How we measure, and where it stops.
If you show a client an AI visibility number, you should be able to explain how it was made and what it does not mean. This page explains that, including the parts that do not make us look good.
Five words that mean different things.
Most mix-ups about AI search numbers come from not knowing which of these words someone means.
- Mention
- The client’s brand name shows up anywhere in an AI answer we collected. It counts even if it is not linked or recommended.
- Position
- Where the client comes in the list of brands named: first, second, third or later. Being named last out of six is not the same as being named first.
- Citation
- The AI linked to a page on the client’s own website as a source for its answer. Being linked as a source is much stronger than just being named. We always report the two apart.
- Visibility
- Out of the answers we collected over a set stretch of time, the share that named the client. It is a percentage. We always show how many answers it is based on.
- Share of voice
- How often the client is named compared with its rivals. It is the client’s mentions out of all mentions of the rivals you named, in the same set of answers.
No one can look up AI answers. So we ask.
A rank tracker can read a search results page. No one can do that with AI answers. The only honest way to measure is to ask again and again, and count.
What we check
A fixed list of prompts (questions we ask AI) for each client. You pick them with us, from the questions that really decide who buyers choose. It is not a keyword list that grows on its own.
How often
Each client's prompts run on the schedule you set, daily or weekly, on ChatGPT, Google AI Overviews and AI Mode, and Perplexity. Gemini and Claude are in beta and checked less often for now. Microsoft Copilot is not tracked yet. Website scans run when you start one.
Where from
We ask each prompt as if from a set place, because answers change by region. The region is saved with the results. It is not a hidden setting that quietly changes the numbers.
What we save
The answer text, the brands named, their order, the sources linked, the AI tool, the model version when it is shown, and the date and time. Old records are kept as they were.
- ChatGPT+14.2%72%
- Google AI Overviews+8.6%61%
- GeminiBeta+11.4%49%
- Perplexity−3.2%43%
- ClaudeBeta+5.9%37%
What we will not claim.
In each of these, a rival tool could quote a bigger number than we will. We would rather lose that contest than have a hard talk with your client six months later.
01One check proves nothing
AI answers change between users, visits, regions and model versions. One check where a client is missing does not prove a problem. The same prompt missing them across two weeks of checks does. That is why we always report over a stretch of time.
02Good for trends, rough as an exact number
Read a visibility score of 68% as ‘named in most answers, again and again’. It is not an exact measure of the real world. When we check the same way over time, the direction the number moves is the part you can trust. That is the part worth showing a client.
03What we can see differs by AI tool, and changes
Each AI company shows a different amount. Some link every source, and some link none. Some block automated tools completely. When an AI tool limits what we can see, the screen says so. We do not fill the gap with a guess.
04Happening together does not prove cause
When a score moves after your team finishes some work, we show both on the same timeline. We do not claim the work caused the change. A rival’s launch or an AI update in the same two weeks could explain the data just as well.
05We do not promise rankings
The AI tool decides whether to link to a client as a source. So each fix is about making a client able to be linked: the conditions that let the AI cite them. That is the part anyone can really change.
The ground moves. The history should not.
An AI update can change a client’s visibility overnight, even if nothing about the client changed. Telling that apart from real change is most of what measuring is for.
- Each stretch of time we measure is labeled by version. So numbers from before and after a change are never quietly averaged together.
- We add a note to the timeline when an AI tool suddenly changes how it acts across all our clients, not just one.
- Old records are kept as they were and never redone later. A report sent in March still says the same thing in September.
Find out which of your clients AI is skipping.
Pick one client. Within 24 hours, we email you a free snapshot of how AI sees them. No call and no card needed.
Name one client
Their website, and where to send the result.
We ask the AI tools
12 questions their customers ask, on three AI tools.
Email within 24 hours
Where they stand, plus the first three fixes.