AI visibility: being named in AI answers
Buyers increasingly ask ChatGPT or Perplexity for a recommendation instead of scrolling a list of links, and that answer names two or three companies. This work is about being one of them: making a site quotable by AI assistants, and measuring whether it actually gets quoted.
What is different from SEO?
Classic SEO optimises for position in a list of links. This optimises for inclusion inside a generated answer. Different mechanism, different metric: citation share across a fixed set of buyer questions, rather than rank.
There is an asymmetry here that favours smaller sites. In a list of links, a new domain does not outrank a competitor with fourteen years of history. Inside an AI answer it can, because the model assembles the answer from facts and structure rather than from domain age.
What does the work involve?
1. Baseline measurement, before any changes. 30 to 40 questions phrased the way your buyers actually phrase them, run across six AI engines. Recorded for each: whether you are named, who is named instead, which sources are cited. This starting point cannot be reconstructed after the fact.
2. Rebuilding for extraction. Crawler access for AI bots, structured data, and text reorganised so a model can quote a fact cleanly. Plus presence on the external sources the engines already read.
3. Re-measurement after four to six weeks. Same question set, same protocol. The comparison is honest because the instrument did not change.
What results has this produced?
A personal services site, new domain, no clients. 34 questions, 6 engines, 116 runs per measurement.
| Query type | 1 Aug | 7 Sep |
|---|---|---|
| Commercial ("who is best", "recommend someone") | 0% | 15.3% |
| Brand and English-language | 31% | 56% |
| Citations of the domain per measurement | 0 | 19 |
Five weeks. On the key commercial queries the site is now named first in the answer, ahead of a school whose domain is fourteen years old.
The site is taichi.valdas.online — you can run the queries yourself.
What did not work?
On problem-oriented queries — "what helps with burnout", "exercises for back pain from desk work" — three measurements produced zero mentions. Those answers cite medical primary sources, and entering them needs a different approach than the one that worked for commercial queries.
This is stated up front deliberately. The criterion for declaring an approach failed is written down before the work starts, not after.
How do we start?
A 45-minute call: your niche, your competitors, and what counts as success. Within one working day after it, you get a written spec.
The spec is yours either way — do it yourself, hand it to your team, or hand it to me.
Related Services
- AI MVP builder in 7 days — Full-stack web app
- custom Telegram bot development — AI bot in 7 days
- analytics dashboard development — Custom tool in 14 days
- ← Back to homepage