Short version: we buy or trial the tool, run it on real tasks, screenshot what happens, and reproduce anything negative before we write it down. Every price in a review is re-checked against the vendor’s own pricing page on the day that review is edited. We have tested nine tools and published reviews of five — the other four did not clear our bar, so there is no review.
This page exists because “we tested it” is the easiest claim in the world to make and one of the hardest to check. Below is the actual procedure, including the parts that are inconvenient for us.
How a review gets made
Testing comes first, always. A review is written from a testing session — never from a vendor’s marketing page, a press release, or our own memory of an earlier test.
- Desk research. Pricing, plan limits and company ownership are collected from official sources and written down before the tool is opened, so we notice when the product contradicts the marketing.
- Hands-on session. The tool is run on real tasks, at the tier a normal reader would actually buy. We record which tier was used.
- Capture. Screenshots are taken during the session and filed per tool and per date.
- Reproduction. Anything that looks like a flaw is re-run before it becomes a finding.
- Write-up. Findings land in a permanent testing-notes file. The review is written from that file.
- Live re-verification. Prices and plan names are checked again the day the review is published or edited.
The rules we test by
Our internal testing methodology runs to twelve principles. These are the ones that most change what you read:
We report what we saw, not why we think it happened
“The outline contained phrasing that appears word-for-word on a competitor’s page” is something we can stand behind. “The tool scraped that competitor” is a guess about machinery we cannot see from outside the product. We write the first kind and flag the second as unknown.
A negative finding has to happen twice
Criticism is the most damaging thing we can publish and the easiest to get wrong. A flaw seen once is a candidate, not a verdict — we re-run the exact step and confirm the same result before it goes in writing. This rule exists because of a real miss: a tool once appeared to ignore user-supplied products, and reproduction showed it was our own procedural error, not the tool.
Old criticism gets re-tested before it is repeated
Tools improve. A documented flaw is re-confirmed before it anchors a new or rewritten article, so we do not keep punishing a product for something it fixed months ago.
Being treated well does not soften the findings
Vendors sometimes give us trial credits or unusually helpful support. That is disclosed, and it does not change what we publish. Where a founder has been generous, we hold thank-you replies until after the honest review is live.
The evidence
We currently hold 55 dated screenshots from five capture sessions, taken while actually operating the tools. We capture pricing pages, configuration panels, generated output, error states, and anything that contradicts the marketing.
Screenshots are filed by tool and capture date and are referenced from the review they support. Where a review makes a claim that a picture can settle, we would rather show the picture.
What we do when we get it wrong
This is the part most methodology pages leave out. Findings we could not reproduce have been withdrawn, including criticisms that made a tool look worse:
- A duplicated-output complaint about one tool did not reproduce across five fresh runs. We removed it rather than leave it standing.
- A claim that a tool’s scores drifted on unedited documents turned out to be our own recording error. It was never published.
- A usability criticism was dropped once the vendor shipped a fix that resolved it.
Where an earlier version of a review was wrong, the correction is stated plainly in the review rather than quietly edited out.
Findings that cost us money
We earn commission on some of the tools we cover. That makes negative findings expensive for us, which is exactly why we publish them. Recent examples, all from hands-on sessions:
- One tool’s AI-visibility feature presented domains as cited sources that do not exist — they fail to resolve. Reproduced across two runs.
- The same tool’s “authority” score turned out to measure how complete an author profile is, not how credible the sources are.
- An internal-linking feature we had previously praised regressed, returning almost nothing across five runs. We said so, and flagged that the feature is not included on the plan we otherwise recommend.
Tools we tested and did not review
We have run documented testing sessions on nine tools and published reviews of five. The gap is deliberate. A tool gets a review only when we have tested the parts that matter and can say something useful; where a flagship feature was unusable, or the paid paths remain untested, the review is held. A missing review is not an oversight.
Keeping reviews current
- Pricing: re-verified against the vendor’s official pricing page every time a review is touched. Never carried over from notes or memory.
- Hands-on re-test: every reviewed tool we hold an affiliate relationship with is re-run at least every 90 days. An underlying model upgrade can change a tool’s output quality with no announcement, and this is the only thing that catches it.
- Dated updates: reviews carry a “last updated” date that reflects real re-verification, not a cosmetic bump.
How we make money, and what it does not buy
Some links on this site are affiliate links. If you buy through one, we may earn a commission at no extra cost to you. Affiliate links are marked as sponsored in the page code.
What that money does not buy: placement, a score, a favourable verdict, or removal of a finding. No vendor has ever been given copy approval or advance sight of a review. Free trial credits or comped access are disclosed in the review itself, near the top, not buried at the bottom.
Who is accountable
Daily AI Reviews is run by Jason Novack, who is responsible for the testing methodology on this page and for the findings published under it. Reviews are bylined to the Daily AI Reviews Team because testing and write-up are a shared process, but accountability for the method sits with a named person, not a brand.
Corrections
If we have published something inaccurate — a stale price, a finding that no longer holds, a feature that has since shipped — tell us and we will re-test it. Corrections are made in the review with the change noted, not silently. Write to hello@dailyaireviews.com.
Last updated August 26, 2026.