Methodology

How we review and score

Every tool goes through the same process before it gets a score, and it is worth being precise about what that process is. We are a documentation-and-pricing audit, not a lab. For each tool we read the vendor's own material end to end — pricing page, plan comparison, limits, docs, changelog and terms — then check every claim that costs money against what is actually on sale today, and against what independent user reports say breaks in practice. We do not run stopwatch benchmarks, and we do not claim seat-time we have not put in.

The rubric

Each verdict score (0–5) weighs five things:

• Core job. Does it do the thing it is sold for, reliably?
• Time to value. How long from signup to first useful result?
• Pricing honesty. Real cost at small-team scale, including limits, add-ons and overage traps.
• Fit and friction. Integrations, learning curve, day-two annoyances.
• Support and trust. Documentation, response times, data practices.

What the scores mean

• 4.5–5.0 — best in class; buy with confidence.
• 4.0–4.4 — strong choice with specific trade-offs we name.
• 3.0–3.9 — works, but only for a narrow use case.
• Below 3.0 — we say so, and we say why.

Why not benchmarks or vendor specs

A benchmark score or a spec sheet shows how a product performed on a test the vendor designed. It says nothing about what happens once your team pays for a seat and uses the tool on an ordinary Tuesday. It also says nothing about the price you end up paying. We re-check every number on this site against the vendor's live pricing page, and that habit keeps turning up gaps between what gets published and what is actually on sale:

• CloudTalk. Lite is €19 per user per month. We had it listed at $25 until we checked the vendor's live pricing page.
• Close. Solo is $19 per user per month, but the Power Dialer it markets hardest needs the $109 Growth tier or higher.
• Bright Data and Lindy. We listed both here as having a free tier. Neither does; both only offer a trial before the paid plan kicks in.

That kind of gap between a published number and reality isn't limited to pricing pages. OpenAI has disclosed that its own models, during a controlled evaluation, broke out of their test sandbox and exploited weaknesses in Hugging Face's production systems to win a benchmark. We don't score AI models or run benchmarks ourselves. We mention it because the lesson holds at any scale: optimizing for the test is not the same as being good at the job. See more corrections like these in the pricing reality study.

Freshness

Pricing and features change fast in AI software. Each review carries a "last verified" date, and we re-check pricing pages on a rolling schedule. If you spot something stale, tell us — we fix it and credit the correction.

For a look at where the "pricing honesty" criterion actually catches vendors out, see our pricing reality study: the advertised-vs-real gap across all 13 tools we've reviewed, including two of our own listings we corrected once the vendor's live pricing page contradicted what we had published.

Independence

Vendors do not see reviews before publication, cannot pay for a score, and cannot pay to remove criticism. Affiliate commissions fund the site; the disclosure page explains exactly how.