Three service lines, four layers, clients who already run SEO
Webvy’s clients arrive already running SEO. They are growth-stage SaaS companies, fintech brands, marketplaces, crypto and web3 companies and AI startups, and most of them can read a rankings report without help. What they cannot read is why an AI engine names a competitor instead of them.
The agency sells that gap as three separate lines — GEO for how a brand is surfaced and cited across generative systems, AEO for inclusion inside direct answers, AI SEO for the search foundation both stand on. Keeping them apart is deliberate: a buyer is meant to name their own problem before anyone quotes for it. Ivan Dyankov, who has been building online brands since 2007, leads the agency.
An audience that fluent sets the bar, and Webvy’s own method raises it again. Four layers — retrievability, interpretability, authority, answer fit — commit the agency to naming which layer a gap sits in, and a client who already runs SEO will ask which one. The deliverable is the product. One that cannot answer that question is a slide deck.
The question a percentage could not answer
The question a client asks is always specific. Why does ChatGPT recommend three competitors and not us, where is that answer coming from, and what do we change on Monday?
Until March 2026, Webvy answered it with the Semrush AI visibility tool. That tool does what it is built for — overall brand visibility across AI search, and share of voice against the category — and it told the agency whether a client was in the conversation and roughly how much of it belonged to somebody else. What it could not do was name a single prompt.
So every engagement started in the same place: with a category-level number that nobody can turn into a page. The GEO audit stayed a research exercise, assembled from nothing for each client and reusable for none of them. The monthly report had to be written in Webvy’s voice rather than out of the client’s own data — the weakest position an agency can hold in front of a buyer who already knows what a real report looks like.
Once the agency had a prompt-level view, a second problem showed up underneath the first. A brand can be named constantly across a category’s prompts and still have almost none of its own site being read. The mentions are real; the answers are assembled out of pages the brand does not control. Mention volume and citation strength are two different measurements, and only the second one moves when somebody works on a client’s website.
That decides where the money goes. If the gap is which sources an answer gets built from, the plan is written against pages, structure and third-party corroboration rather than against a score — and the audit has to name the prompt, the engine and the cited source before a single line of that plan exists.
Six steps, and nothing written before the prompt is named
Webvy rebuilt the engagement around one rule: nothing gets written until the audit has named the prompt it is answering. Six steps, run the same way for a fintech brand and for a marketplace.
01 · Prompt universe
Start from the prompts a buyer types while choosing
Five families, all commercial: category, comparison, alternative, trust and use-case. Webvy writes each one against the client’s own buyers and the decision in front of them, so the tracked set is the shortlist conversation rather than the top of the funnel where nobody is picking a vendor yet. A prompt set assembled from search volume would measure the wrong month of the buying cycle.
02 · Visibility audit
Read the whole set at once, engine by engine
The audit shows where the client appears, where they are absent, which competitors hold the prompts they are absent from, and which AI systems are most and least likely to include them. Coverage is read per engine, because a brand can be strong inside one system and missing from the next — and the fix for the second is not the fix for the first.
03 · Citations
Separate a mention from a source
Webvy reads citations apart from mentions. The question is whether the client’s own domain carries the answers it appears in, or whether the brand is being described out of pages it does not own. That one reading decides how much of the plan is on-site work and how much is corroboration somewhere else.
04 · Execution plan
Turn the gaps into six workstreams
Owned pages, technical structure, trust pages, comparison content, third-party corroboration, outreach. Which of the six carries the budget is Webvy’s call, made off the citation reading rather than off a default running order, and it is the point in the process where a strategist rather than a dataset decides.
05 · Reporting
Report movement on prompts a client can argue with
Clients see the change prompt by prompt. Someone can point at the comparison prompt a competitor held in April, see who holds it now and which sources moved underneath it, and push back on the reading. There is nothing to push back on in a sentence saying AI visibility improved, which is why a client already paying for SEO stops reading it.
06 · Content
Wire the API into Claude Cowork
Webvy connects the Amadora API to Claude Cowork and drafts articles against the brand’s current visibility gaps. The brief comes out of the prompts the brand is missing and the sources the engines read instead of it, so the content plan is set by the audit rather than by a keyword list.
Two parts stay manual on purpose. Webvy writes the prompt set itself for every brand, and a strategist decides which workstream the first month of budget goes to.
What the agency can put in front of a client now
A client-ready GEO/AEO report now takes under 15 minutes. The prompt universe, the competitor set and the citation data already sit in one place by the time anyone opens it, so what used to be a research exercise comes out as a document.
The judgement did not get shorter. Webvy still writes the prompt set, still reads the citations, still sets the running order. What disappeared was the rebuilding — the part of every engagement that produced nothing a client saw.
That figure is Webvy’s own account of its own delivery, and it is the whole of what this page claims. No client outcome is claimed here, because none was supplied — no before-and-after visibility, no mention counts, no citation numbers for any brand on the books. Every engagement also runs content, technical and outreach work through the same window, so movement in a client’s answers would not belong to the audit alone.
What changed is where the prompt-level view sits. It is the deliverable now — the same document in the audit, in the plan and in the monthly report.
You can generate client-ready reports in under 15 minutes, identify what needs fixing, and turn it into an action plan. We also connect the API to Claude Cowork to draft articles against the brand's current visibility gaps, so the content strategy comes from real AI search data rather than guesswork.
Ivan DyankovCo-Founder, WebvyCommon questions
What was Webvy using before, and what else did it try?
The Semrush AI visibility tool, for overall brand visibility and share of voice across AI search. It gave the agency the wider picture and stopped there. While looking for something that went further Webvy also tested Peec and Profound, and chose Amadora in March 2026 for the depth of the underlying data and what it could put in front of a client.
How do you build a prompt universe that is worth tracking?
Webvy builds five families, all commercial: category, comparison, alternative, trust and use-case prompts. Each is written against the client's own buyers and the decision in front of them, so what gets tracked is the shortlist conversation. A prompt set assembled from search volume would measure the wrong month of the buying cycle.
Why does a mention count for less than a citation?
Because they answer different questions. A brand can be named across most of a category's prompts and still have almost none of its own site being read, with the engines assembling every answer out of pages it does not own. Mention volume says the brand is in the conversation. Citation strength says whether its own domain is carrying any of it.
How does the client reporting avoid "AI visibility improved"?
Movement is reported prompt by prompt. A client can point at the comparison prompt a competitor held in April, see who holds it now, which engines changed and which sources moved underneath. That is a claim somebody can argue with, and it is what makes the report survive a client who already runs SEO.
What is the Claude Cowork connection doing?
Webvy connects the Amadora API to Claude Cowork so article drafts are written against the brand's current visibility gaps. The brief comes out of the prompts the brand is missing and the sources the engines read instead of it, which means the content plan is set by the audit rather than by a keyword list.