How Is the AI Visibility Index Calculated?
The short version
Think of it as a poll — but the respondents are AI engines. We ask ChatGPT, Claude, Gemini, Grok, and Perplexity things like "who should I call for an emergency pipe burst?" and record who they name. Do that 380 times across 38 questions and two passes, and patterns emerge: in Edition 01 (measured 2026-07-09), Roto-Rooter took 10.0% of all 1,103 brand mentions — the most AI-visible brand in its category. Full detail: methodology; full data: the live index.
Mention share vs. AIR Score
- Mention share (the index rankings): a brand's percentage of all logged mentions in a run. Category-level, comparative, built from the national brand list.
- AIR Score (individual businesses): a 0-100 composite of mention, recommendation, and sentiment across the five engines — measured per business by the free scan. Benchmark: the published Houston metro study found a median of 31/100.
Both come from the same pipeline; they answer different questions — "who owns the category?" versus "how visible is MY business?"
Why dated editions matter
AI engines re-crawl, re-ground, and shift their answers constantly — pages untouched for ~90 days measurably lose citations (iQDigital, 2026). So the index never claims real-time truth: every number carries its run date, editions supersede each other, and the honest citation format is always "as of [date]." That discipline is also why the data stays citable by the same AI engines it measures.
Frequently asked questions
Is this like a search ranking?
Closest analogy, but for AI answers instead of links: rather than tracking your position on a results page, it measures whether AI engines name and recommend you when customers ask. There's no position #3 in a ChatGPT answer — you're named or you're absent, which is why mention and recommendation rates are the units.
Why do you run each prompt twice?
Because AI engines give varying answers to identical prompts. Two passes per prompt per engine (380 calls total in Edition 01) reduce the noise of any single response while keeping runs affordable and reproducible. The variance itself is part of the finding — visibility that survives both passes is robust.
What does the Houston 31/100 median mean for me?
It's the calibration point: in the published Houston metro study, half of measured businesses scored 31 or below — barely visible to AI. If your scan lands above 50, you're well ahead of a typical metro business; below 31, you're behind even the median — and the causes are almost always the five mechanical, fixable gates.
Do engines know they're being measured?
No — the harness asks the same questions any customer could type, through the same interfaces, and logs what comes back. No special access, no partnerships with the engine vendors, no gaming. That's what keeps the measurement honest and repeatable by anyone who doubts it.
How do I get my own number?
Run the free scan at deepaivisibility.com — about a minute, five engines, AIR Score baseline with the failing gates flagged. It's free because measured baselines are the best argument for the paid engineering; you keep the number either way.