G2 Reviews and AI Recommendations: What Moves ChatGPT and Claude

As of October 2026, this reflects what ChatGPT, Claude, Perplexity and Google's AI features showed, and what the published studies measured.

If your G2 reviews beat a competitor's and that competitor still shows up in ChatGPT every time, more reviews will not close the gap. Review volume barely moves AI citations, and the pages ChatGPT does cite are lists, comparison pages and vendor sites. OnlyAEO is an AEO and GEO agency that measures citation share, mention rate and recall across ChatGPT, Claude, Gemini, DeepSeek, Perplexity and Grok. This is how we would answer a Series B CMO who has spent years building a review-site program and wants to know what it is worth in AI search.

Do G2 reviews help ChatGPT and Claude recommend your SaaS brand?

Yes, but indirectly: reviews help you get considered, and they almost never get you cited. The clearest evidence for both halves comes from studies published between May and August 2026, and they line up better than their headlines suggest.

What the citation studies found

DerivateX published a study on May 31, 2026 of 233 ChatGPT software recommendations across 40 B2B SaaS categories. G2 and Capterra each received zero citations, and review aggregators as a group were 0.9% of everything cited.

The method, category list and source breakdown are in the DerivateX B2B SaaS citation study.

Nathan Mzumara's August 2026 analysis looked at a different slice: 367 citations across 60 ChatGPT software buying answers. The five established aggregators (G2, Capterra, Software Advice, TrustRadius and TechnologyAdvice) took 4.1% of citations between them. They were also only 1.9% of what ChatGPT retrieved, 52 retrievals and 15 citations in total.

That second number explains the first. When ChatGPT did read a review page, it cited it 28.8% of the time, the highest conversion rate of any source type in the dataset. Mzumara's conclusion, in his breakdown of who ChatGPT actually cites, is that aggregators still have the model's trust and lose at retrieval, because the searches ChatGPT runs rarely surface them. OpenAI's own help pages say ChatGPT rewrites a question into one or more targeted queries before it searches, and "best invoicing software for agencies" tends to return listicles and vendor pages, not a G2 category page.

Why a profile still counts

The same research points the other way on eligibility. A study summarized by Strivelabs in August 2026 found that 100% of tools ChatGPT recommended had Capterra reviews and 99% had G2 reviews. Strivelabs reads this as review platforms acting as an eligibility gate: they confirm a tool exists and is described consistently, and other pages take the citation. DerivateX says much the same: review profiles shape buyer perception and conversion, but they are not the pages ChatGPT pulls from when it cites software.

Mzumara adds one caveat. In his corpus, 69.3% of brand recommendations arrived with no evidence attached, and review data may still be doing work there, inside what the model already believes. DerivateX found only 7.7% of its recommendations had no source, and that gap most likely reflects a difference in method. Neither dataset can settle it.

Our read: a missing or thin G2 profile can keep you out of the candidate set. A strong one probably gets you to the gate, and on this evidence it rarely walks you through.

Why does a competitor with worse G2 reviews show up in ChatGPT every time?

Because the competitor is either already in the model's memory or sitting on the third-party pages ChatGPT retrieves for your category, and neither depends on review scores. DerivateX found that established category leaders are often named from model memory with no citation at all, while lesser-known tools get cited from fresh third-party listicles fetched at answer time.

The second route is the one you can act on. When ChatGPT recommends a B2B SaaS tool, it cites that tool's own website only 11.6% of the time, and credits a third party the other 88.4%. In 25 of 40 categories, no recommended vendor was cited through its own site.

That looks like it conflicts with Mzumara's finding that vendor sites took 65.9% of citations. It does not. DerivateX measured whether the recommended tool's own site backed its own recommendation. Mzumara measured whether a citation pointed at any vendor's website, including a vendor's "best tools" list that recommends its rivals. DerivateX's best example: one Procurify list of procurement software was the cited source for five different brand recommendations in a single answer. A vendor can own the answer for a whole category by publishing the page ChatGPT cites for everyone else.

Visibility is also won one category at a time. Of the 219 tools DerivateX saw named, 94% appeared in only one category. If your competitor wins "expense management for startups" and you win nothing, we usually find a page problem in that one category long before we find a brand problem. We walk through the full diagnosis in what a Series B CMO does when a competitor owns every AI answer.

How much does review count actually move AI citations?

Very little, and the two published versions of the same analysis disagree on exactly how little. Strivelabs reports a G2 analysis of 30,000 citations across 500 software categories: categories with 10% more reviews had 2% more citations, but the R-squared was 0.009, so review volume explains under 1% of the variation in citations.

G2's own page on AI citations, published in November 2025 and written by G2 about G2, attributes the same 30K-citation analysis to Kevin Indig and reports the same 10% to 2% correlation. It then says reviews explain about 2% of variance in AI visibility, set against brand authority and content quality. Under 1% or about 2%: the sources do not agree on the size, and neither figure supports a review-volume push as an AI visibility plan.

G2 publishes its own figures too. Read them as vendor claims. On that page, G2 cites Radix data from 10K+ searches putting G2 at a 22.4% share of voice for software queries, and says G2 pages are cited in 1 out of 5 product discovery searches. G2 also says its category pages and best-of listicles are what usually get cited, with actual reviews increasingly quoted word for word. That last point should shape what you ask reviewers to write (see the table below).

Then there are G2's buyer surveys. Strivelabs notes that the finding that 51% of B2B buyers start in an AI chatbot comes from G2's own survey of 1,076 buyers, and G2's August 2025 survey found 87% of B2B software buyers said AI chatbots are changing how they research. Both can be true and still come from an interested party. Mzumara's advice is blunter: stop treating G2 rank as your AI visibility metric, since review sites earned 15 of 367 citations in his data.

Do ChatGPT, Claude, Perplexity and Google treat review sites the same way?

No, and the differences are large enough that one blended "AI visibility" number tells you very little. Strivelabs puts the conflict plainly: one analysis found review platforms in 34.5% of Google AI Overview responses, while another found zero G2 or Capterra citations across 233 ChatGPT recommendations.

Two rows in that table lean on a Reddit post by the founder of Sanbi.ai, a tracking vendor, who followed 42 B2B brands for 74 days and found only 19% appeared consistently in all four engines. It is a single vendor's dataset, not a peer-reviewed study. Its source mix also disagrees with Mzumara's: the founder found Reddit threads were 22% of cited sources against about 6.5% for brands' own sites, while Mzumara saw community content retrieved 31 times and cited 0 times. When studies split this hard, measure each engine yourself.

Google also says AI Mode and AI Overviews may use different models, so the links they show will vary. If your board deck has one number for "AI search," split it.

What review-site work is still worth doing?

The work that makes your profile accurate and your reviews specific is worth doing; the work that only raises review count or profile tier is not, at least for AI visibility. Here is how we would sort a typical review-site budget.

Most rows in that table come from Strivelabs' review of G2 and Capterra in AI answers, which is worth reading in full before your next budget review.

One structural change affects how you plan this. G2 acquired Capterra, Software Advice and GetApp from Gartner in February 2026, so three of the five aggregators in Mzumara's list now sit under one owner. Spread effort across those three and you are spreading it across one company's data.

What actually changes whether ChatGPT and Claude cite you?

Three things move citations more than review work: being on the third-party pages engines retrieve, publishing pricing on your own site, and making sure AI crawlers can read you. Each is cheaper to test than a review campaign.

Get onto the lists and comparison pages engines retrieve

In DerivateX's data, independent and niche blogs plus vendor content were 81.9% of ChatGPT's software citations, major media 8.8% and community content 8.4%. Every cited page used list structure, 78% had a year in the title or headline, 68% had a comparison table and 56% had an FAQ.

That gives you two jobs. Get included, accurately, on the independent "best X for Y" pages that already rank for your category, which is earned media work; our guide to the earned media sources AI engines cite covers how to find them. And publish your own honest comparison and category pages in the shape those cited pages share. Google's AI features use query fan-out, running related searches across subtopics, so a page per real use case gives the engine more ways to find you than one long features page. How that third-party work stacks with your own pages over time is the subject of our AEO channel mix guide.

Publish your pricing

Mzumara found that categories with public pricing had vendor citation shares of 80-91%, while categories with gated pricing sat at 48-55%. That is a correlation across categories, with no controlled test behind it. It is also the largest gap in any of the studies here. If "contact sales" is the only answer to "how much does it cost," an engine has to cite someone else to answer the buyer.

Let the crawlers in

None of the above matters if ChatGPT cannot read you. OpenAI says a site must allow OAI-SearchBot to be eligible for ChatGPT search, and sites opted out of it will not be shown in ChatGPT search answers. After a robots.txt change, OpenAI says its search systems can take about 24 hours to adjust.

For Google's AI Overviews and AI Mode, a page must be indexed and eligible to show with a snippet. Google adds that no special schema.org markup is needed for its AI features, so treat markup as a way to state facts clearly, as our guide to structured data for AEO explains, and not as an entry ticket. Check your robots.txt and your CDN's bot rules before you spend anything on content. We check it first because, in our experience, it is usually the cheapest fix on this page.

How do you test whether review-site work changed your AI citations?

Run a fixed prompt set before and after each change, repeat every prompt, and record the cited source type, not just whether you appeared. Strivelabs suggests twenty category prompts through ChatGPT, Gemini and Perplexity, recording whether you appear and whether a review site is the cited source. We would add Claude, because its citations overlap with ChatGPT's but are not the same set.

In our experience, repetition is the step teams skip. The Sanbi.ai founder re-ran the same prompt on the same engine within a 24-hour window and saw the brand mention set change in 38% of cases. A single run per prompt will show you noise and call it a result.

A test plan you can run this quarter:

1. Write the prompt set from real buyer language: category queries, "best X for Y," "X vs Y," and pricing questions.
2. Baseline: run each prompt several times per engine and log whether you appear, whether you are cited, and which source type carries the citation (vendor site, list, review site, community, media).
3. Change one thing: fix the G2 profile's category, integrations and pricing, or start a specific-review ask, or publish pricing, or get onto one independent list.
4. Wait for the change to be crawled, then re-run the same prompts the same number of times.
5. Compare appearance rate and cited source type per engine; do not average across engines.
6. Keep what moved citations and drop what did not, then test the next change.

If step 2 shows review sites almost never carry the citation for your prompts, your review budget belongs in conversion, not AI visibility. That is a useful answer too.

How to choose a partner for this work

Pick a partner that measures citations per engine, repeats prompts, and can tell you which source carried each citation. A partner that only reports "mentions" cannot tell a review-site win from model memory. Ask how they separate the two before you sign anything.

OnlyAEO measures citation share, mention rate and recall frequency in monthly reports across ChatGPT, Claude, Gemini, DeepSeek, Perplexity and Grok. Our free AI visibility audit is delivered within 48 hours and covers ChatGPT, Claude, Gemini and DeepSeek. Most clients see first measurable citation improvements within 60 days, and FastTrackr AI went from 0.1% AI visibility to the #3 most-cited brand in its category in about 100 days. For the full owned and third-party program around this, see our guide to getting a B2B SaaS brand recommended by ChatGPT and Claude.