Independently agent-testedAnthropic · OpenAI · Gemini · Grok
Try Tested®

Methodology

How we score AI tools

Every score on Try Tested comes from the same process, and every part of it is published: the panel, the rubric, each model's score and its reasoning, and how scores change over time.

The panel

Each tool is rated by four AI models from four different labs. Each model scores the tool on its own, without seeing the others' scores. The headline score is the average of the four. Tool pages show all four scores side by side, with each model's one-sentence rationale, so you can see where they disagree.

Using four labs is deliberate. A single model has blind spots and can favour its own maker's products; averaging across competitors damps both.

The rubric

Models are asked to rate a tool's overall quality and usefulness for its category on a 0–10 scale, calibrated against the real market for that category rather than an ideal tool. A typical tool in a crowded category is average and should land near 6.

9.0–10
The one or two tools that define the category today.
7.5–8.9
Strong, widely recommended alternatives to the leader.
6.0–7.4
Solid but unremarkable; a reasonable pick for some users.
4.0–5.9
Niche, dated, limited, or a non-AI product with AI features bolted on.
Below 4
Discontinued, in maintenance mode, or not really an AI tool.

The models see the vendor's own description and feature list, and are told to treat it as marketing copy: they judge the real product from what they know about its adoption, reputation and current status. A tool that has been discontinued or shut down scores below 4 whatever its description says.

We recalibrated the rubric on 7 October 2026. Before that, the scale had no anchors and most tools scored between 7 and 9, so the score stopped telling tools apart. Score-history charts show a step on that date: a change of method, not a change in the tools.

What the score is not

It is not a hands-on lab test. The panel doesn't sign up for each tool and run it through tasks. It is a structured, repeatable assessment by four independent models, built on what they know about each product. That makes it broad and consistent across more than a thousand tools, but it can lag when a product changes quickly. Treat a score as a well-informed starting point, and use the rationales to see why.

When scores update

New tools are scored within a day of being added. The whole catalog is re-scored on the first of every month, and each run is kept, so every tool page shows its score history. When the same product is listed twice, we merge the listings and the old page redirects to the one we keep.

Pricing

Prices are collected automatically from public information and can go stale; vendors change plans often. Always confirm on the vendor's pricing page before you buy.

Money

Some outbound links are affiliate links and pay us a commission if you buy. The panel never sees whether a tool has an affiliate program: the models get only the tool's name, category, description and features. Rankings are sorted by score alone, affiliate links are labelled, and a commission never changes your price. Details are in our terms and privacy policy.