Models via API
The response obtained for a question, with provider, model, date, and the panel's country/language.
It does not reproduce personalization, memory, session, or the consumer app interface.
Public methodology · version 2
What SeoWave measures, and what it doesn't.
SeoWave runs buying questions against model APIs, records the technical context of each response, and separates observed visibility from website readiness.
Updated September 13, 2026
Measurement surfaces
An API response is not an exact reproduction of a consumer app. That is why we disclose the surface and search mode.
Web search
Grounding when the engine or gateway supports it; Perplexity uses intrinsic search.
It does not mean every source or consumer interface will return the same answer.
Google Search
Google results and an AI Overview are labeled only when the source returns them explicitly.
Grounded Gemini is not presented as Google AI Overview or AI Mode.
Website readiness
Technical, content, structure, entity, trust, and citability signals from the site.
It does not add observed-visibility points when the brand is absent.
AI Visibility Score (observed)
- The headline uses brand-blind questions: how many responses mention the brand across the complete panel.
- Questions that name the brand are shown as aided but do not inflate the headline.
- Only a citation to the brand's own domain improves citation quality; a competitor source does not count.
- AI Readiness is separate: it explains whether the site is prepared, not whether it is already visible.
Coverage and comparability
- We do not publish a score, delta, PDF, or alert when a question × engine cell is missing.
- A delta is comparable only when method version, question set, and engine configuration match.
- Each measurement preserves model, search mode, date, coverage, and available structured sources.
- Model responses vary. A repeated trend is more useful than an isolated observation.
A score is an estimate over a finite panel of questions × engines, not an exact truth. We use its margin of error to decide which changes to show and we do not present sampling variation as a trend.
Measurement precision and noise
- The score is computed over a finite set of answers, so for every measurement we compute a 95% confidence interval (Wilson score interval). We use it to decide whether a change is real; the dashboard shows the result of that comparison, not the interval itself.
- The variation between two measurements is compared against their COMBINED margin: sqrt(se1² + se2²), not a single measurement's.
- When the change is smaller than that margin, the interface labels it "within the measurement margin" instead of showing it as a rise or a fall.
- With a weekly sample of 10 questions × 4 engines (40 answers), the panel cannot resolve changes under ~15 points. Any tool reporting 2-3 point moves as a trend is measuring noise.
- If a measurement has too few comparable answers, we say it is not enough to compare instead of inventing a delta.
Limits you should know
SeoWave measures a controlled, repeatable panel; it cannot observe every user's private personalization. Providers may change models, policies, or availability. When the panel changes, we start a new series instead of drawing a misleading delta.