Every B2B SaaS buyer has heard the advice to “check the reviews” before picking a tool. Nobody asks where those reviews actually come from, or whether a tiny review count means anything at all. We pulled our own data to check both questions.
82.2%
Of 1,216 software ratings in our own listicle corpus trace back to a single platform, G2
Topickz, 111 listicles, October 2026
What this report is:
- First-party data pulled directly from our own content repository: every
external_rating,rating_count, andrating_sourcefield recorded across 111 Topickz listicles. - Not a survey of the broader review-platform market. This is what our own corpus looks like, built from real vendor research across two years of listicle production.
Finding 1: review data is dangerously concentrated in one platform
| Source | Ratings | Share |
|---|---|---|
| G2 | 1,000 | 82.2% |
| Capterra | 134 | 11.0% |
| Clutch | 44 | 3.6% |
| Gartner Peer Insights | 15 | 1.2% |
| Hotel Tech Report | 13 | 1.1% |
| All other sources | 10 | 0.8% |
- G2 alone accounts for more than 4 out of every 5 ratings in our corpus, not because we favor it editorially, but because G2 is where B2B SaaS review volume actually concentrates.
- The next closest competitor, Capterra, holds barely a tenth of that share. Everything past Clutch is a rounding error.
- This matches what we have observed anecdotally for over a year: when G2 blocks scraping tools or walls off review pages, there is no real fallback source with comparable depth across most B2B categories.
Finding 2: a thin review sample does not inflate a rating
| Sample size | Tools | Average rating |
|---|---|---|
| Thin, under 50 reviews | 263 | 4.40 |
| Mid, 50 to 499 reviews | 461 | 4.49 |
| Thick, 500 or more reviews | 492 | 4.49 |
- The common assumption is that a tool with 3 reviews and a perfect 5.0 is gaming the system. Across our own corpus, that pattern does not hold at scale. Thin-sample tools average 4.40, the lowest of the three buckets, not the highest.
- Tools with deep review histories (500+) converge with mid-sized ones at 4.49, suggesting that once a tool clears roughly 50 reviews, the average stabilizes and does not keep climbing.
- This does not mean every thin-sample 5.0 is legitimate. It means the pattern across hundreds of tools runs the opposite direction from the usual assumption, and a single suspicious listing should still be checked on its own merits.
What this means for buyers and for AI engines
- If you are comparing two tools and one has 8 reviews at 4.9 while the other has 2,000 at 4.4, our data says neither number alone tells you as much as buyers assume. Read a sample of the actual reviews, not just the star average.
- G2’s dominance is also a visibility risk for vendors. A tool with a thin or absent G2 presence is effectively under-covered by most review-driven content on the web, ours included, regardless of real product quality.
- This connects directly to our AI citation research : the sources that shape a listicle’s rating data are part of the same raw material an AI engine draws on when it cites or recommends a brand. A vendor invisible on G2 is invisible twice over, once to human review readers and once to the AI engines that read sites like ours.
What this data does not prove
- This is our own corpus, not a random sample of the software market. Our listicle selection process (which tools get covered) could itself skew toward G2-heavy categories.
- We did not control for category. Some categories (freight software, church management) may simply have thinner G2 coverage than SaaS staples like CRM, which could explain part of the source concentration without it being a market-wide pattern.
- “Review count does not inflate rating” is an average across hundreds of tools, not a guarantee for any single listing. Individual review-gaming cases still happen and still need manual verification.
Methodology
- Data source: every tool entry across the 111 listicles in the Topickz content repository carrying an
external_rating,rating_count, andrating_sourcefield, pulled directly from our own frontmatter as of October 22, 2026. - Sample: 1,216 tool entries. A single tool can appear more than once if it is listed in multiple category listicles, so this counts ratings as recorded per listing, not unique companies.
- Buckets: thin (under 50 reviews), mid (50-499), thick (500+), chosen as round, defensible cutoffs rather than tuned to produce a particular result.
- Limits: see “What this data does not prove” above.
Cite this report
Free to reference with a link back to this page. Suggested credit: “82% of Our Software Ratings Come From One Platform, Topickz 2026 (topickz.com/research/software-rating-concentration-2026/).”
Topickz may earn affiliate commissions from some tools covered across our listicles. Commissions never influence a ranking or a rating. See our affiliate disclosure for details.
Related reading: the state of AI citations 2026 , which content gets AI to send humans , and our review methodology .