Do reviews affect AI answers
Less than most people expect. Reviews help when the engine happens to read a review site, and do nothing when it doesn't. In several categories I've measured, G2 and Capterra weren't cited once.
What reviews actually do
They put you on a page the engine might read.
A G2 profile is a page. If the engine pulls that page as one of its three or four sources, your name is in front of it. That's the whole mechanism. The star rating matters less than being on the page at all.
When they help
Software categories where review sites rank well. For "best project management software" or "best CRM", G2 and Capterra rank near the top, so they get read.
When the answer is about price or features. Review site profiles have structured pricing and feature tables, which engines pull easily.
As corroboration. If your own site says one thing and a review site says the same thing, the engine has two sources agreeing.
When they don't
When the engine doesn't read them. In a pet care software category I ran twenty buyer questions through, the cited sources were competitor blogs, category roundups and App Store listings. Not one review site profile.
In service categories. Nobody reviews agencies on G2. Clutch matters there instead.
As a volume game. One company I looked at had 127 reviews at 4.9 on Capterra and appeared in zero of three answers. The tools that did appear had fewer.
What matters more
The pages that mention you in prose. A category roundup, a comparison page, a Reddit thread where someone answers a question.
Those get cited more often than review profiles in most categories, and they're the ones nobody thinks to work on.
What to actually do
Check first. Run the questions your buyers ask and look at the sources listed. If review sites are cited, get your profile filled in and get reviews. If they're not, spend the effort elsewhere.
Fill the profile properly if you have one. An empty profile with reviews on it is worse than a complete one — the engine can read the reviews but has no description, pricing or feature list to pull.
Don't chase the count. Going from 40 reviews to 90 doesn't move an answer if the engine wasn't reading that page in the first place.