasons to choose the offer over alternatives.
Recommendation probes
AnswerCue creates three category-relevant questions from the supplied context, records the selected model and time, then checks whether the target appears and how the answer frames it. A grounded probe uses web search only when the provider and configuration support it.
Evidence rules
The report may use only the crawled public pages, the supplied product context, and the recorded probe output. Missing information must be marked as missing. The analysis prompt prohibits invented customers, metrics, certifications, pricing, or model citations.
Owned evidence and external corroboration
A product website can explain its own category, fit, features, and use cases, but independent sources may be needed to corroborate reputation or comparative claims. AnswerCue keeps those roles separate: a probe can reveal which outside sources shaped an answer, while the website scan evaluates the evidence the team directly controls.
What the score cannot show
AI answers vary by model, prompt, location, personalisation, retrieval source, and time. A probe is an observation, not a traffic forecast or inclusion guarantee. Stronger evidence can improve eligibility and accuracy, but no website optimisation can force a model to recommend a product.
Search, retrieval, and model training
Traditional indexing, live AI retrieval, and model training are different processes. AnswerCue focuses on public, indexable evidence that can help search and retrieval systems understand an offer. It does not claim that an llms.txt file or training-bot access creates rankings.