Primary research · August 2026
How ChatGPT Shortlists Software Brands
An audit across 10 categories and 60 buying questions. I recorded what ChatGPT reads, throws away and links to when a buyer asks it which software to buy, and what that decides.
- Every figure is measured from captured data, none are estimates.
- The full method, so you can re-run the audit on your own category.
- What to do about it, split by whether you run the company, the marketing or the site.
Nathan MzumaraAI Search & Digital Growth Strategist, twelve years in search
- 60
- Questions asked
- 10
- Software markets
- 2,680
- Results read
- 367
- Links shown
- 150
- Brands named
Free download
Get the full report
35 pages · PDF · 536 KB. Enter your details and it downloads straight away.
What is in this report
Your buyers have started asking ChatGPT which software to buy, and nobody can tell you how it picks. Most of what gets written about it is guesswork. The common story goes: AI reads G2 and Reddit, so collect reviews and get talked about on Reddit. I wanted to know whether that was true, so I stopped guessing and measured it.
I ran 60 software buying questions through ChatGPT, across ten B2B categories and six different ways of asking, and recorded everything it did. Not just the answer, but the searches it ran, the pages it read, the pages it threw away, and the handful it finally showed. Then every one of those websites was sorted by hand into what kind of site it actually was.
Every AI visibility tool on the market shows you citations: the links that appear in an answer. None of them show you what the model read and discarded. This audit captured both, and the gap between them turns out to be where most of the story is.
A preview of the findings
3 of the 6 headline findings, in full. The rest, and the numbers behind them, are in the report.
65.9%
Your own website is the biggest source of evidence
Two in every three links ChatGPT showed pointed at a software company's own site. In 26 of the 60 answers, every single link was a vendor page. Pricing, plans and feature pages are doing more work than anything else you publish.
69.3%
Most recommendations are not backed by anything
248 of 358 brand recommendations had no link to that brand anywhere in the answer. The links sit next to checkable facts like prices. The advice itself, who is best, usually arrives with nothing attached.
4.1%
G2 and its rivals are barely in the room
All the review sites combined produced 15 of 367 links. Reddit, LinkedIn and YouTube produced none at all. Meanwhile 43 small comparison sites nobody tracks produced 14.4%.
3 more findings in the full report
Download the free reportContents
21 sections across 35 pages.
- 01Why I audited ChatGPT's software picks4
- 02What I tested: 10 markets, 60 questions5
- 03How I recorded what ChatGPT was doing6
- 04Why the free model, and why it matters7
- 05How a shortlist is actually assembled8
- 0688.9% of what ChatGPT reads never shows9
- 0765.9% of links point to vendors' own sites10
- 084.1% of links come from G2 and its rivals11
- 0914.4% of links come from sites nobody tracks12
- 1069.3% of recommendations cite no source14
- 1186.3% of dated pages were published in 202615
- 12Only 34% of a shortlist survives rewording16
- 1348.7% to 90.9%: vendor share by category19
- 14Four real answers, taken apart23
- 15Six things I would do about this25
- 16What this audit does not tell you27
- 17Four follow-up tests worth running28
- 18Glossary of terms used in this audit29
- 19The full numbers behind every chart30
- 20Ten more things worth reading next33
- 21How to cite this audit and its data35
Free download
Get the full report
35 pages · PDF · 536 KB. Enter your details and it downloads straight away.