NamedraHow ChatGPT chooses which vendors to recommend
When a buyer asks ChatGPT for the best vendors in a category, it runs web searches, reads a handful of pages, and names the vendors that appear most consistently across those pages with the clearest descriptions. Your homepage is rarely one of those pages. Review sites, comparison articles and communities are.
What happens in the ten seconds after the question
With search enabled, the model turns the question into one or more queries, fetches the top results, and reads them. It then writes an answer that summarises what it read, attaching citations to sentences. The list of vendors it returns is largely the intersection of the vendors named on the pages it fetched. If three of five pages name Vanta, Drata and Secureframe, those three are the answer, regardless of what Vanta's homepage says.
Four things that decide whether you are named
1. Whether the pages that rank for the query mention you
The query is usually close to the buyer's words: "best compliance automation software for startups". The pages that rank are listicles on review sites, vendor blogs and comparison sites. If you are not on those lists, you are not in the answer. This is why "earning mentions on third-party pages" is most of the work.
2. How clearly those pages describe you
Engines reuse descriptions. If G2 says "compliance automation with in-house auditors" and your site says "the trust platform for modern companies", the engine will use G2's words, or skip you when the prompt asks for something specific. One clear sentence, used everywhere, is worth more than a tagline.
3. Entity confidence
The model has to be sure that "Acme" the compliance vendor is the Acme in the fetched page and the Acme in its training data. Renames, inconsistent descriptions and missing structured data lower that confidence. Organization schema, sameAs links to your profiles, and a consistent description on LinkedIn, Crunchbase and review sites raise it.
4. Extractability
Pages with direct answers, tables and short definitions get quoted. Pages built from hero images and adjectives do not. A comparison page with a table of features and prices is the most citable thing most vendors can publish.
What about answers without search?
When the model answers from training data alone, the list reflects what was written about the category before the cutoff. It changes slowly and rewards years of consistent coverage. You can still check it: our audits include an offline baseline for exactly this reason.
Why answers vary
Two runs of the same prompt can fetch different pages and name different vendors. That is why we run each prompt three times and report a mention rate rather than a single answer. A vendor named in 2 of 3 runs is "usually there"; 0 of 3 is a gap worth fixing.
What you can influence this quarter
- Get on the listicles that rank for your money prompts; updating an existing article is easier than earning a new one.
- Publish comparison and alternatives pages with tables.
- Refresh reviews on the review sites that get cited.
- Fix entity data: one name, one description, schema, sameAs.
- Measure monthly with the same prompts.
To see where you stand today, run the free audit or follow the do-it-yourself checklist.