How we check a tool before it gets listed

Most "review" sites in this niche publish a list they copied from a press release, add a five-star score nobody can explain, and monetise the outbound link. This page exists to be the opposite of that: an explicit statement of what I actually check before a tool appears here, and — just as important — what I refuse to pretend I measured.

Read it before you trust anything else on the domain. If a profile ever contradicts this page, the profile is wrong and I want to know.

What a profile must contain

Every entry in the profile index is written to one of two structures: a long verdict-first profile for the tools most people are deciding between, and a shorter one for the rest. Neither structure has a slot for a field I cannot fill honestly — if a field can't be filled, the tool doesn't get listed yet.

• What it is, in plain terms — one paragraph that would still be true if the marketing site disappeared.
• Where it wins and where it falls short — at least three points each side. A profile with no cons is an advertisement.
• Who it fits and who should skip it — including the situations where a free or cheaper option is good enough.
• What it costs — the tiers as the vendor currently publishes them, stamped with the month I checked, phrased as an indicative ballpark rather than a quote.
• A 30-minute test plan — the checklist this desk works from when it writes a profile, handed to you so you can reach your own verdict on your own work.
• Alternatives — real competitors, linked, not straw men.

The checklist behind every profile

The same eight questions run through every tool, whatever category it's in. They're deliberately boring, and they're the questions that predict whether you'll still be using something in month three.

1. Does it do the one job you're buying it for, on your material? Not on the vendor's example — yours.
2. What happens when the input is messy? Real work is badly labelled PDFs and half-finished notes. Tools break in the same order every time.
3. Can you get the output out? Export formats, lock-in, and whether your work survives if you stop paying.
4. What does the free tier actually limit? Volume, model quality, features, or just speed. This decides whether you can trial it properly.
5. How does it fail? A tool that fails loudly is safer than one that fails smoothly and confidently.
6. What are you agreeing to on data? Training use, retention, sub-processors, and whether a business tier changes the answer.
7. Who owns the output, and can you use it commercially? Especially for generated images, voice and music.
8. What's the realistic total cost? Seat price plus credits plus the thing you'll buy next because this one stops short.

What we do not do

This is the part that makes a directory trustworthy, so it gets its own list.

• No invented benchmarks. I don't publish accuracy percentages, generation counts or "10x faster" claims, because I have not run a controlled evaluation and a hand-made one would be theatre.
• No star ratings out of ten. A number implies a measuring instrument. What you get instead is a written verdict and a test plan you can run in half an hour on work that actually resembles yours.
• No affiliate links, no sponsored placements, no paid inclusion. Outbound links to vendors are plain nofollow links. If that ever changes on a page, the page says so at the top.
• No press-release copy, and no claimed experience. Product descriptions here are written from documentation and published pricing read on the vendor’s own pages — not pasted from a vendor, and not described as used, trialled or tested, because nothing here has been.
• No listing everything. A directory that adds a tool to add a tool is noise. Categories stay deliberately incomplete.

How pricing is checked

Prices on this site are stamped with the month they were last read against the vendor's own pricing page, and written as an indicative ballpark ("about $20/month") rather than a quotation. That's on purpose: regional pricing, annual-versus-monthly billing, credit systems and tier renames change faster than a static site can track, and a precise wrong number is worse than an honest approximate one.

When you see a price here, treat it as the figure that was true when I checked, and confirm it on the vendor's page before you pay. When you see a qualitative cost instead — "per-seat", "credit-based", "usage-billed" — it means the vendor doesn't publish a simple number and I'm not going to invent one. Errors in pricing are the most common thing I get told about, and they get fixed and logged on the updates page.

That month is not a maintenance field. It moves forward only when someone has actually re-read that vendor's pricing page. Refreshing it to make a warning disappear would be a fabricated record, and a fabricated record is worse than a stale one: the stale date is a fact you can act on, while a false fresh one is a guess dressed up as evidence. So some profiles will admit that their reading is more than 45 days old, and the figure on those pages stops printing a check date at all. That notice is not an apology — it is the only honest statement this site can make about what it has, and it is the reason the date is worth anything.

Where AI is used in making this site

Openly: AI models assist with research summarisation, first drafts and checking my own spelling. Every published page is then read end to end, corrected, and had its claims verified or removed by me before it goes live. What you will not find is an auto-published page — no scheduled generation, no scraped feeds, no tool listings imported from a dataset. Roughly a dozen pages a month is the pace this site can hold while still being worth reading, and that's the pace it keeps.

Why disclose it? Because "AI wrote this" is not a disqualifying fact, but hiding it is a credibility one. The value in this niche is judgement and honesty about limits, and you're entitled to know which parts of a page you can lean on.

Corrections and complaints

1. Email [email protected] with the page and what's wrong. A vendor correcting me about their own product gets the same treatment as anyone else.
2. If it's a factual error, the page is fixed — usually within a few days, and the change is recorded on Updates & corrections with the date and what changed.
3. If a tool's behaviour has degraded enough that the profile misleads, it gets a dated note at the top, and if it's no longer worth recommending it comes off the front page rather than being quietly rewritten.
4. Disagreement is not a correction. "I like this tool more than you do" gets a polite reply and no edit.