methodology — how we measure AI search visibility

this page documents exactly how the vøiddo geo audit works. transparency matters, because most "AI visibility" tools obscure their method.

one audit, five queries, real AI search

when you submit a domain we do three things, in order:

1. brand profile. an LLM looks at the domain and infers (a) the actual brand name as a buyer would type it, (b) the tight product category, (c) the top four competitors that an actual buyer in that category would compare against, and (d) five real buyer-intent queries an actual buyer would ask an AI assistant. queries are buyer-intent ("best X for Y under $Z"), not brand-name lookups ("is X a good company"). brand-name lookups are useless for visibility scoring — every brand "appears" when you ask about itself.

1. AI search. each query is sent to Gemini with live Google search grounding. grounding is the same mechanism Google's AI Overviews use to ground answers in real web pages. the assistant returns a natural-language answer plus the list of sources it actually used.

1. citation detection. we check whether the answer contains a literal mention of your brand name and each competitor name. the brand cited in the most queries wins. we report a 0–100 score (percentage of queries where you were named) and a per-query breakdown so you see exactly which queries you are losing.

why Gemini + grounding, not raw GPT-4 or Perplexity

on the free tier we use Gemini with Google grounding because it is the cheapest provider of AI-search-with-citations, which lets us offer the tool free without a paywall in front. paid plans add ChatGPT (gpt-4o), Perplexity, and a Google AI Overviews scraper for cross-provider comparison. for most categories, the four providers agree on which brands are visible to within a percentage point, because they pull from heavily overlapping document sets.

what the score actually represents

a 60/100 score means: of the five buyer queries your real customers might ask, you appeared in three of the five answers. a 0/100 means you are invisible. a 100/100 means you are cited in every buyer query — congrats, but also: re-run the audit on the cohort of high-intent queries you'd lose first if you slipped (we surface those in the report).

what we deliberately don't measure

• traffic to your site from AI assistants. nobody reliably measures this yet; analytics referrer headers are stripped or spoofed. anyone who claims to measure "AI search traffic" with high confidence is mostly guessing.
• backlinks. those are a SEO signal. some of them still matter for AEO, but they are not the metric.
• "AI sentiment." sentiment is mostly noise; what matters is whether you got the citation, not whether the surrounding sentence had a positive adjective.

known limitations of the free tier

• only five queries per audit. paid plans run the full 50-query set across providers daily.
• single snapshot, no historical tracking. paid plans track week-over-week movement.
• brand-name match is case-insensitive substring. if your brand name is a common English word ("Apple", "Square") we may overcount. we'll fix that for the categories where it bites.
• we don't yet handle multi-brand parents (e.g. Estée Lauder companies). roadmap.

what to do with the result

if your score is below 50: you have a real visibility gap. the queries you are losing are usually category-defining buyer questions where competitors have shipped better first-party content. the fix is content + earned citations, not paid acquisition.

if your score is above 70: focus on the high-intent comparison queries you are still losing. these are usually the most valuable five percent of the funnel and the easiest to win.

re-run weekly. AI assistants update their training and grounding indexes faster than you'd think, so a fix you ship on monday can show up in citations by friday.