How we review

What that adds up to

Review sites usually advertise how many reviews they hold. That number says nothing about whether any one of them helps you decide. These are the figures we can show instead, measured across all 528 published reviews on 10 October 2026.

Hands-on tests are the exception, not the rule: 33 of 528, and every review says which kind it is. The month-over-month check runs on our Price Watch, which archives the pricing pages every month.

Who does what

We use AI, and we would rather say exactly how than leave you guessing. Most of the reading and writing on this site is done by an AI assistant, Claude (made by Anthropic), working under fixed rules. Daan Haket, who owns the site, sets those rules, chooses what gets tested, creates the test accounts and runs some tests himself. He does not read every review before it goes live.

• The AI assistant reads the vendor’s own pricing page in a real browser, records plans and prices with the date, builds the dossier, writes the review and runs many of the hands-on tests, failures included.
• The rules are what you can hold us to: no price without a literal quote from the vendor’s own page, no “tested hands-on” label without a recorded test, no Price Watch entry whose evidence has gone. The build stops when one of them is broken.
• You can check every figure: each one carries its source and the date it was read. When something is wrong, tell us and it goes on the corrections log.

The method, step by step

1. Live pricing research. We open the vendor's own pricing page (rendered in a real browser, so JavaScript-built pages and regional pricing don't fool us) and record tiers, meters, billing conditions and trial terms — with the year attached.
2. The dossier. Every tool gets the same five-part treatment: real pricing · the honest knock (the catch the sales page won't tell you) · who it's genuinely best for · who should skip it · and the natural head-to-head comparison.
3. Evidence screenshots. Reviews carry dated captures of the vendor's own pricing page and product — shown as evidence for the figures above, marked with when we took them, and refreshed automatically.
4. Bottom line up front. Every review opens with the verdict in two sentences — worth it for whom, skip it when — so you can decide in ten seconds whether to read on.
5. The same dossier everywhere. Our video reviews, comparison pages and the Honest Software Atlas are all generated from the same researched dossier — one source of truth, no channel where the story conveniently changes.

What happens after we publish

A review is only as good as the day it was written. Most of what we do to keep one true runs after publication, and none of it depends on anyone remembering to check.

1. Monthly captures. Since July 2026 we archive the pricing page of every tool we cover — 397 of them as of August 2026 — once a month. We keep the price lines exactly as they appeared, not a summary of them, plus a fingerprint of the page text. Each month is then compared against the one before.
2. A candidate is not an entry. The comparison finds movement; reading the raw lines decides what it means. In the first round, July against August 2026, it flagged 24 candidates and 9 became entries on the Price Watch. The other fifteen were a currency selector set differently, pages that failed to load properly, annual rates disclosed for the first time, a feature change with no price change — and one pricing page that was announcing a funding round. A script cannot tell those apart from a price rise. That is why the raw lines from both months are read side by side before anything is published, and the entry keeps them on the page.
3. Dates instead of adjectives. Every dossier carries the date the price was last read against the vendor's own page; 484 of 528 were re-read since 1 September 2026. A rebuild is not a new reading: the date only moves when the page is read again.
4. The build refuses to publish a claim whose evidence is gone. Every entry on the Price Watch is tied to a literal sentence in our own dossier. If later research rewrites that sentence away, the page does not build — it stops, names the entry, and waits for a person. This is not hypothetical: on 22 August 2026 a re-verified dossier no longer carried the sentence behind one of our entries, and the registry stayed frozen until we established that the finding had been rewritten rather than retracted, and restored it.
5. Researched, and we say so. We cover hundreds of tools and cannot run all of them. What we do is read the vendor's own pages, weigh what users report, and keep the two clearly apart. Where we have not used something ourselves, we do not write as though we have — the synthesis is the value, and dressing it up as hands-on time would cost you the ability to tell the difference. When we do test hands-on, we write the checks down before the first attempt and publish every result, failures included — for example One booking app, three AI builders, where Lovable, Replit and Emergent each built the same app from one prompt against 18 fixed tests.

What we never do

• No invented numbers. If a price isn't public, we say so — we don't guess.
• No laundered star ratings. Where a review quotes G2, Capterra or Trustpilot, the score is shown as attributed third-party context with the month we read it — never as our own verdict. Our claims carry their own evidence.
• No pay-to-say. Vendors can't buy a verdict, a ranking, or the removal of an honest knock. Marketplace listings by vendors are always labeled as vendor-submitted.
• No vendor copy-paste. Partner announcements are verified against the live product before a word reaches a review.

Corrections — public, dated, fast

When we get something wrong, we fix it quickly and say so on the page, with a dated correction note — not a silent edit. Every correction is also listed in one place: the corrections log. Vendors are welcome to flag inaccuracies; the fastest way is a reply through any of our partner channels or the contact route on our About page. Recent example: when a vendor pointed out our pricing framing was outdated, the review, its screenshot and the video were corrected within a day — with the note to prove it.

How we make money — and why it doesn't bend the verdict

Reviews contain affiliate links: if you buy through them, we may earn a commission at no extra cost to you. Every review discloses this. The reason it doesn't bend the verdict is structural: our “skip it if” sections and critical knocks apply to paying partners exactly as to anyone else — you'll find them on every single review, including our best-earning ones. An honest “skip it” today is why you'll trust the “worth it” tomorrow.

The facts behind that, as of 29 September 2026:

• Most of the tools we review can pay us. 438 of the 484 tools we review have an affiliate programme with us. The other 46 get the same research, the same page and the same verdict format; their links are plain links that earn us nothing.
• Each review says which kind it is, at the top. A transparency line directly under the title tells you whether the links on that page can earn us a commission, before you read the verdict. The same note sits next to the Visit button and at the foot of the page.
• No money, no mark. When a programme ends or turns us down, the tracking link comes off the review and the commission wording goes with it; the review itself stays.
• Marketplace pages say who wrote them. A tool page in our marketplace states whether the vendor has claimed it. An unclaimed page is our research only, and says so.

Reader reviews

Readers can leave their own experience on every review page. Critical reviews are published — our only filter is spam. We don't aggregate them into a score while the volume is small, because a rating built on three opinions is a number pretending to be data.

Who does this

The research and every verdict come from Daan (Daniel Haket) — twenty years in international banking (trade finance, operations, risk & control) before building this platform on the tools he actually runs. The ex-banker filter is simple: does a tool remove a real cost bigger than what it charges?