What a five-star average hides

Review sites, ours included, put an average at the top of every product. It is the number people quote and the number vendors screenshot, and it is close to the least useful thing on the page.

Read the shape, not the average

Two products both average 4.4. In the first, almost every rating is a four or a five. In the second, the ratings split into a mass of fives and a hard cluster of ones. Those are completely different products. The first is broadly adequate for most buyers. The second works extremely well for some companies and fails badly for others — and your only real question is which group you are in.

The split usually has a cause you can find in the text: a pricing model that punishes a particular usage shape, an onboarding that assumes an admin you do not have, a feature that only makes sense above a certain team size. Bimodal ratings are an invitation to read, not a warning to avoid.

Five questions to ask of any single review

1. When was it written? Software changes. A two-year-old complaint about a missing feature may describe a product that no longer exists, and a two-year-old rave may predate a pricing change that reshaped the whole value proposition.
2. How big is the reviewer, and are they you? A 400-person company complaining about permission granularity is telling a six-person company nothing at all — and vice versa.
3. What were they replacing? "Much better than what we had" is a statement about the old tool. Reviews from people moving off a spreadsheet and reviews from people moving off a competitor rate the same product very differently.
4. Was it solicited, and when? A review requested two weeks after onboarding captures enthusiasm. One requested at renewal captures a decision. Both are legitimate; they are not the same signal.
5. Does it describe a task, or a feeling? "Intuitive" tells you about the writer. "It took us three days to import 4,000 contacts because the CSV mapping does not handle multiple emails per record" tells you about the product.

One-star reviews are the most informative

Not because they are more honest — angry reviews are often unfair — but because of what they cluster on. Read fifty of them across a category and the same four subjects come up: billing surprises, support response times, migration and export, and reliability. Those are precisely the things a free trial cannot show you, because a trial is short, low-stakes and staffed by a sales team that wants you to succeed.

Feature complaints in one-star reviews are usually mismatches rather than defects. Process complaints are usually real.

Trials tell you about the product. One-star reviews tell you about the company. You are buying both.

How review sets get distorted

• Timing of the ask. Prompting at the moment of successful setup produces a different distribution than prompting at random.
• Gating. Asking "how happy are you?" first and only routing the happy answers to a public form is the oldest distortion in the category, and it is prohibited when the result is presented as an unfiltered picture.
• Incentives. A voucher for a review is not automatically illegitimate, but it must be disclosed, and it changes who bothers to write.
• Sample size. Twelve reviews is an anecdote. The average of twelve reviews is a very confident-looking anecdote.
• Segment mismatch. A page dominated by enterprise reviewers, read by a five-person team, is worse than no information.

What the rules actually require

Under the EU consumer rules as amended by the Omnibus directive, a trader who gives consumers access to reviews of products has to say whether and how it ensures the reviews come from people who actually used the product. Presenting fake reviews as genuine, or paying someone to submit them, is prohibited outright. In Poland, UOKiK is the authority that acts on this.

Practically, that means every serious review platform should have a page telling you what it verifies. Ours is our review policy. If a site does not have one, the absence is the finding.

Turning reviews into a shortlist

1. Filter by company size and industry before you read anything.
2. Read the three-star reviews first. They are written by people who are still thinking.
3. Search the reviews for your own dealbreaker words: "export", "invoice", "migration", "support", "API".
4. Count how many independent reviews describe the same friction. Two is noise. Seven is a product characteristic.
5. Take the top three frictions into the trial as tests, and see for yourself.

Then compare on structured criteria rather than sentiment. Products like Pipedrive, Asana and Zendesk all carry high averages across the industry while suiting very different companies — which is exactly the case where an average tells you nothing and a filtered comparison tells you everything.