How an AI visibility scan works
AIV Report answers one question: when a potential customer asks an AI assistant who to buy from, does it name you? This page is the full account of how we get that answer.
Buyer-discovery prompts, not vanity questions
Each scan generates 10 prompts a real buyer would ask — things like “who should I hire for same-day appliance repair in Austin”, not “what is Acme Inc.”. We crawl your website first and build a profile (what you do, where, who your peers are), you confirm it at the checkpoint, and the prompts are written in your site’s language and anchored to your market, so a Portuguese business is compared with Portuguese peers, not Brazilian banks.
Four models, grounded with live web data
The prompts go to ChatGPT, Claude, Gemini and Perplexity — the assistants consumers use to research purchases today. Every call is grounded with live web search, matching how the consumer versions of these tools actually behave: an assistant that can look things up answers very differently from one working from memory.
Deterministic scoring — no LLM guesswork
From each answer we extract the list of recommended businesses, and your position in it is computed from the text, never estimated by a model. Scores weight being recommended first above being mentioned last, and being named at all above polish: for a small business, presence beats sentiment. The analysis and recommendations in the report are rule-based and fully traceable — every score can be audited against the raw prompts and excerpts shown right below it.
Ranked against peers, not the Yellow Pages
AI answers mix real competitors with adjacent entities — brokers, newspapers, regulators, platforms. Each name mentioned across all answers is classified (peer, platform, media, institution), and the leaderboard ranks peers only. Adjacent names still appear, badged as context, because they shape the answer your customer reads.
Limitations, honestly
- A report is a snapshot: AI answers shift as models and their sources change. That’s why a fresh scan opens 7 days after the last one.
- Answers vary run to run; one scan is a strong signal, not a census.
- Grounded API calls approximate consumer chat closely, but no third party can replay a consumer session exactly.
- A low score is a starting point, not a verdict — the report ends with concrete, deterministic recommendations for what to fix first.