Our method
A score is only worth something if you can trace where it comes from. Here is exactly how Joute evaluates a tool.
The Joute score, out of 10
Each tool gets an overall score out of 10, calculated from six weighted criteria. The weighting is public, no black box.
The overall score is a weighted average, with a safety cap: if any single criterion falls below 4/10, the overall score is capped at 7.5. An expensive but broken tool cannot display a 9.
The obsolescence risk, across five factors
This is Joute's signature indicator, a completely separate axis from the quality score. The question: will this tool still be useful in eighteen months, or will its use case be absorbed by ChatGPT, Claude, or Gemini?
It is scored out of 10 across five precise factors:
- Absorbability (weight 30%): can the core value be taken over by a major model at the next refresh? A GPT wrapper for generating marketing copy = 2/10. A code editor that maintains its own index = 8/10.
- Technical moat (weight 25%): proprietary data, specialized infrastructure, privileged access to a proprietary model. A tool with 5 years of structured user data = 9/10. A tool consuming a third-party public API = 3/10.
- Verifiability (weight 20%): does the tool expose what it does step by step (logs, citations, traces)? An agent that shows its actions = 8/10. An opaque tool that just delivers a result = 4/10.
- Distribution (weight 15%): does the tool have a real paying user base and a sustainable acquisition channel? GitHub presence, measurable traction, distribution partnerships.
- Evolution speed (weight 10%): public changelog, release frequency, responsiveness to new models. A tool shipping a major feature every month = 8/10. A product frozen for 6 months = 3/10.
A low score is not an insult, it is a warning before you pay a subscription that goes nowhere.
How we test, concretely
No paraphrasing product pages. Each tool is used on real projects, in real conditions.
- Minimum duration: 1 week of daily use for a simple tool (chat, image, voice), 3 weeks for a dev tool (IDE, agent), 1 month for team platforms.
- Hardware: MacBook Pro M3 (16 GB RAM) for Mac tests, Windows 11 (Ryzen 7, 32 GB) for PC tests. 1 Gbps fiber connection. Degraded network tests (4G throttling) for web apps.
- Paid accounts: every subscription is taken at the public price, on monthly billing (no annual discount). No partner code, no free seat, no internal version. Invoices available on request for skeptical publishers.
- Reference prompts: we systematically use a set of 8 reference prompts per category (writing, code, image, voice, etc.) to compare tools on the same inputs. The set is available on request.
Update policy
- Prices: re-verified monthly via a bot that scrapes official pricing pages on the 1st of each month. Changes are logged on the public changelog.
- Joute scores: reviewed quarterly on the top 30 tools, or earlier if a major release justifies it. A score goes down as much as it goes up.
- Obsolescence risk: reviewed at each major release of the main models (Claude, GPT, Gemini, Llama), so 4 to 6 times a year.
- Comparisons and reviews: the last update date is shown at the top of each page (
updatedAtin the frontmatter). Beyond 6 months, the page is flagged for re-review.
What excludes a tool from the catalogue
We remove a tool from the catalogue in four cases:
- Dead: official shutdown announcement, redirect to another product, or domain returning 404 for more than 30 days.
- Trivial wrapper: a product whose core functionality is a system prompt on top of a public GPT API, with no measurable added value.
- Questionable practices: deceptive billing (hidden renewal, intentionally complex cancellation), abusive terms of service, documented data breach history.
- Out of scope: a tool that is not a genuine generative or agentic AI application (classic feature extensions, marketing layers).
Money and independence
Joute uses affiliate links. A commission may be paid when you sign up via the site, at no extra cost to you. This commission never influences a score or a ranking, and every page containing affiliate links states this clearly.
To make this independence concrete, the affiliation page lists tools we actively recommend without being affiliated, and programs we refused for ethical reasons. That is the most direct proof that money does not drive recommendations.
Reporting and contact
If you spot a factual error (outdated price, misdescribed feature, disappeared tool), write to hello@joute.io with a source. We correct within 7 days and date the correction in the changelog. For publishers contesting a score, same email: we document the counter-argument, retest if necessary, update publicly.
Method last revised: 23 May 2026. Next revision planned: 1 September 2026.