Method
How we review AI tools.
We read the pricing pages and the documentation vendors publish, then put the tools side by side. Every figure on this site names where it came from and the day we checked it.
- 01
We pick an axis that separates the field
Before ranking anything we score every tool in the category on the axis we plan to rank by. If they all score the same, that axis decides nothing and we throw it away and find one that does. A rule that cannot rule against anything is decoration.
- 02
We read the vendor's own pages
Pricing page, docs and changelog. Every figure we publish is recorded with the URL it came from and the day a human opened it, and both appear in the source ledger at the foot of the page carrying that figure.
- 03
We cite benchmarks by name
Where an independent group has measured something, we link to their published result and say whose it is. We do not run our own benchmarks. No page here currently carries one, and when one does it will name the group that ran it.
- 04
We publish an editorial score against a fixed rubric
The rubric is below. The score is our judgment, weighted the same way for every tool in a category. The facts underneath it are not a judgment, which is why they are sourced separately and you can check them without agreeing with us.
- 05
We name who each tool is not for
A profile that never rules against the product is an advert. Every scored review here carries a Not for line, and it is required before the page will build.
The rubric
What the editorial score is made of.
Five weighted components, applied the same way to every tool inside a category. Weights differ between categories only where the category demands it, and where they do, the review says so.
| Component | Weight | What it reads |
|---|---|---|
| Fit for the stated job | 30% | How well the published capability matches the use case the category is about |
| Cost to get started | 25% | What the free tier actually allows, and what the first paid step costs |
| What the vendor commits to | 20% | Limits, terms and compliance the vendor publishes in writing, not in marketing copy |
| Portability | 15% | Export formats, API availability, and how hard it is to leave |
| Documentation quality | 10% | Whether the docs answer a real question without contacting sales |
Scope
What this method does not cover.
Our scores rest on published material: what a vendor commits to in writing, what their documentation specifies, and what independent groups have measured and published. That is a narrower base than running every tool for a month, and it is also a base you can check yourself, line by line, from the ledger on each page.
Where a vendor publishes nothing on a point, we leave the field empty rather than fill it, and the ledger prints no row for it. A missing row means we did not find it. It does not mean the feature is missing, and we are careful never to write it as though it did.