Software directories took 165 of our 9,198 citations
Ask whether you need G2 and Capterra reviews to get into AI answers and you'll find two confident answers. One says review profiles became the main source behind software recommendations, so claim yours and fill in every field. The other says they barely get cited. So we counted.
Between 23 August and 2 October 2026 we pulled every citation our sampler stored for six B2B software brands in our own accounts, across HR tech, security and construction estimating. Four engines return a source list we can read: our ChatGPT engine on OpenAI's API, Google AI Overviews, Perplexity and Claude. 1,099 answers carried at least one citation, 9,198 citations over 1,995 domains. Read every number here as ours.
- 165 of the 9,198 citations pointed at a software directory. That's 1.8%, with a 95% confidence interval of 1.5% to 2.1%. We counted ten sites: G2, Capterra, Gartner, Software Advice, GetApp, TrustRadius, SourceForge, PeerSpot, SelectHub and Slashdot.
- 125 of the 1,099 answers carried at least one directory link, 11.4%.
- 19 of those 165 pages belonged to the brand being measured. The other 146 were a competitor's page, or a ranking the brand doesn't control.
- For scale, two HR trade publications took 240 citations between them, more than all ten directories combined.
The cited page is a ranking, or a rival's alternatives list
Share of the pool tells you whether to care. Which page got taken tells you what to do.
- 49 were a category or market ranking, 29.7%. The "best X software" listing, or a Gartner market page.
- 47 were an alternatives page keyed to one named product, 28.5%. Titled something like "top alternatives to <competitor>", with you somewhere in the list.
- 39 were a single product profile, 23.6%. Seven of those 39 were the measured brand's own.
- 24 were editorial, research or press, 14.5%. Mostly Gartner research documents and press releases, which is the analyst firm, not the peer review site.
- 6 compared two named products head to head, 3.6%.
The profile page was 8 citations out of 9,198
Add up the own-profile hits across every site and you get 8. Eight citations landed on the page a review collection campaign is built to improve. One of the eight is a near name match we couldn't confirm is the same company, so call it seven or eight.
The most cited directory page in the sample was a brand's own alternatives tab, pulled 11 times. That's the page listing its competitors.
G2 publishes a list of what a vendor can't influence
G2 keeps a page at /ai-instructions written, in its own words, for AI assistants such as ChatGPT, Claude, Perplexity, Gemini and Microsoft Copilot. There's no separate version served to automated clients. It prints two lists.
A vendor can claim and update its profile, answer reviews publicly, ask customers for honest and unbiased reviews, and buy advertising in sponsored positions. A vendor cannot, in G2's words, influence report or live Grid ratings, placement, or sort order.
Line that up against the page shapes. The most cited directory page in our sample was the category ranking, and G2 says plainly that a vendor can't influence placement or sort order on it. The lever it does offer is asking customers for reviews, which feeds the score that decides placement. That's a real lever, and it's slower than "optimise your profile fields" suggests. It also works through a page you don't own.
ChatGPT reached for a directory four times as often as AI Overviews
The 11.4% hides a split big enough to change what you'd spend on.
- Our ChatGPT engine: 68 of 306 answers carried a directory link, 22.2%, interval 17.9% to 27.2%. Directories were 4.1% of its citations.
- Google AI Overviews: 42 of 739 answers, 5.7%, interval 4.2% to 7.6%. Directories were 0.71% of its citations.
- Claude: 8 of 24 answers. Perplexity: 7 of 30. Both lean the way ChatGPT does, with intervals too wide (18% to 53%, and 12% to 41%) to separate from anything.
Access isn't what's holding the directories back
The easy explanation for a thin share is that engines can't read these sites. They can.
G2's robots.txt puts GPTBot, ClaudeBot, Google-Extended, Amazonbot, CCBot and two others in one group, and the rules under it are URL parameter and account path patterns, not a site wide block. Bytespider and DeepSeekBot do get blocked outright. The file's last line declares an llms.txt as a sitemap, and that file returns 200 as markdown: category pages listed first, the /products/{product-slug}/competitors/alternatives pattern documented, every category's Grid exposed as a PNG, an SVG and a JSON feed. Capterra names no AI crawler at all, so they're allowed under its wildcard. PeerSpot goes furthest, with a block its own file calls an AEO AI agent allow list.
PeerSpot got one citation in our entire sample.
So the thin share isn't a blocking problem. These sites are readable, several are instrumented for models, and the engines reached elsewhere anyway.
What we'd actually do
Cheapest first.
- Split the directory question by engine before you spend anything. One blended number hid a four fold difference in ours.
- Find out which directory page your category's answers cite, then work on that one. If it's the category ranking, the job is placement, and the platform documents what moves it.
- Treat your competitors' alternatives pages as a surface you care about. They took 47 citations against the 8 that landed on an own profile.
- Keep collecting reviews. Just stop expecting the profile to be the cited link. Reviews feed the ranking, and the ranking is what gets cited.
- When a tool quotes you a review-site influence figure, ask for the denominator. Share within review-platform citations and share of all citations are different questions, and only the second one tells you whether to invest.
Where this could be wrong
Six brands is thin, and one supplied most of the answers. Three categories, all B2B software, weighted heavily to HR tech. A CRM or project management sample could land differently, and a consumer category differently again. The directory list is hand built from ten domains, so add a site we missed and the share goes up.
We can't tell you whether a directory citation goes with getting named. Only 24 of these answers had both a brand mention score and a directory link, far too few, and the two engines with enough rows pointed opposite ways. Nothing here says a G2 presence doesn't help you get mentioned. It says the directory page is rarely the one the engine shows.
Our counts come from the citation annotations each engine emits, and OpenAI documents that its sources field runs longer than its citation list. Retrieval is undercounted, so a directory page could inform an answer without appearing in these 9,198. Claude and Perplexity only started returning citation rows on 30 September, hence the small cells.
The robots.txt reading is a snapshot from 2 October 2026 and these files change. We read six of the ten directly. Gartner and TrustRadius returned 403 to us, and we didn't check SelectHub or Slashdot.