Methodology
Exactly how an Avanti scan measures AI visibility — so you can defend every number in the report.
1 · What we measure
When a buyer asks ChatGPT or Claude for a recommendation, either your brand is in the answer or a competitor's is. A scan sends a controlled panel of real buyer questions to the AI engines and records, verbatim, what comes back. Every score in the report is computed from those raw responses — all of which you can read, unedited, in the Response Explorer.
2 · How the questions are built
You pick a department (e.g. "Automotive accessories"); nobody shops by department, so we first resolve your brand to the category a buyer actually types — e.g. Vantrue → "dash cam", Cal.com → "scheduling software". The category we measured is printed at the top of every report as “Measured as”. Questions are then drawn from three intent tiers: High Intent (ready to buy, weighted 1.5×), Comparison (weighing options, 1.2×), and Informational (researching, 1.0×). Consumer categories additionally sample generational phrasings (Gen Z social-driven queries through Boomer trust-driven queries). No question ever names your brand — the AI must bring you up on its own.
3 · Engines and versions
Free scans query OpenAI (gpt-4o-mini) and Anthropic Claude (claude-fable-5) with neutral system prompts — the engine answers exactly as it would answer a real user, and is never told a measurement is happening. Paid monitoring adds Gemini and Perplexity. Every run is date-stamped (top of the report); AI answers drift over time, which is precisely why visibility is worth monitoring.
4 · Mention detection
Latin-script brand names must appear as whole words ("Sudowrite" never matches the word "write"). Chinese/Japanese/Korean names match as substrings, since those scripts don't delimit words. Known aliases (绿联 ↔ UGREEN) count as the same brand.
5 · The scores
AI Visibility Score = your intent-weighted share of voice (0–100) among the brands tracked in this scan — it is relative to the competitors you listed, not an absolute market number. Mentions = the count of AI answers naming you, out of all answers sampled. AI Recommendation Risk = how far you trail the strongest competitor per intent tier, weighted (0 = no gap; requires competitors to mean anything). Sentiment is a directional keyword-window signal — treat it as a hint, not a verdict.
6 · Sampling honesty
A free scan is 15 questions × 2 engines = 30 answers. That is enough to see structure (who dominates, where your gaps are) but individual scores can move roughly ±8 points between runs — AI engines are not deterministic. Paid monitoring runs larger panels on a schedule and reports trends, which is where the signal gets sharp. If a number surprises you, run it again before you act on it.
7 · Fair use
Anonymous: 3 free scans per day. Free account: 8 per day. Paid plans: unlimited. Each scan costs us real inference — the caps keep the free tier alive.