G2 Reviews in AI Answers: Why the Citation Studies Range From 0.9% to 84%

One study found G2 and Capterra cited zero times across 233 ChatGPT software recommendations. Another found G2's family of sites holding 84% of all review platform citations in ChatGPT. Both are real, both were published in 2026, and neither is wrong. They count different things, and that difference is the whole story about G2 reviews in AI answers.

The short answer

Yes, G2 reviews still feed AI answers, but unevenly. Review sites are cited most on Perplexity, on explicit review and comparison queries, and at the evaluation stage of a purchase. They are cited least on ChatGPT discovery prompts such as "best CRM for startups", where vendor pages, blogs and Reddit take most of the citations.

Why G2 citation studies disagree so badly

Put the published figures side by side and they look contradictory. G2 is cited in roughly 1 in 5 product discovery searches by one count. It is cited zero times by another. It owns 84% of a category by a third.

The contradiction dissolves once you ask three questions of every study. What is the denominator? Which engine was tested? What kind of query was asked? Nearly every bad take on review sites and AI search skips at least one of them.

The denominator problem

There are at least three different things a study can be counting. The first is the share of all citations in an answer set, where review sites compete with vendor pages, Reddit, news and blogs. The second is the share of review platform citations only, where G2 competes with Capterra and TrustRadius. The third is the share of answers that cite any review site at all.

These produce numbers an order of magnitude apart from the same underlying behaviour. A review site can hold a dominant share of a small slice. That is roughly what the ChatGPT data shows, and it is why I would never quote an 84% figure without the words "of review platform citations" attached.

The engine and query problem

The second split is the engine. Google AI Overviews, ChatGPT and Perplexity retrieve differently, and the studies below show they lean on review sites to very different degrees. The third split is the query. A buyer typing "HubSpot reviews" and a buyer typing "best CRM" get answers built from different sources, even on the same engine. The mechanics of how engines pick sources in the first place are covered in the analysis of whether ChatGPT citations follow Google rank.

Six G2 citation studies, compiled with their methods

I found six studies with a stated method that report review platform citations in AI answers. None is peer reviewed. Three were published by G2 or rely on figures G2 chose to publish, which matters because G2 sells the thing being measured. The rest come from search tool vendors and an agency, who sell the measurement.

Every row is either vendor research or published by an interested party. Read them as six flashlights on one room, not six measurements of one number.

The two ChatGPT studies look like a direct conflict, and they are not. DerivateX ran one buyer-intent prompt per category, the "recommend me a tool" style of question, and recorded every citation. Review aggregators took 0.9% of those citations, Reddit and community sites 8.4%, and blogs plus vendor pages 81.9%. Indig pooled thousands of prompts across the whole buying journey, including evaluation prompts where review sites do much better.

Both findings point the same way. On a cold discovery question, ChatGPT barely touches review sites. When a buyer is already comparing named tools, review sites show up more, and when they do, it is mostly G2's domains.

Which review platforms each AI engine favours

The per-engine evidence is thinner than the confident blog posts suggest. Only one dataset compares G2's share across three engines on a common definition, and it reached the public through a G2 roundup rather than a methods paper. With that caveat, the pattern is consistent across every source that reports it.

Perplexity cites G2 most heavily

Profound defines G2's share as G2 citations divided by citations to eight review sites, including Capterra, TrustRadius, Gartner and Product Hunt. On that definition, G2 took about 33% of review site citations in ChatGPT and Google AI Overviews, and closer to 75% in Perplexity.

Radix points the same way. Its page reports an engine split of 60% Perplexity, 30% Google and 10% OpenAI. The page labels this coverage but does not define the base, and the three figures sum to 100%, so I read it as the share of G2's citations by engine. That reading is mine, not Radix's.

Google AI Overviews spreads citations across more platforms

Google is the least concentrated engine in the data. In SE Ranking's sample, Gartner Peer Insights held 26.0% of review platform links, G2 23.1%, Capterra 17.8%, Software Advice 12.8% and TrustRadius 8.3%. No single site dominates, and Gartner's own review property leads.

Placement is weak even when the citation lands. SE Ranking found the median position of a review link in the sources panel was about fifth, and only about 7% of review platform links appeared inline in the answer text. A citation in position five of a collapsed panel is a long way from a click.

ChatGPT concentrates on the G2 family

In Indig's December 2025 ChatGPT data, the G2 ecosystem took 84% of review platform citations, gartner.com 13%, SourceForge 2% and TrustRadius 1%. TrustRadius, a real presence in Google's answers, is close to invisible here.

Columns use different samples and slightly different platform sets, so compare rankings within a column, not numbers across columns.

Query type moves review site citations more than the engine

If you only remember one finding, make it this one. The wording of the buyer's question changes how often a review site is cited by a factor of almost 3, inside a single engine.

SE Ranking split its 30,000 keywords into three equal groups. Explicit review queries contained words like "review" or "rating". List queries contained "best" or "top". General queries contained "software" or "tools" with no review intent. The results diverged sharply.

The share of sources tells the same story. Review platforms made up 16% of cited sources on explicit review queries, 5.7% on general software queries, and just 2.7% on "best" and "top" queries. That 2.7% is consistent with DerivateX's near-zero result for ChatGPT discovery prompts.

Radix found a gentler version of the same gradient across engines. G2 was cited in 19.70% of product discovery queries, 15.03% of branded queries and 14.01% of comparative queries. The rank order differs from SE Ranking's, which is a reminder that "discovery" means different things in different taxonomies.

My reading is that review sites are answer sources for questions about a named product, not for questions about a category. The category answer is being built from listicles, Reddit threads and vendor pages, which is the ground covered in the Reddit and Perplexity citation playbook and in the analysis of community mentions and AI visibility.

G2 reviews get cited at evaluation, not discovery

Indig's February 2026 study is the only one that tags prompts by buying stage. It used about 35,000 Profound citations from US ChatGPT in December 2025, reduced each citation to its domain, and counted unique domains by site type within each stage.

Across all stages, review platforms took 8.6% of cited domains, user-generated platforms 17.1%, publishers 4.0% and vendor or other domains 70.4%. Review sites beat publishers. They lost heavily to community sites and to vendors' own pages.

The 13.2% evaluation peak is the most useful number in this whole literature for a budget owner. It says G2 reviews earn their keep when a buyer already knows your name and is checking whether to believe you. That is a proof job, not a discovery job, and it should be funded and measured as one. How shortlists form before that point is set out in the analysis of the AI shortlist in the B2B buyer journey.

More G2 reviews buy very little extra AI citation

The obvious vendor response to all this is to run another review campaign. G2's own published analysis is the best argument against making that the main lever.

Indig's October 2025 study took 30,000 Profound citations across 500 G2 categories. It compared approved reviews over 12 months with citations over 4 weeks, and excluded categories with fewer than 100 reviews or fewer than 10 citations. Categories with 10% more reviews had about 2% more citations, and reviews explained less than 2% of the variance.

The relationship is real but small. An R-squared of 0.009 means 99% of the difference in citations between categories comes from something other than review count. I would credit G2 for publishing a result that undercuts its own sales pitch. I would not let anyone sell me a review drive on the promise of AI visibility after reading it.

What a review campaign still does

None of this says reviews are worthless. A profile with recent, specific reviews gives an engine something accurate to quote when it does cite G2, and it gives a buyer something to read when they click through. The value sits in the content of the reviews, not the count.

A review that names the integration, the team size and the problem solved is quotable. A review that says "great tool, easy to use" is not. If you run a campaign, prompt reviewers for the specifics a buyer would ask an assistant about. Stay within the FTC's rule that incentives cannot be conditional on sentiment, covered in the analysis of the review site business after the Gartner sale.

Capterra vs G2 after the merger: does the choice still matter

For years "Capterra vs G2" was a budget question with two vendors on either side. G2 now owns Capterra, Software Advice and GetApp, so the commercial negotiation has collapsed into one counterparty. The citation question has not collapsed with it.

The domains are still separate, and the engines treat them separately. In Google's AI Overviews, G2 and Capterra held 23.1% and 17.8% of review platform links respectively, close enough that neither is optional if Google is where your buyers start. Capterra was also among the platforms SE Ranking found most often in the top three source positions.

Gartner Peer Insights is the platform most vendors underweight. It was not part of the sale to G2, it led Google's review citations at 26.0%, and it held 13% of ChatGPT's review citations in Indig's data. If your buyers are enterprise IT, I would put a Peer Insights profile ahead of a third G2 campaign.

The ChatGPT discovery row is the one most teams get wrong. If a cold "best tool for X" prompt is built from blogs and vendor pages, then the asset that matters is a page you own that answers the comparison honestly. The structure for that is in the template for B2B comparison pages.

Where the evidence is weakest

The honest summary is that this is a literature of vendor research, and the conclusions above should be held loosely. Three weaknesses are serious enough to change how you act on it.

The interested parties problem

Three of the six studies were published by G2 or reached the public through G2. G2 has an obvious interest in showing that its reviews feed AI answers. The search tool vendors and the agency have an interest in showing that AI visibility is measurable and needs their help. Nobody in this literature is neutral, including the tool every study relies on.

The G2 work deserves some credit on this score. The October regression and the low discovery-stage shares both cut against G2's commercial interest, and they were published anyway. That is not proof of rigour, but it is evidence against cherry picking.

Samples, dates and engines that do not line up

No study measures all three engines on the same queries in the same month with the same definition. The Profound per-engine figures have no published sample or date range. DerivateX ran only 40 prompts, ten times each, which is a test rather than a census. SE Ranking is a single-day Google snapshot from December 2025, almost a year old by now.

Engines also change their retrieval quietly and often. A study from December 2025 describes ChatGPT as it behaved in December 2025. I would treat every number in this post as having a half-life of months, not years.

The case that citation is the wrong metric

The strongest counter-argument is that citation counts miss the main channel. An engine can absorb what G2 reviews say about your product and repeat it without linking to G2 at all. In that world, a near-zero citation share understates review influence, and the only way to measure it is to check whether assistant descriptions of your product match your review themes. Nobody has published that test at scale. The broader measurement problem is set out in the review of the GEO versus SEO citation evidence.

Frequently asked questions

Where to start: run your own engine split this week

Every number above came from someone else's sample. The fastest way to know which applies to you is a test that takes an afternoon.

Write 20 prompts in three groups: category discovery such as "best tool for X", named comparison such as "your product vs competitor", and review questions such as "your product reviews". Run each one on ChatGPT, Perplexity and Google, and log every cited domain in a sheet. If G2, Capterra or Peer Insights show up only in the comparison and review groups, your review budget is a proof budget. Judge it by win rate on evaluated deals rather than by AI visibility dashboards.

Related reading