AI answers name a few brands and skip the rest. We measure the questions your buyers ask AI, find the sources that decide each answer, then own and earn those sources, so your brand gets mentioned, cited, and recommended.
Published June 2026 by Creative Data Engineers.
People read an AI answer instead of clicking a list. The brands named in the answer win the consideration. The rest stay invisible while the buyer decides.
of organic clicks disappear when an AI answer appears, and the search stays inside the AI.
of answers in one documented case cited a brand's own listicle, while the AI recommended a rival named inside it.Cited is not recommended.
The metric moved from rankings and clicks to share of voice, citations, and recommendation. AI search is an early, higher-converting touch point, because the visitor arrives more informed.
It reads the sources that already rank and names the brands that recur across them. Change what those sources say about you and you change the answer.
A brand can be a footnote the AI read and still not be the name it gives. In one documented case a brand's own listicle was cited while the AI recommended a rival named inside it.
Every period of absence is a set of buyers who never see you while they are deciding. The gap does not pause when you do.
An AI answer is one question fanned out into many hidden sub-questions. The model forms a hypothesis of what a good answer needs, runs grounding searches for each sub-question, pulls the top-ranked passages, and names the brands that recur. Win the sub-questions and you get named.
Puts you on the shelf. The AI summary sits on top of the existing search index, so you have to be in the index it reads.
Answering the fan-out gets you cited. When your content fits the sub-questions a buyer asks, the AI pulls your passage.
Co-occurrence gets you recommended. When you appear across the sources that decide a category, you become the name it gives.
Two surfaces, and we win the summary first. The quick AI summary and the deeper browse mode share most answers but draw on different sources.
GPTBot fetches analysed found no JavaScript execution. The major assistants read raw HTML.Vercel and MERJ, 2026
JavaScript run by GPTBot, ClaudeBot, or PerplexityBot. Gemini is the exception.Content behind client-side rendering can arrive as a blank shell.
Server-side or static HTML is the entry ticket. If the content you want cited only appears after client-side rendering, most assistants cannot see it.
We run the same loop every engagement. It starts from real demand, reads each answer and the sources behind it, finds the gaps and the sources that decide them, then reacts by owning or earning, and re-scans to prove the lift.
Start from real demand. The questions your buyers actually ask AI, derived from search data, not guessed.
Read each answer and its sources. We record how you show up: absent, cited, or cited and recommended, and every source the answer cites.
Map the fan-out. The sub-questions where you fail to appear become the target list.
Find the deciding sources. The high-authority publications that keep getting cited in your category, the North Stars, and the ranking sources that cite competitors but not you.
Own or earn. Improve the content you control, and get included in the sources you do not. Editorial and review listicles get a digital-PR pitch; retailers get a merchandising or data-feed conversation; community gets authentic presence.
Re-scan for the lift. We A/B the new questions against the old set and confirm your mention, citation, and recommendation share moved.
Four pillars carry the work, with measurement running through all of them. Two are about knowing (the questions and the sources), two are about acting (what you own and what you earn).
Data-derived questions and their fan-out. The foundation everything else targets.
The North Star map, the gap list, and what AI believes about you, traced back to the sources that say it.
Answer-fit content, schema, llms.txt, and rendering so crawlers see real content. What you control.
Editorial, retailers, and community. High authority beats volume, once the North Stars are mapped.
Measured throughout: share of voice, citation share, recommendation share, and sentiment, tracked per market and per product.
We run two tracks at once, so you get results while the engine is still being built. The quick wins are no-regret and need no engine; the engine gives the targeting and the earned-source strategy.
Across every market in parallel: server-rendered content, schema and FAQ, llms.txt, answer-first page copy that leads in the first 150 to 200 characters, video captions, and owned social. No engine required.
Secure the inputs, build the engine, scan on the real questions, build the source intelligence, and earn placements in the gap sources. This is what makes the targeting precise.
A first scan proves the method and produces a first source map in days, the owned quick wins begin immediately, and the question engine follows in a market where the data is ready.
A first scan on a sensible question set, to prove the method and produce a first source map. We treat it as starting evidence, not the foundation.
The owned quick wins begin immediately, across every market, in parallel. They are no-regret and need no engine.
We build the question engine in a market where Search Console access is ready, replace the hand-picked questions with demand-derived ones, and measure the lift.
What we need from you: Search Console access for the markets we measure, and the green light to roll out the owned quick wins. The first scan needs nothing from you.
What teams ask before working with us on AI search visibility.
SEO gets you ranked. AI search visibility gets you named. We work the layer on top of ranking: the hidden sub-questions an AI expands a query into, and the sources it reads to decide which brands it mentions and recommends. Ranking puts you on the shelf; answering the fan-out gets you cited; appearing across the deciding sources gets you recommended.
You control your owned levers: answer-fit content, schema, llms.txt, and server-side rendering so crawlers see real content. You earn inclusion in the sources you do not own: editorial and review publications, retailers and aggregators, and community platforms. We work both tracks, and we prioritise high-authority sources over volume.
Most read raw HTML and do not run JavaScript, so content that only appears after client-side rendering can reach them as a blank shell. This is documented for GPTBot, ClaudeBot, and PerplexityBot; Gemini is the exception. Server-side or static HTML for the content you want cited is the entry ticket, not the prize.
The owned quick wins start in the first weeks and need no engine: schema and FAQ, llms.txt, answer-first page copy, and server-rendered content. The deeper lift from the question engine and earned sources builds over the following months, and we confirm it with re-scans that compare the new question set against the old.
Search Console access for the markets we measure, and your approval to roll out the owned quick wins. The first scan needs nothing from you: we run it on a sensible question set to prove the method and produce a first source map before any deeper build.
We recommend. We provide ready-to-use content and schema, brief your development and content teams, and re-run the scan after the changes go live to confirm the lift. We do not implement directly in your CMS.
This is how we work. The service page covers what you get and what it costs, and the free AI Visibility Report shows the method in action every two weeks.