Your buyers' questions
We start from the buying moments in your market: the situations in which someone starts thinking about your category, and the questions buyers then ask. That is where the questions we measure come from.
Our vision
How we look at AI search
Our goal is simple: that AI assistants recommend your brand at the moment a buyer decides. More and more, people ask an AI assistant for advice before they buy. Who gets recommended then depends mostly on what is known about your brand and how well that fits the question. Below is how we look at it, and what it means for how we work.
We are different
Almost every GEO tool is a prompt tracker. It regularly fires a fixed set of prompts and measures whether your name appears in the answer and whether your site is cited. That gets wrapped in a slick dashboard. Then it looks at what the sites that are named do differently, and helps you produce content at speed on the topics you are still missing. It sounds logical, but it does not get you recommended more often. Fast, shallow content can even backfire.
Back to basics: first principles
Once we saw how limited the existing tools were, we started at the beginning: how does an AI assistant actually arrive at an answer? We applied first principles thinking. You reduce a problem to the facts you are sure of and reason again from there. That way you avoid taking over other people's assumptions. We read almost every scientific paper on LLMs, tested the findings ourselves and looked at what they mean in practice.
The outcome: an AI assistant first builds its answer from what it already knows, from its memory, and then refines it with live search results. Whether your site is cited therefore matters less than many tools would have you believe. What counts is what the model knows about your brand and whether that fits the buyer's question.
Only once we had taken that apart did we start building our own solution.
How an answer comes about
Two ways into an AI answer
An assistant names you because it already links you to the subject, or because the pages it reads name you. Which of the two you have decides what work pays off.
Model memory
- 1Velora
- 2Brisk Bikes
- -Acme Cyclesnot recalled
- -Pedalonot recalled
Pages it reads
- 1Pedalo
- 2Acme Cycles
- 3Velora
- 4Brisk Bikes
The answer
- 1Velora
- 2Pedalo
- 3Acme Cycles
- 4Brisk Bikes
The pages it reads name you, while the model itself does not know you. You make the answer, behind the names it already had.
Keep those pages. Moving up takes brand work: the same story about who you are, in the places models learn from.
Pages: leave you outPages: name you
Memory: does not know you
Memory: knows you
An illustration. In the platform we count this per subject: for every live answer, whether the model already knew you and whether the cited pages name you.
What we measure differently
Because we know how an answer forms, we measure different things than a prompt tracker. In plain language:
Memory and live, separately
We measure separately what an assistant knows about you on its own and what it says after it has gone searching. So you see which of the two routes above needs work in your case.
What the models look up
When an assistant searches, we see which searches it fires itself and whether your pages land on them. So you know which words and topics your site is still missing.
Reputation: what AI already believes about you
We ask ChatGPT, Claude, Gemini and Perplexity open questions about your brand and set the answers beside how you want to be seen. You see which qualities and topics they already link you to, what is missing and what is wrong.
A margin on every number
A language model gives a slightly different answer to the same question every time. One measurement says little. So we ask every question several times and put a 95% confidence interval on every number: the margin within which the real value almost certainly lies. We only call a change a change when it falls outside that margin.
Conversations up to the choice
Nobody stops at the first answer. A simulated buyer from your audience holds a whole conversation with the assistant, with follow-up questions, until a choice is made. So you see whether you are recommended at the end, where you drop out and why.
What we believe
Getting recommended is brand strategy first
Only then a technical discipline.
An AI assistant recommends you on the strength of what can be found about you: on your site, in reviews, in trade press and on forums. The work is tying the right associations to your brand, in the places a model learns from and the pages it reads before it answers. ChatGPT, Claude, Gemini and Perplexity sum all those sources up in one piece of advice. That is where you see most sharply whether your brand story lands.
Your own site is usually the starting point. Logical, because it is the one place you fully control. The evidence a model uses to back its recommendation, though, comes largely from outside your site: what customers say, what trade press writes, where you get named. That is also where most of the work is.
Where people search depends on what they want. Our Search Intent Framework describes exactly that: nine intents and fifteen platforms, from Google and the AI assistants to YouTube, TikTok and Reddit. Every topic in an audit carries one of those intents.
Your positioning
Who you are, for whom you are the best choice and for whom you are not. Crystal clear, because a model can only place you if you know where you stand yourself.
Your site
Technically sound for Google, readable for AI and convincing for buyers. The foundation, and the only thing you fully control.
Beyond your site
Reviews, trade press, forums and the pages the assistants cite. This is where buyers and models collect their evidence, and where most brands have the most work to do.
Together that is your body of evidence: what you claim, what your site proves and what others say about you. We measure that evidence, what the model has made of it and what it looks up before it answers. Then the answer itself, measured repeatedly, and all the way into the conversation where a buyer decides.
That is why getting recommended more often goes straight to your core: your positioning, your proof and what customers say about you. It takes your whole marketing team: brand, content, PR and tech. That is why our audits end in an action plan ordered by expected impact and effort, with who picks up each action.
How we work
With us you buy an outcome: advice and a plan, for a fixed price. Our platform measures, repeatedly and with the margin shown. Then one of us reads every finding, writes the conclusion and builds the presentation. Nothing goes out unread. We are level-headed about AI: it cannot be hacked, so we sell no shortcuts. Writing content we leave to you, your agency or your own AI assistant. We tell you what is missing, why it matters and who should pick it up.
Who we are
Recommendable was built by Peter Minkjan and Erik Mus, who between them have spent almost thirty years on how brands get found.
About usSee it on your own brand
Pick a package at a fixed price, and we plan the intake together.
Request now
Fixed price · no subscription · results within two weeks