Methodology
How to measure what AI says about a business
Ask an AI assistant for the best dentist in your city. You get a short list of names. Maybe four. Maybe two. If a business is missing from that list, that buyer never hears of it.
Search rankings never had this problem, because there was always a second page of results for a patient or a landlord to scroll through before deciding who to call. AI answers have no second page. So the old metrics stop working, and two published indices were built to replace them: AIVI and AIRS.
Both come from research by Gokul Mohanty, co-founder of CTRL Reality, Inc. and first author of the five preprints listed at the bottom of this page. This guide explains each index, shows the arithmetic, and lets you run the numbers yourself.
AIVI: does the AI name you, and how early?
The AI Visibility Index answers one question. Across the questions real buyers ask, how often does an AI assistant name this business, and where in the answer does the name land?
Coverage is the share of questions where the business shows up at all. Ask twenty questions, appear in five, and coverage is a quarter of the way to 100.
Rank is about position. People read the first name and skim the rest, so each answer gets points for where the business appears:
| Where the business appears | Rank points |
|---|---|
| Named first | 100 |
| Named second or third | 75 |
| Named fourth to sixth | 50 |
| Named after sixth | 25 |
| Not named | 0 |
The average rank score is the mean of those points across every question, including the misses. That matters. A business named first once and absent everywhere else gets a low average, as it should.
AIVI is never read alone. Its partner, AIRS, comes next.
A worked example
Take a made-up practice and twenty buyer questions. It gets named in five answers:
| Outcome | Answers | Rank points |
|---|---|---|
| Named first | 1 | 100 |
| Named second or third | 2 | 150 |
| Named fifth | 2 | 100 |
| Not named | 15 | 0 |
Coverage is 5 of 20, so 25. The rank points add up to 350, spread over 20 questions, so the average rank score is 17.5.
AIVI = 0.6 × 25 + 0.4 × 17.5 = 15 + 7 = 22.
Low. Fixable, too. Showing up in more answers moves the score faster than climbing one slot in answers it already wins, because coverage carries the bigger weight.
Five named answers is also too few for a reputation read. Its AIRS stays blank for now. Why is explained below.
AIRS: when the AI names you, what does it say?
Getting named is half of it. An assistant can name a business and then call it overpriced. The AI Reputation Score looks only at the answers where the business appears and asks two things. How warm is the language? How credible are the sources behind it?
Sentiment strength
Each mention is read as positive (100 points), neutral (30) or negative (0). Words like "best" or "highly rated" count as positive. A plain factual listing is neutral.
Authority confidence
Each answer is scored by the strongest source signal behind it:
| Source behind the answer | Points |
|---|---|
| Major publication or established media | 100 |
| Aggregated reference platform, such as an encyclopedia | 75 |
| First-party or user-generated: the company site, reviews, social posts | 10 |
| No identifiable source | 0 |
Notice the gap between the top two rows and the third. A business described only by its own website and its own reviews tops out low on authority, however glowing the words are. Independent coverage changes the picture.
Why an AIRS can be blank
Reputation needs evidence. On the CTRL REALITY platform, AIRS is reported only once a business has been named in at least 8 answers, and below that line it shows as unmeasured, never as zero.
Blank is a finding. It means the assistants barely mention the business, so there is too little said about it to judge. A zero would claim something else entirely: that the AI speaks badly of it. Different problem. Different fix.
That is also why the two numbers always travel together. AIVI tells you whether you are seen. AIRS tells you how you come across once you are, and if you read only one of them you can end up celebrating a loud complaint thread as visibility.
How CTRL REALITY runs the measurement
The formulas are public. Consistency is the hard part.
- Fixed questions. Every business in a category is measured on the same written prompt set, phrased the way buyers phrase things.
- Four engines. ChatGPT, Claude, Gemini, Grok, each with web search switched on, so answers reflect what the engine can find today.
- Questions that name the business don't count toward AIVI. Asking "Is Acme any good?" guarantees Acme appears. Those answers are reported separately as a direct-query check.
- Repeated on a schedule. The same prompts run again each cycle, so a change in score reflects a change in the answers.
The scoring rules for AIVI and AIRS come from the papers. The prompt sets, the schedule and the blank-AIRS threshold are CTRL's own conventions, built on top of that research so that one business can be compared with another and with its own history.
Try it on your own business
Pick ten to twenty questions a customer would ask, in their words, about the thing you sell and the place you sell it. Ask them in one assistant. Count where your name lands. Then fill this in.
For AIRS, take only the answers that named you. Mark each one's tone and its strongest source.
The calculator runs in your browser. Nothing you type is sent anywhere.
The research
Gokul Mohanty is first author of the five preprints that define these frameworks. Cite them as Mohanty (2026).
- Artificial Intelligence Visibility Index (AIVI): The Framework to Quantify Entity Presence in Generative Information Engines. doi:10.5281/zenodo.19153037
- Artificial Intelligence Reputation Score (AIRS): The Framework to Quantify Entity Sentiment in Generative Information Engines. doi:10.5281/zenodo.19122428
- Generative Information Engine: Generative Engine Optimization and the Next Layer of Global Commerce and Decision Making. doi:10.5281/zenodo.19107148
- Distributed Authority Domain Placement and How to Influence Artificial Intelligence. doi:10.5281/zenodo.19122809
- GEO: Influencing Commerce, Human Decision Making, and Global Policy Through Generative Information Engines. doi:10.5281/zenodo.19154674
Want it measured properly?
Doing this by hand once is useful. Doing it across four engines, every cycle, against the names that keep beating you, is the job CTRL REALITY does.
Industry guides: AI visibility for property managers · AI visibility for dental practices