46% of SaaS companies aren't named when AI recommends their own category

Method

We picked 45 software categories with real competition, deliberately unglamorous ones: legal practice management, gym management, podcast hosting, uptime monitoring, form builders, veterinary practice software, expense management.

For each category we ran 5 prompts through ChatGPT, with web search enabled, including what is the best [category] and what are the alternatives to [company]. For each of 114 real companies we recorded whether the model named them, where they appeared relative to competitors, and which sources the answer cited.

Every run was on 2026-09-10. Scores are the same 0–100 visibility score QueryQue reports to its own customers, so the numbers here are directly comparable to a scan you run yourself.

What came back

1. It is close to winner-take-all

Of 114 companies, 47 were named in all five answers and 31 in only two. The middle is thin. A company is mostly either in the answer or it is not, with little gradient between.

2. Nearly half are missing from the plain question

53 of 114 were absent from the straightforward what is the best X answer. These are not obscure products. They are established companies with real customers and real revenue, and in several cases more features than whatever the model recommended instead.

3. Generalists crowd out specialists

151 distinct companies were named across 114 answers, but a handful appear again and again, including in categories that are not theirs.

A specialist tool competing in its own narrow category is often losing to a generalist that simply has more written about it on the open web.

4. Some categories are effectively closed

In the worst category we measured, every specialist tool we checked was absent from every answer, while the same three generalist names were returned each time.

Every category

Sorted by the share of measured companies that were absent from the category's own "best X" answer. "Top named" is who the model returned instead.

A methodology note worth stealing

Brands whose names are ordinary English words score badly wrong if you match naively. A company called Convert initially scored 86 out of 100, until we found it was matching the word "convert" in ordinary prose. Corrected, it scored 37. Reputation.com went from 88 to 37 the same way.

Anyone running this kind of measurement needs to handle dictionary-word brand names explicitly, or the numbers are noise dressed up as data.

Limitations

One engine, ChatGPT, with web search enabled. One run, on one day. Answers move between runs, between models, and as the web underneath them changes.

45 categories is not a representative sample of all software, and 114 companies is not a representative sample of those categories. Companies were selected for having a public website and a clear category, not at random.

This is a snapshot, not a benchmark. We would not make a decision on a single run of it, and neither should you. The pattern that held across all 45 categories is more durable than any individual number in the table above.