Why is a number about AI visibility usually not enough?
You have a number. It moves. Whether it moved because you did something or because a model answered differently that week is written nowhere — and that is the one thing you wanted to know.
What does each layer do, and what does it not do?
The layers sit in the order in which they feed each other: no signals, no graph; no graph, no explanation; no explanation, no fix that means anything. The right-hand column is as wide as the left, and that is deliberate — what a layer does not do is what makes the rest believable.
| Layer | What it does | What it does not do |
|---|---|---|
| Source Signals | Gathers what is publicly out there about you and which sources an engine fetched to answer. | Does not say you are mentioned there. “Reddit was consulted” and “you are discussed on Reddit” are two different things. |
| Evidence Graph | Gives every fact about your company a source, a method and a confidence figure, and flags where two sources contradict each other. | Does not decide who is right. A conflict stays a conflict until someone confirms or refutes it. |
| Decision Drivers | Measures the 25 factors that steer an engine's choice, per engine and per market, and ties each factor to a kind of fix. | Does not add them up into one number, and does not measure all of them: whatever needs a connection that is not there stays dormant, with the reason. |
| Execution Engine | Places the fix on your connected channels, with preview up front, approval per action type, verification afterwards and rollback on failure. | Does nothing without a connection and nothing without approval. The safety gate is shut by default and opens per platform. |
| Fixproof | Asks the same prompts again with the same method and puts before → after side by side. | Promises no rise. No result is claimed that does not come from a re-measurement, and what drops goes back. |
Which sources does ceeme actually read?
Most of what an engine knows about you does not come from your site. It comes from review sites, directories, press releases, marketplaces, social profiles and encyclopedias — places where you do not publish and which you often do not even know about. Source Signals reads those channels, plus the sources you connect yourself.
What does this rest on, beyond our own word?
On research that is not ours. Aggarwal et al. set nine content changes against a baseline across roughly ten thousand questions and several generative engines: statistics, quotes and clear source attributions gave 22 to 41% more visibility in AI answers, and keyword optimisation did less. That is why a fix here follows from a finding and not from a word list. And the limitation belongs with it, because it is half the argument: this is about visibility in a generated answer, not about visitors and not about revenue.
Why five layers and not one score?
One composite number hides which lever is broken. Going from 14 to 41 still does not tell you whether that is because you get named more often or because one engine suddenly started citing you. We count mentioned, cited and chosen separately, and we publish the weighting.
where you stand in the answer
why the winner wins · winner / you
- Onafhankelijke bronnen
- Recente reviews
- Entiteit-duidelijkheid
- Casebewijs
- Lokale verankering
market share of the engines ceeme measures · Statcounter · AI chatbot market share · April 2026
- openai 70.83%
- gemini 10.56%
- perplexity 6.78%
- claude 3.85%
Where does the human come in?
In two places, both on purpose. In the proof: what you confirm as the owner weighs more than anything else, more than schema and far more than what an AI claims. And in the execution: nothing goes live without approval, and there is always a way back.
What is in place when this runs inside a larger organisation?
Five things that are in place, and for each one where it stops. That is the whole table below, and the right-hand column is not there out of politeness: a list of five compliments cannot be checked, a list with a boundary can — because you can run into that boundary.
| What you get | What it does | Where it stops |
|---|---|---|
| A trail of every action | Every agent action with its timestamp, input, output, who signed off — and the REASON, because that field is mandatory. Exportable as JSON or CSV, so it can go into your own system. | It is our own log and not a third party's statement. If you need a statement an auditor signs, this is not it — see below. |
| One click back | Every execution keeps the previous version and has a way back. Rolling back itself needs no approval: asking permission to repair something broken is making the mistake twice. | Outside your own channels nothing rolls back. A sent email and a posted update are seen before you take them down — that is the only action in this system with no way back, and that is why a human stands in front of it. |
| Roles and seats | Who may look and who may sign off, per seat. The number of seats depends on your plan and sits with the price, not in the fine print. | It is not single sign-on. Anyone wanting to connect their own identity provider needs an agreement, not a checkbox — that too is below. |
| Several brands, several clients | One view across several brands or clients, with a report under your own name and approval per client instead of one switch for all of them. | No amount is attached. The scope differs too much between agencies to put a price on it that is true for everyone, so this is always a conversation. |
| A door for machines | An MCP endpoint where an AI agent fetches your verified company picture directly instead of crawling your site. 12 tools, 8 of which answer without anything being connected. | Nothing runs ahead here, and that is deliberate. An answer to a machine is passed on as fact by the agent, so the tools waiting on a measurement return an explicit error rather than an empty answer — empty would mean “visible nowhere”. |
And what if I already have an agency or a tool for this?
“I already have an SEO agency.” Keep it. Kargaev put SEO and GEO factors side by side in one normalised framework and lands on a reweighting, not a replacement: authority keeps carrying weight in both, and the claim that SEO is over goes, in his own words, further than the evidence allows. What you built over ten years is not thrown away here — it gains a second weighting, and that one you can see.
How far does ceeme go for you?
Four routes, and for each one what it does not do. That boundary is not fine print — it is enforced in the code. Watch never delivers a fix pack, DIY delivers the fix but does not apply it, and ProOps applies but never without your approval.
Frequently asked questions
Is ceeme One an SEO tool?
No, though it overlaps. SEO is about ranking high in a list of blue links. ceeme One is about being chosen in one AI answer: which facts an engine knows about you, whether they are verifiable, and whether you become the logical recommendation. The technical base — crawlability, schema, speed — is part of that, because a page a crawler cannot reach will not be cited either.
Do I have to take all five layers?
No. Measurement and explanation come with every route. Execution only exists in ProOps, and only on the channels you connect yourself. Anyone who just wants to know where they stand stays in Watch and does not pay for an engine they never use.
Which AI engines are in the measurement?
The core measurement covers ChatGPT, Google AI, Perplexity and Claude. Per market and sector, more engines are added. The table shows per engine whether it was measured and whether it counts towards the weighting: an engine whose market share in your market is unknown gets measured but not weighted — giving it an invented percentage would pollute the number.
What happens if a fix does nothing?
Then it says so in the Fixproof report, with the same prompts and the same method as the baseline. We claim no result that does not come out of a re-measurement, and if something drops, we roll that change back. A tool that only re-measures its successes measures nothing.
Read on
Free, no account and no card.