- Two real Kevin Indig/G2 studies look contradictory on the surface. One (learn.g2.com) found total review count barely predicts citation volume: 10% more reviews, only about 2% more citations, across 500 categories and 127,554 products. The other (Jun 2026, sell.g2.com, already cited elsewhere on this site) found a genuine review burst produces a measurable citation lift in a median of four days.
- They're not contradictory. They're measuring two different variables. One is lifetime review count. The other is recent review velocity. G2's own documented scoring mechanics explain why velocity dominates: an exponential decay function, compounding daily to roughly -3%/month, means an old review carries a fraction of the weight of a recent one.
- A review holds its strongest scoring weight for its first 90 days, decays more sharply past 18 months, and retains around 3% of its original weight after three years, unless the reviewer updates it, which resets the decay clock.
- The practical takeaway inverts the naive "just get more reviews" advice: a large but stagnant historical review count is being actively discounted by the platform's own scoring mechanics, while a concentrated recent burst is not.
- This is the full-mechanics piece G2/Capterra profile hygiene points to but doesn't cover. That piece is deliberately scoped to profile completeness, not review generation.
Two real, disclosed-methodology studies from the same researcher, on the same platform, look like they disagree. G2/Capterra profile hygiene and the five stages for winning in the AI search era already cite one of them: Kevin Indig's finding that a genuine review burst produces a citation lift in a median of four days. A second, less-cited Indig study on the same platform found something that sounds like the opposite, that total review count barely predicts citation volume at all. Both studies hold up on their own terms. G2's own documented scoring algorithm is what reconciles them, and once you see how it works, it changes what a review-generation strategy should actually chase.
Finding one: does lifetime review count matter?
A Kevin Indig/G2 analysis (30,000 AI citations traced across 500 software categories on G2, drawing on 127,554 tracked products over 180 days) measured the relationship between total review count and total AI citation volume at the category level. The result: categories with 10% more reviews had only about 2% more citations, a regression coefficient of roughly 0.097. Reviews explain under 2% of the variance in how often a category's products get cited. The remaining 98%-plus comes down to brand authority, content quality, organic search visibility, and what a model's training data already contains, factors that have nothing to do with how many reviews a product has accumulated.
Read on its own, that finding would argue review count is close to irrelevant for AI citation. That's not the right conclusion. It's the wrong variable being tested for the question "does chasing reviews work."
Finding two: does a review burst produce a fast, measurable lift?
A separate Indig study, published on G2's sell-side blog in June 2026, tracked 2,514 G2 products that had a genuine review burst (five or more published, non-stub reviews landing inside a 14-day window) against a matched control group of similar products without one. The median time from the burst to a measurable citation lift (defined as citations clearing 1.5x baseline): four days. The same body of research found the review-count effect within this burst-based framing dwarfing the paid-listing effect: going from zero to 500+ reviews lifted median citations by roughly 2,100x within the free tier alone, while a paid listing added a comparatively modest 1.5-2.5x multiplier on top, once review volume and company scale were controlled for.
Read next to the first finding, this looks like a straightforward contradiction: how can review count both barely predict citations (coefficient 0.097) and produce a 2,100x lift?
How do you reconcile these two findings?
The two studies aren't measuring the same thing. The volume study measures a static snapshot: total reviews accumulated to date, compared against total citations to date, across categories. The burst study measures a dynamic event: a concentrated burst of recent activity, compared against a matched baseline, tracked forward in time. A category can have a large total review count built up slowly over several years and show almost no relationship between that count and current citation volume, while a product experiencing a genuine recent burst shows a fast, large, measurable lift. Both can be true simultaneously, because "how many reviews exist in total" and "how much recent review activity just happened" are different variables, and G2's own documented scoring mechanics explain exactly why the recent one carries far more weight.
G2's Research Scoring Methodology documentation describes an exponential decay applied to review weighting, compounding daily to roughly -3% a month, or about -30% a year. The practical timeline: a review holds its strongest scoring weight for its first 90 days, continues to hold reasonably well through about 18 months, then decays more aggressively, retaining only around 3% of its original weight after roughly three years. If a reviewer updates their existing review, the decay resets and the review regains its full weight. Recency, not the review's original publish date, is what the mechanism actually tracks.
That decay function is the missing variable that reconciles the two studies. A large historical review total, most of it more than 18 months old, is being actively discounted by G2's own scoring, which is consistent with why raw lifetime count shows such a weak relationship to current citation volume in the volume study. A concentrated burst of new reviews, by contrast, is exactly the kind of activity the decay function weights most heavily, consistent with why it produces a fast, large, measurable lift in the burst study. The two findings aren't in tension. One measured a stock; the other measured a flow, on a platform whose own scoring mechanics are explicitly built to reward the flow over the stock.
What does this change about a review-generation strategy?
The naive version of "reviews help AI visibility" implies a steady trickle of reviews accumulated over time should work about as well as the same total delivered in a concentrated window. The mechanics above say otherwise: a large, aging review base that stopped growing is being discounted in real time by the platform's own decay function, while a genuine, concentrated burst (five or more real reviews inside roughly two weeks) is the pattern with actual measured citation-lift evidence behind it. Getting existing satisfied customers to leave a review now, together, rather than whenever they happen to think of it, is a materially different (and better-evidenced) lever than a slow, indefinite request campaign.
This also means an existing large review count isn't wasted. Updated reviews reset their own decay clock, so re-engaging past reviewers to update a review (a new use case, an updated rating, additional detail) is a legitimate way to refresh weight on an existing base rather than needing every review to be brand new.
Frequently asked questions
Do more reviews mean more AI citations?
Barely, in aggregate. A Kevin Indig/G2 analysis of 30,000 AI citations across 500 G2 categories (127,554 products, 180 days) found categories with 10% more total reviews had only about 2% more citations, a regression coefficient of roughly 0.097. Total review count explains under 2% of the variance in citation volume; the other 98%-plus comes from other factors entirely.
Then why does a review burst produce a fast citation lift?
Because velocity and recency are a different variable from lifetime count, and a separate Indig study measured that variable specifically. Tracking 2,514 G2 products with a genuine review burst (five or more new, non-stub reviews inside a 14-day window) against a matched control group, the median time to a measurable citation lift (clearing 1.5x baseline) was four days. A burst is a concentrated recency signal; a large total count built up slowly over years is not the same thing, even on the same platform.
How does G2's own scoring treat old reviews?
G2 documents an exponential decay applied to review scores, compounding daily to roughly -3% a month, or about -30% a year. A review holds its strongest weight for its first 90 days, decays more sharply after about 18 months, and retains only around 3% of its original scoring weight after three years, unless the reviewer updates it, which resets the decay entirely.
What does this actually mean for a review-generation strategy?
Chase recent bursts, not lifetime totals. A large historical review count that stopped growing gets discounted by the same recency mechanics that make a burst effective: old reviews are worth a fraction of their original weight. Five or more genuine reviews landing inside a two-week window is a better-evidenced lever than slowly accumulating reviews at a steady trickle over years.
Sources referenced
- G2/Kevin Indig: Do More G2 Reviews Mean More AI Visibility? Insights from 30k Citations (500 categories, 127,554 products tracked, 180-day window)
- G2/Kevin Indig: How long does it take for software reviews to become visible in AI answers? (Jun 2026, 2,514 products)
- G2 documentation: Review Refresh
- G2 documentation: Research Scoring Methodologies
This piece is the full-mechanics deep dive behind the review-velocity point made in G2/Capterra profile hygiene and the five stages for winning in the AI search era. For the profile-completeness lever this doesn't cover, see G2/Capterra profile hygiene directly.
