Data
Do reviews increase conversion? What we can and cannot prove
Probably, but nobody can tell you by how much on your site, including us. The study behind most claims, from Northwestern University’s Spiegel Research Center, found that a product with five reviews was 270% more likely to be bought than one with none, which compares reviews with no reviews on retailers’ own product pages. It says little about what a review widget adds to a site whose customers can already find you on Google. When we made our own review widget visible again in July 2026, sign-ups rose 126% in four weeks, and we still cannot credit the widget with any specific share of that, because we fixed a broken sign-up box on the same day.
The figures below come from our production database, our GA4 property and our Google Ads account, read on 21 October 2026. The two comparison windows are 29 June to 26 July 2026 and 27 July to 23 August 2026, four weeks each.
What the research measured, and what it did not
The Spiegel Research Center study is careful work that usually gets quoted carelessly. Its finding is that the purchase likelihood for a product with five reviews is 270% greater than for a product with no reviews. The same research found the effect was larger for expensive products than for cheap ones, 380% against 190%, and that purchase likelihood typically peaked for ratings between 4.0 and 4.7 and fell as ratings approached 5.0.
Three things limit how far that result travels. The data came from product pages on retailer sites, where reviews were the shopper’s main evidence about a product. The comparison is zero reviews against some reviews, which is not the decision most businesses face when they consider a widget: their reviews already exist on Google, Booking.com or Amazon, and a visitor who cares can look them up. And it was published in 2017. Repeated as “reviews increase sales by 270%”, it turns into a promise no study made.
Most other figures you will find come from companies that sell review software, and that includes us. It does not make them wrong, but a vendor rarely publishes the test that came out flat. So instead of quoting another benchmark, here is the cleanest test we have run on ourselves, and why it still does not settle the question.
Our accidental experiment: a widget most visitors never saw
Our homepage shows a carousel of reviews about JustReview, pulled from Facebook, Capterra and G2 by our own widget. In July 2026 we found out that most visitors had not been seeing it. Our consent banner ran in an automatic blocking mode, which rewrites any script it does not recognise as essential so that it stays switched off until the visitor clicks “accept”. Our widget script was one of those. Because the blocking depended on a race between the page and the banner, a single test usually passed. Repeated fresh visits showed the truth: the widget script was blocked on 6 of 8 visits to our homepage and on 5 of 8 visits to our paid-traffic landing page.
We exempted it on 27 July 2026, because the widget sets no cookies on page load and loads nothing from other domains; we explain that check in our post on review widgets and GDPR. From that day the carousel rendered on every test visit. On paper this is the experiment everyone wants: the same page, almost the same reviews, roughly three visitors in four without the widget before the fix and all of them with it after.
What happened to sign-ups
| Measure | 29 Jun to 26 Jul 2026 (widget mostly blocked) | 27 Jul to 23 Aug 2026 (widget visible) | Change |
|---|---|---|---|
| Sign-ups in our database | 72 | 163 | +126% |
| Of those, accounts that connected a review source | 17 | 40 | +135% |
| Sessions recorded by GA4 | 1,182 | 1,254 | +6% |
| Sign-up page openings recorded by GA4 | 317 | 314 | -1% |
| Sign-ups recorded by GA4 | 3 | 117 | 39x |
| Google Ads clicks | 257 | 148 | -42% |
Sign-ups more than doubled, and the jump is not noise. If the underlying rate had stayed the same, a split of 72 against 163 between two equal periods would happen by chance roughly once in 700 million tries. The new accounts were not empty shells either: 40 of the later 163 connected a review source, against 17 of the earlier 72, so the share of accounts that actually started using the product held at about a quarter. Something real changed in the last week of July.
Paid traffic did not cause it. Google Ads clicks fell 42% between the two periods as we cut the budget ahead of pausing the account in mid-August, and we ran no Meta ads in either window. Total sessions in GA4 moved by 6%.
Why we cannot credit the widget
The widget was not the only thing that changed. In the seven days from 27 July we pushed 325 commits across our landing page, app and backend repositories. Among them:
- The main sign-up box was broken. On 44 landing pages, typing your domain into the field at the top and pressing Enter threw a JavaScript error and did nothing. We fixed it on the evening of 27 July, about 18 hours after the widget.
- The blog went live on 26 July, with a review demo and a sign-up link on every post.
- A Polish homepage launched on 28 July, and the pricing buttons started carrying the chosen plan into sign-up.
- The sign-up form stopped showing raw error messages on 31 July.
- Even the reviews changed. The day before the fix we had added G2 reviews to the carousel, which pushed most Facebook reviews out of the top 20 it shows.
- The consent fix unblocked more than the widget. The same blocking had also switched off our carousel library, the image lightbox and the general page scripts.
Any of these could explain a doubling, and the one number that points somewhere points away from the widget. If the reviews had persuaded more people, more of them should have opened the sign-up form. GA4 recorded 317 openings of the sign-up page in the four weeks before and 314 in the four weeks after. GA4 only sees visitors who allow it, so this is a hint rather than proof, but it suggests the gain came after people reached the form, in a step a carousel on the homepage is unlikely to influence.
So the honest summary of the cleanest experiment we have is this: the widget went from mostly invisible to visible, sign-ups more than doubled, and we can prove each of those facts but not the link between them.
The analytics trap
There is a second lesson in that table, and it is the one most likely to mislead you. Our GA4 property recorded 3 sign-ups in the four weeks before the change and 117 after. Read from a dashboard, that is a 39-fold lift. It is almost entirely measurement. From early June the account creation event had stopped reaching GA4, so over seven weeks it recorded 4 sign-ups while our database recorded 138. The event started arriving again in the last week of July, the same week as everything else.
Attribution moved the same way. Sign-ups carrying a Google Ads click ID doubled from 12 to 25 while ad clicks fell 42%. The ads did not get better: two days before the widget fix we had stopped the consent banner from blocking the script that stores the click ID, so clicks that had always been there finally got recorded.
If your before-and-after comparison relies on analytics, first check that the analytics itself did not change between the two periods. Ours changed at least three times in one week, and each change looked like a result.
How much data a real test needs
The cure for all of this is a proper split test: half the visitors see the widget, half do not, at the same time, so every other change hits both groups equally. The catch is volume. The number of conversions you need barely depends on your conversion rate. It depends on how small an effect you want to detect.
| Lift you want to detect | Sign-ups needed, both groups together | Time at our pace of 68 a week |
|---|---|---|
| +5% | about 12,500 | about 185 weeks, three and a half years |
| +10% | about 3,300 | about 48 weeks |
| +20% | about 900 | about 13 weeks |
| +50% | about 190 | about 3 weeks |
| +100% | about 65 | about 1 week |
The table assumes two groups of equal size, 95% confidence, 80% power and a 5% base conversion rate; at a 2% base rate the totals are about 3% higher. Our pace is the average of the eight weeks to 18 October 2026.
A 20% lift from a widget would be a very good result, and we could detect it in about a quarter. A 10% lift would take most of a year with the page frozen, because any other change during the test muddies it. A 5% lift is out of reach for a business our size. Most small sites are in the same position, which is why a vendor promising a specific lift on your site is promising something neither of you can check any time soon. The flip side is that big effects show up fast: our July doubling would have been visible in a week. Seeing an effect is the easy part. Attributing it is the hard one.
What we would do on a small site
- Count conversions where they happen. Orders in the shop backend, sign-ups in the database, bookings in the booking system. Use analytics for the path, not for the total.
- Check the widget is seen before asking whether it works. Ours was blocked for three visitors in four. In our scan of 282 live homepages, the first review block sat a median of four screens down. A widget nobody reaches cannot convert anybody.
- Change one thing, then wait. Write down the date and hold other changes to the same page for a few weeks. We did the opposite in July, which is why this post is about what we cannot prove.
- Test big changes, not small ones. At small-business volumes only large effects are detectable. Reviews on a page that had none, or a rating moved next to the main button, is a test you can finish. A different carousel colour is not.
- Prefer true numbers to perfect ones. The Spiegel research found purchase likelihood falling as ratings approach 5.0. On 17 October 2026 we replaced a hard-coded “5.00” image in the top section of 67 of our landing pages with our real rating, 4.8, calculated from our Facebook, Capterra and G2 reviews. Given the table above we expect no measurable result, and we did it anyway, because the old image was not true.
For what our wider data says about review counts and ratings across businesses, see our online reviews statistics for 2026.
How we measured
Sign-ups are accounts in our production user table, read-only, grouped by creation date. An account counts as having connected a review source if it has at least one connector today. GA4 figures are event counts from our own property across justreview.co and app.justreview.co, grouped by ISO week: session starts, the event that fires when the sign-up page opens, and the account creation event. Google Ads clicks come from the weekly report of our own account. The blocking rates come from repeated fresh visits in a headless browser on 26 and 27 July 2026. The test-size table uses the standard formula for comparing two proportions. No customer’s data is used anywhere in this post.
Where JustReview fits
If you decide reviews belong on your site, the most useful thing a tool can do is make sure they actually appear, stay current and do not depend on a cookie banner. JustReview pulls reviews from Google, Facebook, Booking.com, Amazon, Capterra and other sources into one widget and refreshes them automatically, and the testimonials widget sets no cookies, which is exactly why we could exempt it from consent blocking on our own site. The setup for the most common case is on our Google reviews widget page, and the plans are on the pricing page: the free plan shows a badge with up to 100 Google and Facebook reviews, and the testimonials carousel starts at 9 EUR a month with a 7-day trial.
You can create an account and see your own reviews in a widget within a few minutes. What we will not do is promise you a conversion lift. If you test it, count the result in your own database, and if the effect is big enough to matter, you will be able to see it.
FAQ
How much do reviews increase conversion rates?
There is no single honest number. The most cited figure, from Northwestern University's Spiegel Research Center (2017), says a product with five reviews had a 270% higher purchase likelihood than a product with none, measured on retailers' own product pages; the lift was 190% for a lower-priced product and 380% for a higher-priced one. That compares having reviews with having none, not adding a widget to a site whose reputation already exists elsewhere. On our own site, sign-ups rose 126% in the four weeks after we made our review widget visible again, but we pushed 325 other commits in the same week, so we cannot say how much of that the widget did.
How long does an A/B test of a review widget need to run?
Long enough to collect the conversions, not the visits. With a 50/50 split, 95% confidence and 80% power, detecting a 10% relative lift takes roughly 3,300 conversions across both groups, whether your base rate is 2% or 5%. A 20% lift takes about 900 and a 50% lift about 190. At our own pace of about 68 sign-ups a week, a test for a 10% lift would run for around 48 weeks. Most small sites can only detect large effects, so test bold changes rather than small tweaks.
Why does Google Analytics show fewer conversions than my backend?
Usually consent and broken events. Analytics only sees visitors whose browser let it load and whose events actually fired. Between 29 June and 26 July 2026 our GA4 property recorded 3 account sign-ups while our database recorded 72, because the sign-up event had stopped reaching GA4. Count conversions where they are created, in the shop or app database, and use analytics to study the path rather than the total.