The recommendation is being made by a source you don't control
Ask ChatGPT or Perplexity for the best software in your category and watch what it leans on. It doesn't invent an opinion. It reaches for the things it treats as neutral evidence: review-site aggregate scores, category rankings, analyst grids, and the occasional roundup post. G2, Capterra, TrustRadius, and the analyst tiers show up over and over because the models read them as consensus.
That means your review profile stopped being a marketing vanity metric and became an input to an automated recommendation engine. Nobody at G2 asked whether you wanted that. Nobody sent a memo. But a 4.2 with 312 reviews sitting next to a competitor's 4.6 with 871 reviews now reads, to a model, as a tiebreaker. The higher number gets the citation. You get left off the shortlist, and you never see the query that decided it.
And here's the part that stings. The gap doesn't have to be real quality. It can be a review count that stalled two years ago, a batch of unanswered one-star reviews dragging the average, or a category page where a competitor kept collecting testimonials and you didn't. The model can't tell the difference between quiet and bad. It just reads the score.
A four-tenths gap is doing more damage than it looks
38%
In this illustrative category view, the brand wins the citation in 38% of AI answers where it gets named at all. The competitor sitting at 4.6 with more than double the review volume takes most of the rest. Closing a 4.2-to-4.6 gap and clearing the 14 open negative reviews is the difference between being the default suggestion and being the afterthought.
You can't fix what you can't attribute
Most teams already watch their G2 score. What they can't see is which sources the AI answers in their specific category actually pull from, and where a weak score is the reason a citation went to someone else. That attribution is the missing link. Knowing your score is 4.2 is trivia. Knowing that ChatGPT cites G2 in six of ten shortlist answers for your category, and that your 4.2 is the line item costing you those mentions, is a decision you can act on.
This is where Crescive comes in. It runs the real prompts buyers use, records which review-site and analyst sources feed the answers in your category, and flags where a low or outdated score is the wall between you and a citation you'd otherwise win. When an unanswered negative review or a stalled review count is dragging an aggregate that a model keeps quoting, you see it named, not guessed at. Then the fixes go through a human approval gate, and you get before-and-after evidence that the citation moved.
How to turn a review score into an AI citation you win
- Find which review and analyst sources actually feed your category. Run the buyer prompts and log which sites the assistants cite. If G2 and Capterra dominate, that's where the leverage is.
- Pull your aggregate next to the competitors that keep getting named. Note the score gap and, just as important, the review-count gap. Volume signals recency to a model.
- Clear the drag first. Respond to open negative reviews and resolve what you can. An unanswered one-star does double damage: it lowers the average and it's the first thing a model can quote against you.
- Close the volume gap with a steady review campaign, not a one-time push. A count that keeps growing reads as an active, trusted product.
- Re-run the category prompts after the score moves and confirm the citation shifted. If the answer now names you, the work paid for itself.
Treat review sites like the ranking surface they became
The old playbook treated G2 and Capterra as places to collect a badge for the website footer. That's not the job anymore. Those pages are now training data and live retrieval sources for the tools your buyers ask before they ever hit your site. A score you're ignoring is a recommendation you're forfeiting.
The good news is this is one of the more fixable AEO problems. You don't need to rewrite your whole content strategy. You need to know which sources matter in your category, close the specific gaps that are costing citations, and watch the answers change. The score was already public. Now it's just doing a much bigger job.
Key takeaways
- AI assistants cite review-site aggregates and analyst rankings as neutral evidence, so your G2 and Capterra scores are now direct inputs to whether you get recommended.
- A small score gap or a stalled review count can lose you citations even when your product is stronger, because a model reads volume and recency as quality.
- Unanswered negative reviews do double damage: they lower your average and give the model something to quote against you. Clear them first.
- Crescive shows which review and analyst sources feed your category's AI answers and where a weak score is costing a specific citation, then proves the lift after you fix it.
FAQ
Do AI assistants really use G2 and Capterra scores to recommend software?
Yes. When you ask AI assistants like ChatGPT, Perplexity, or Google AI Overviews for the best software in a category, they frequently cite review-site aggregate scores and rankings from sources like G2, Capterra, and TrustRadius as evidence. A higher aggregate score or larger review count can be the tiebreaker that earns a citation, which means your review profile now directly influences whether an AI recommends you.
How do I know if a low review score is costing me AI recommendations?
Run the actual prompts buyers use in your category and record which review and analyst sources the assistants cite, then compare your aggregate score and review count against the competitors that keep getting named. Crescive automates this by tracking which sources feed your category's AI answers and flagging where a low or outdated score is the reason a citation went elsewhere, so you can prioritize the fixes that move a specific recommendation.