Data

Review sources comparison: which ones still work in 2026, measured

22 August 2026 · JustReview

Most reviews about “dead” review sources are guesswork. This one is not: the numbers below come from our own production database on 2026-08-22, where 1,375,512 active reviews sit against 4,834 live source connections belonging to 3,595 businesses. The short answer is that review sources almost never die. 94.6% of every review we hold comes from a source that imported something new within the last 30 days. What dies, constantly and silently, is the connection between a business and that source, and the failure looks identical from the outside to a source going dark.

What was measured

Every figure here is a count over two tables. One holds reviews, with the review’s own date and the date we imported it. The other holds connections: one row per business per source, carrying the authorisation and the last known import state. A connection counts as live if it still has a target URL; when someone disconnects a source we blank that field rather than delete the row, so 1,552 of our 6,386 connection rows are ghosts and are excluded throughout. That leaves 4,834 live connections spread over 62 source identifiers, and 53 identifiers that have ever produced a review.

Two dates do the work in the table below. “Newest review” is the most recent review that exists anywhere in our copy of that source. “Last import” is the last time any row from that source landed in the database. When those two dates agree, the source is healthy. When the import date is recent but the review date is old, the pipe works and the businesses on it simply are not getting new reviews. When both are old, the source or the route to it is finished.

The connection dies long before the source does

Here is the number that reframes the whole question. Of the 3,595 businesses holding at least one live connection, only 495 received a new review in the last 30 days. 158 more had something land between 31 and 90 days ago, 282 between 91 days and a year, 2,183 have received nothing at all in over a year, and 477 have never imported a single review since the day they connected. That is 60.7% of connected businesses sitting on a review section that has not moved in twelve months, while the sources those sections point at were, in aggregate, importing normally the entire time.

Split by authorisation state and the mechanism is unmistakable. Google connections in our database come to 2,212 live rows, of which 1,072, or 48.5%, currently carry a rejected refresh token. Among the 1,140 with a healthy token, 564 imported something in the last 90 days, which is 49.5%. Among the 1,072 with a dead token, four did. Four out of 1,072, or 0.4%.

Facebook is worse and it is not close. 1,540 of 1,812 live Facebook connections, 85.0%, hold an expired authorisation. The healthy remainder imported at a rate of 18.0% over 90 days, 49 connections out of 272. The expired group managed eight out of 1,540, 0.5%. Facebook page tokens lapse for reasons the business never sees: an admin leaves, a page changes ownership, an app review cycle ends. Nobody is emailed. The widget keeps rendering the last batch it received, and the older those reviews get, the more actively they hurt the page they are on.

This is the practical meaning of a review sources comparison in 2026. The interesting axis is not which source has the nicest API, it is which source can keep a connection alive for two years without a human touching it, and how loudly a tool tells you when it cannot.

Yelp: one source, two routes, opposite answers

Yelp is the cleanest natural experiment in our data because we hold both routes to it at once. 50 live connections still sit on the older native integration. Their most recent successful import is dated 2024-02-05, and across the 90 days ending 2026-08-22 they imported exactly zero reviews. The other 18 connections were migrated to a third-party collection route: 14 of them imported within those same 90 days, most recently on 2026-08-09.

Same source, same reviews on Yelp’s website, opposite outcomes, decided entirely by which pipe the tool happened to be built on. Anyone asking whether the Yelp API still works is asking a question with two correct and contradictory answers, and the 50 businesses on the older route have been showing a list frozen since February 2024 with nothing on screen to indicate it.

The identifier trap: renamed sources look dead when they are not

Etsy shows the second failure mode. We carry two identifiers for it. The legacy one has nine live connections which have imported precisely nothing, ever, and the last row that identifier ever produced is dated 2023-04-03, leaving 31,812 reviews frozen in place. The current identifier has 19 live connections, 23,094 reviews, and eight of those connections imported within the last 30 days. Etsy did not go anywhere. The route did, and the businesses left behind cannot tell the difference.

Sitejabber is the same story with a rebrand attached: four connections, last import 2023-11-19, and the platform now trades as SmartCustomer, which we wrote up separately in what the Sitejabber rebrand changes. Moment is the one genuine death in the set, 6,808 reviews frozen at 2022-04-04 and not a single connection left pointing at it.

Counted together, 31 live connections across 31 businesses point at a source that has imported nothing anywhere in our system for over a year, and another 12 point at identifiers that have never produced a single review at all. That is 43 connections which will never deliver anything, sitting in accounts whose owners believe they are connected.

Last import and newest review are different questions

Capterra is the row worth staring at. We last imported from it on 2026-06-19, so the pipe works, yet the newest Capterra review we hold is dated 2025-11-19. Nothing is broken. Those seven businesses simply have not been reviewed on Capterra in nine months. Angi is the extreme version: imports ran as recently as 2025-10-27 and the newest review is from 2022.

The distinction matters because the fix is completely different. A stalled pipe needs re-authorisation. A working pipe with no new reviews needs you to go ask customers for reviews. Tools that only show you the review list cannot tell you which situation you are in, which is why both of ours are surfaced per connection rather than averaged into a single green tick.

What we would actually connect

If you are choosing sources to display rather than auditing ones you already have, our data points at a short list. Google first, always: 2,212 of our connections, the deepest general-purpose history at seventeen years, and the only source whose reviews are read directly in the search results your customers are already looking at. Then whichever marketplace you actually sell on, because eBay, Allegro and Amazon carry by far the highest review volume per business in our corpus and their reviews are transaction-verified in a way a public profile never is. Facebook third, on the honest condition that you accept re-authorising it roughly once a year.

We built JustReview around the maintenance problem rather than the collection problem, because collection stopped being the hard part years ago. The reviews are copied into a database we control, so a source having a bad afternoon does not blank your page; every connection carries its own last-import date instead of a shared status light; and when an authorisation lapses, the account gets a re-connect prompt rather than a slowly staling widget. Marketplace sources are covered in the same account, which we walk through in putting marketplace reviews on your website, the full current source list lives in the supported sources docs, and what Google’s own APIs will and will not let a tool do is in the Google reviews API breakdown. Plans are on pricing, or you can create an account and connect one source to see its real import date rather than a guess.

The caveats, plainly. This is a census of businesses that chose to put reviews on a website, so it over-represents sources with a self-serve connect flow and under-represents anything requiring a sales call. A source with three connections tells you about those three businesses and nothing more, which is why the thin rows are labelled as such. Import dates prove our importer ran, not that a source is healthy for everyone. Roughly 14,209 review rows, almost all on one Amazon marketplace, carry an unparseable date that lands on the Unix epoch and are excluded from the history columns above. And every figure here is a snapshot taken on 2026-08-22; the honest way to read the dead rows is not “this source is gone forever” but “on this date, nothing had come through it for years”.