The method behind
the number.
The home page makes the case. This page is the work. Here is exactly what we measure, how the measurement actually runs, what we change to move it, and why every result sits on the record instead of being a claim.
Watch the method before you read it.
Two numbers, defined precisely, and always shown together.
Everything we do resolves to two published scores on a 0 to 100 scale. We never report one without the other, because being named often but described badly, and being described well but named almost never, are different problems with different fixes.
AIVI, the AI Visibility Index
AIVI is how often and how prominently the models actually name you across the real buyer prompts in your category. Frequency is how many of those prompts surface you at all. Prominence is where you land when they do: named first and unprompted counts far more than a brief mention after the buyer already asked for you by name. We fold both into one 0 to 100 reading each cycle.
AIRS, the AI Reputation Score
AIRS is how favorably the models describe you when they do name you, from 0 to 100. It reads the sentiment, the claims they repeat about you, and whether the framing is a recommendation or a warning. Until we have at least eight scored observations, AIRS reads null, never zero, so a thin sample is never dressed up as a bad reputation. We print AIRS beside AIVI every single time.
A fixed ruler, run the same way every month.
A measurement is only trustworthy if it holds still while the result moves. So the prompts and the engines are fixed, and only your position inside them is allowed to change.
We build the real buyer questions
We assemble the questions a buyer in your category actually asks an AI: the best option requests, the head-to-head comparisons, the is it worth it checks, across the personas who buy from you. That fixed set becomes the ruler, not a sample we reshuffle each time.
We put them to four engines
Every prompt goes to the four frontier engines, ChatGPT, Claude, Gemini and Grok, and we capture each answer exactly as written, verbatim, along with the sources the engine leaned on to compose it.
We read every answer and score it
For each captured answer we record whether you were named, how high you placed, and the exact words used about you. That is what rolls up into your AIVI and your AIRS for the cycle.
Then we do it again next month, same prompts, same engines. Because the ruler never changes, the before and the after are directly comparable. A move in the number is a real move, not the result of quietly asking a different question.
We engineer the inputs, never the model.
A model does not invent its recommendation. It synthesizes the sources it already trusts, then names whoever those sources make the obvious answer. So we work on the sources, in the open, with nothing hidden from the engines and nothing done to the model itself.
In practice that means writing content shaped like the answer a model is trying to assemble, earning your presence in the directories, datasets and references the engines treat as ground truth, cleaning the structured data that tells them what you are, fixing how their crawlers actually read your site, and working the review and AI Overview signals that feed how you get summarized. Each one is a real input a model weighs when it decides who to name.
Six agents, each on one input.
The work above is split across six agents, so every input a model weighs has something accountable to it, and each is measured back on AIVI and AIRS.
Owns how models describe you
Sets the language, positioning and facts you want repeated, aligns them everywhere a model can read them, and tracks the result on AIRS so a change in wording shows up as a change in the score.
Writes what engines pull from
Produces the question-led, answer-shaped pages a model lifts from when it builds a recommendation, mapped to the exact prompts in your set rather than to keywords.
Places you in ground truth
Gets you into the directories, datasets and references the engines treat as fact, because a claim a model finds in a source it trusts outweighs the same claim on your own site.
Builds the standing models reward
Earns the third-party coverage and references that make an engine treat you as a safe answer to recommend, not a name it is unsure about.
Fixes how your site is read
Checks how each engine crawls and interprets your pages, finds what they misread or cannot reach, and clears it so the right facts are actually ingestible.
Works reviews and overviews
Manages the review signals and the AI Overview surfaces that feed how models summarize your reputation, so the sentiment they repeat matches reality.
Every result traces back to a recorded answer.
None of this is self-reported. Every answer an engine gives us is captured verbatim, signed and time-stamped the moment we take it, and kept. Each AIVI and AIRS reading points back to the exact answers it was built from, so you can open any score and read the raw text behind it in your dashboard.
That is what makes a lift auditable instead of a claim. When the number moves, you see the before answer and the after answer side by side, in the models' own words. And payment cannot touch it: the only thing that moves AIVI or AIRS is what the engines actually say.
See the method run on your own category.
Your free report runs the measurement against your business and hands you the first reading. Book a demo to walk through it and the plan to move it.