Your buyers have stopped searching and started asking. I make sure the answer names you.
No account manager. No junior writing your category page. No team you never meet. When you send a question, the reply comes from the person who ran the measurement and wrote the page — which is only possible at three clients, not thirty. Everything below is what that involves, what it costs you in scope, and who to call instead when it is not a fit.
In short
- Three clients at a time, so the person who writes your pages is the person reading your results.
- Six years in crypto — positioning, content, go-to-market, launches — now pointed at the engines that answer instead of listing.
- Every engagement starts with a measured baseline. Without one, nobody can tell you whether the work did anything.
- The method is public and so is the evidence: 1,500 AI responses and 12,518 citations, downloadable, no email gate.
- No KOLs, no paid media, no community management, no press. Those are real services and I do not sell them.
Why only three clients
This is the question I get asked most, and the answer is not scarcity marketing.
The work is writing and judgement, not production. Getting a protocol named in an AI answer means writing the passage a model will quote — the definition of a category, the comparison that is actually fair, the paragraph that stands up on its own once it is pulled out of the page. That is slow, and it does not get faster by adding people.
It cannot be delegated without losing the thing that makes it work. A junior can write around a category. They cannot write the definition of a perpetual DEX or a stablecoin rail, because that takes knowing the category well enough to be right about it in one sentence. Six years in crypto is the input here, and it does not transfer to whoever an agency assigns to your account.
The person writing is the person measuring. No account layer, no status call where someone relays what someone else found. I read your engine results, so the next thing I write is a response to them rather than to a brief.
Three is what fits. A fourth client means one of the four gets a version of this done at arm’s length, which is the thing I built this to avoid. When there is no slot, I will say so and point you at someone with capacity.
What an engagement involves
Four steps, in this order, because the first one is what makes the rest measurable. The method page has the long version, with the evidence behind each step.
| What happens | What it is not | |
|---|---|---|
| 1. Baseline | I ask your buyers’ real questions across four engines, several times each, and record every answer and every citation | A keyword report |
| 2. Rewrite | I rewrite the pages and passages a model can actually quote, in the words your buyers use rather than the ones your site uses | A content calendar |
| 3. Off-site | Documentation, aggregators and the places the citation data shows these engines actually read | A press release campaign |
| 4. Re-measure | The same questions re-run every month in kymo, showing what moved and what did not | A dashboard nobody opens |
The baseline is not a formality. AI answers produce no impressions and no sessions — there is nothing in your analytics to compare against later. If nobody recorded what the engines said in month zero, nobody can honestly tell you what changed by month six.
What I bring to it
Six years marketing crypto from inside it. Positioning, content, go-to-market, launches and the community work that actually moves numbers — not answer engines alone. I specialised here because I kept watching teams do everything right and lose deals anyway, with nothing in the analytics to explain it. The buyer had asked a model, got three names, and none of them was the client. That is a retrieval problem, not a campaign problem.
Published measurement rather than case studies. 1,500 AI responses across four engines and nine markets, 12,518 citations, every scoring script included, under CC BY 4.0 with no email gate. Two of six pre-registered hypotheses in edition one failed and are published as failures. You can check the method before you pay for any of it — it is all here.
My own tooling. I built kymo because nothing on the market recorded which crawler read what, or which citation actually sent a human. Every client report runs on it.
A control I do not get paid for. RemoteStack is a job board I own, with no client and no budget, now pulled into AI answers about remote hiring. It demonstrates the method rather than the brand, which is the point of having it.
Categories I have measurement in
I take work in categories where there is already a published corpus, or a clear route to one. These three have the data behind them:
- Perpetual DEXes — including Velar, where the category had no written definition, so I wrote one and placed it where the engines read.
- Stablecoin rails and cross-border payments — the category edition one measured directly, across ten real money problems.
- Crypto cards and neobanks — the category edition two measured, across nine markets.
If your category is not on that list, it is not a no. It means there is no published measurement for it yet, so a baseline would be the first piece of work rather than an assumption I make on your behalf.
Where I am the wrong choice
This is on the services page rather than buried three clicks down, because a badly matched engagement costs a solo operator more than a lost one does.
If you need a team. Twelve people, an account layer, coverage across paid, social and community. That is a real requirement and I am structurally not it.
If you need KOL or paid campaigns. I do not run them and I have no measurement to offer on them.
If you need tier-one press. That is a relationships business built over years, and other people have those relationships.
If you want guaranteed placement in AI answers. Nobody can offer that honestly. In my own index, ChatGPT ran a web search on 54% of its answers and Claude on 65% — the rest came from training data, which nothing published afterwards can reach. Anyone guaranteeing an outcome across that is guaranteeing something they do not control.
If the category genuinely has no buyer questions yet. Pre-narrative categories need the narrative first. That is sometimes the same work and it is usually a different budget.
For a direct comparison against the firms people usually weigh me against — including what each of them does better than I do — the alternatives pages and the five-firm comparison are written to be fair rather than flattering.
How to start
A twenty-minute call. I run your questions across the four engines before we speak, so you are not spending the call explaining your protocol to a stranger — you are looking at what the engines already say about it.
Book twenty minutes, or read the questions I get asked most first.
Sources
- Google Search Central, “AI features and your website”. Google. Checked 2026-09-12.
- Aggarwal and colleagues, “GEO: Generative Engine Optimization” (2023). arXiv:2311.09735. Checked 2026-09-12.
- Lewis and colleagues, “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks” (2020). arXiv:2005.11401. Checked 2026-09-12.
Published 2026-09-12. Written by Narender Charan, who runs Kunzum.