Learn · Guide
How AI engines choose which businesses to name
When someone asks an assistant for a recommendation, the answer looks like one decision. It is three, made in order, and a business can pass two and still be dropped at the third. Knowing which stage you fail at decides what to fix.
Updated 19 October 2026
Stage one: what to read
The engine rewrites the question into one or more web searches, runs them, and takes a handful of results — often five to fifteen pages. Nothing outside that set can appear in the answer, however good it is. Two things decide whether you are in the set: whether the crawlers behind the engine are allowed to read your pages, and whether a page of yours matches the way the question was rewritten.
- A robots.txt that blocks an AI crawler removes you before anything else happens.
- A missing or broken sitemap makes pages harder to find.
- No page for the question at all is the most common cause: the buyer asks about a named procedure, a price, or a comparison, and nothing on your site is about it.
Stage two: what it can extract
From the pages it read, the model lifts what it can resolve in one pass. A direct opening answer, a list, a table and marked-up facts are easy to lift. A price in paragraph nine, a claim inside an image or a service described only in a PDF are effectively absent. This is why two businesses with the same substance get different answers.
- A page that opens with a slogan gives the engine nothing to quote; one that opens with the answer does.
- Lists and tables carry comparisons and prices better than paragraphs.
- Structured data states who you are and what you offer in a form the machine does not have to guess at.
- A thin page that mentions a topic without answering it is read, and then not used.
Stage three: who to name
The model composes a recommendation, usually naming three to five businesses, and where its sources disagree it has to pick a version. Third-party sources carry real weight here: directories, review sites, press and comparison pages. An out-of-date price in a directory can outweigh your own page. The names chosen also vary from run to run, which is why a single run tells you very little.
- Being absent from the directories and review sites the engines cite for your category means competitors are named from sources you are not in.
- Facts that disagree across the web give the model a conflict to resolve, and it does not always resolve it in your favour.
- One competitor that owns most of the sources can dominate the answer.
Telling which stage you are losing at
The stages are observable. In a NamedFirst diagnosis, every gap is assigned a cause from a fixed set, each backed by a measurement rather than an opinion:
| Stage | Causes we test for | What the evidence looks like |
|---|---|---|
| Read | Crawler blocked · sitemap unhealthy · no page for the topic | robots.txt rules, sitemap status, no crawled page matching the question |
| Extract | Page too thin · no answer format · missing schema | Answer density in the first 200 words, lists and tables, structured-data types on the page |
| Name | Not cited · absent from third-party sources · contradicted by third parties · one competitor dominant | The URLs the engines actually cited, and who owns them |
The order matters. A plan that starts with new content before anyone has checked that the crawlers can read the site wastes the first two months, and a plan that chases directory listings while every page opens with a slogan wastes the next two. Fix the earliest stage first.
Questions
Common questions
Keep reading