Obility Editorial · · 4 min read
The study everyone cites shows less than it seems
Ask whether G2 and Capterra reviews help a brand get recommended by ChatGPT and most answers lean on one piece of research. Quoleady, a SaaS marketing agency, asked ChatGPT for alternatives to well known tools, with prompts such as “Trello alternatives” and “Slack alternatives”, then collected review data for every tool it named. It reports that 100 percent of those tools had reviews on Capterra, 99 percent had reviews on G2 and 78.8 percent had a Wikipedia page. Many posts turn that into a rule: no review profile, no recommendation.
The data cannot carry that rule. The study only looked at tools ChatGPT had already named, so it has no group of tools without reviews to compare against. And the alternatives to Trello or Slack are, almost by definition, established products that would have review profiles whether or not those profiles mattered to the answer. Finding that nearly every named tool has a G2 page tells you that G2 lists nearly every established tool. It does not tell you the page caused the mention.
Ratings and review counts barely moved the order
The more useful findings are the weak ones. Within each answer, Quoleady found only a faint link between how many reviews a tool had and where it appeared in the list, with correlations of 0.21 for Capterra and 0.16 for G2, and close to none for average star rating. Its own example makes the point: Coda, with 97 Capterra reviews, came third for Notion alternatives, ahead of ClickUp with 4,490.
The study gives no prompt count beyond “dozens” and covers ChatGPT alone, so treat it as directional. The direction makes sense, though, once you look at how an answer engine actually uses a review site.
An engine quotes the page it retrieves, and that page is rarely your score
When ChatGPT, Perplexity or Google’s AI Mode searches the web for a question like “best help desk software for a small team”, the review platform pages that match are category pages, comparison pages and lists of alternatives. They put several products side by side with short descriptions and review excerpts, which is the shape a shortlist answer needs. G2’s own announcement of its partnership with Profound, an AI search analytics vendor, puts categories at the centre: the dashboard is meant to show when a customer’s product categories are mentioned by language models with G2 as the source, and which G2 categories move the needle.
What an engine can lift from those pages is text. A 4.6 average says little about whether a product suits a 20 person agency or connects to the buyer’s CRM. A review along the lines of “we moved our six person support team off a shared inbox and were live in a week” says a great deal, and it matches the way buyers phrase qualified questions. So the parts of a profile that plausibly matter are the category you are listed under, the description you control and what reviewers write about who they are and what they used the product for.
Four review sites now belong to one company
A common recommendation is to build profiles on G2, Capterra, GetApp and Software Advice so that AI engines see several independent signals. That advice has aged. G2 announced in January 2026 that it had agreed to acquire Capterra, Software Advice and GetApp from Gartner, and said it would bring their data and offerings into its own platform. The same announcement says G2 will apply its answer and search engine optimisation playbook to lift customers’ visibility in AI driven search.
If the reviews and categories on the four sites converge, as that plan suggests, they will be less independent than four logos imply. Keep the source of the advice in mind as well. G2’s Profound announcement lists increasing recent reviews as built in guidance, and calls G2 the most cited B2B software source across language models, a claim drawn from its partner’s data. G2 also sells the review programmes. The claim may be right, but it is not neutral evidence.
Check which review pages your answers actually cite
Instead of chasing review volume, look at the evidence in your own answers. Take the buying questions that matter to you and record which review site URLs each engine cites when it answers them. If the cited category page lists your competitors and not you, the problem is how your profile is categorised. If you appear but the excerpt describes a use case you no longer serve, the problem is your description and the age of your reviews. If no review site is cited for that question at all, reviews are not the lever, and the effort belongs elsewhere.
Obility’s Answer Engine Insights keeps the answer, the competitors it names and the sources it cites with each observation, so the review pages behind a shortlist are visible question by question and engine by engine. The engines covered for your programme are confirmed at the demo. When a cited profile says something about you that is out of date, FactCheck & Sentiment records the answer, engine, date and source as one item with an owner, so the fix to the profile can be rechecked later.
Two limits apply. Review profiles matter most in software categories where review platforms rank for the buyer’s question; in many service and consumer categories they never appear among the sources. And asking customers for honest reviews that describe their situation is fair, while buying reviews or rewarding only positive ones breaks the platforms’ own rules and would not change what the engine can quote anyway.