How we measure AI visibility
Most “AI visibility” tools ask the model a leading question — “How good is BrandX?” — and then count the brand being echoed back. That measures nothing. FoundGeo uses a blind-query method: the model never sees your name, so a mention is a real recommendation, not a reflection of our own prompt.
We read your site, then write buyer questions — without your name
We fetch your homepage and identify the category you actually compete in. Then a research model writes 12 buyer-intent queries a real customer would type into an AI assistant — covering recommendations (“best X for…”), problem-first (“how do I…”), alternatives to a known leader, and comparison/buying criteria.
Every query is validated to guarantee it contains neither your brand name nor your domain. Any query that leaks either is dropped. If we can't read your site, we stop — we never let a model fill the gap with guesses.
We ask 6 real AI models, 3 times each
Each query runs across ChatGPT, Claude, Gemini, Perplexity, Grok and DeepSeek — three independent times per model at a natural temperature, so we capture variance instead of a single lucky (or unlucky) answer. That's up to 216 real completions per scan.
We deliberately excluded low-traffic models (Qwen, Mistral, Llama). They carry negligible real search share and only inflate scores with easy mentions.
We score with code, not with another AI's opinion
Mention = your brand appears via a word-boundary, case-insensitive match. Position = we order every known brand by where it first appears in the answer and read off your rank. Share of Voice — our headline metric — is your mentions divided by all brand mentions in the same answers: the most honest number of how often AI picks you over the field.
If you're never mentioned, Average Position shows “—”, never a fake 1 or 0. Sentiment is measured only on the sentences that actually name you; if you're absent, it's “N/A”.
We verify every citation — and expose the hallucinated ones
AI models routinely invent source URLs (a review page that returns 404, a profile that never existed). We send a live request to every cited URL and only show links that actually resolve. The fabricated ones are counted and surfaced as Hallucinated Citations— a risk signal you won't find anywhere else.
Why your score might be low — and why that's the point
Because we never feed the model your name, a brand that AI genuinely doesn't know yet will score near zero. That's the honest starting line. Every point you gain from here is a real customer conversation you started winning.
Run an honest scan →