We start from the outcome you need, not the longest feature list or the loudest benchmark. This page explains exactly what sits behind every verdict on this site.
Why there is no single score
A universal score out of ten implies one product is better than another at everything. That is almost never true. The best tool for drafting a long document and the best tool for refactoring a codebase are usually different products, and collapsing that into one number hides the only thing that matters: which one fits your job.
So we publish a verdict per outcome instead. Where we name a winner, we name what it won at.
What each review label means
Every profile on this site carries one of four labels. They are not marketing words. They describe what evidence exists.
- Source review. Every fact comes from the provider documentation linked on the profile, checked on the date shown. We have not run the product ourselves for this assessment. No performance claim is implied.
- Initial editorial review. An assessment based on documented capabilities, published pricing, vendor privacy terms, and hands on familiarity with the product. Controlled comparative testing has not been run yet, and we say so on the page rather than implying a benchmark.
- Controlled test. We ran a defined set of comparable tasks against this product under the same conditions as the others in its outcome group, and recorded the version and plan used.
- Full review. A controlled test plus extended use, with pricing and privacy terms verified against vendor documentation and the tested version recorded.
What we examine
- Output quality for the specific outcome, judged on clarity, accuracy, structure, and whether the result needed rework.
- Verifiability. Can you see where a claim came from, and how much checking does the output need before you can use it.
- Workflow fit. Files, memory, projects, integrations, collaboration, and the number of steps between starting and finishing.
- Reliability. Consistency across attempts, behaviour at the edges, and how gracefully it recovers when it gets something wrong.
- Value. Capability and real usable limits relative to what you pay, including what the free tier genuinely allows.
- Privacy. Whether your content trains the vendor models by default, whether you can turn that off, and where the data is held.
Versions, plans, and prices
AI products change monthly. Every profile records the version or tier assessed, the plan used, the price we found, and the date we last checked it. Prices are taken from the vendor pricing page and are shown in the currency the vendor publishes. Where a vendor does not publish a global price, we say the price is unverified rather than guessing.
If a price on this site is more than sixty days old, our internal maintenance checklist flags it for recheck. If you find one that is wrong, please tell us.
How the Popular Now list is ordered
The eight apps on the homepage are an editorial snapshot of what is shaping how people work, based on our own coverage, product significance, and reader interest. It is not a measured global ranking, we do not have usage data, and we do not publish invented user numbers. The list carries the date it was last reordered.
Independence
No vendor pays for a position on this site, and no vendor sees a review before publication. Where affiliate links exist they are disclosed and they do not affect a verdict. See our editorial standards and affiliate disclosure.
Corrections
Send anything that looks wrong to hello@18byte.com. We date every comparison and correct material errors with a public note.