How We Test
Most AI tool roundups are rewritten press releases. This page explains exactly how we test, so you can judge whether our verdicts are worth anything.
Rule 1: we actually use the tool
Every review on this site is based on hands-on use of a real paid or trial plan — the same plan a reader would buy. We don't review from feature lists, vendor demos, or screenshots on a pricing page. If we couldn't get working access to a tool, we don't review it.
Rule 2: identical tasks across competitors
Tools in the same category get the same work. When we compare AI video generators, each one gets the same script, the same style instructions, and the same constraints. When we compare writing tools, each one gets the same brief and the same tone requirements. Same input, scored output — that's the only fair way to compare.
[HANDS-ON TEST: describe the exact test to run — e.g., "Generate a 60-second product explainer from this 150-word script, avatar presenter, 1080p export; record render time, word-error count, and export failures."]
Rule 3: a fixed scoring rubric
Every tool is scored on the same five dimensions:
| Dimension | What we check |
|---|---|
| Output quality | Does the result look/sound/read professional, or does it need heavy fixing? |
| Ease of use | How long from signup to first finished result? Where did we get stuck? |
| Speed | Render, generation, and response times on a typical task. |
| Pricing fairness | What the plan actually costs at real usage levels — including overages, seat minimums, and features locked behind higher tiers. |
| Support | We contact support with a real question during testing and report what happened. |
No single dimension decides a verdict. A fast tool with bad output loses to a slower tool with great output, and we show our reasoning.
Pricing fairness, specifically
This is where AI tools most often mislead. A "$20/month" plan with a 30-minute monthly video cap is not a $20 tool if you need two hours — it's a $20 tool plus overage fees. We calculate the real monthly cost at light, medium, and heavy usage, and we call out seat minimums, annual-only discounts, and features that only exist on the top tier.
Rule 4: screenshots of real outputs
Reviews include screenshots or exports from our actual test runs — not vendor marketing images. If a tool's output had flaws, the screenshot shows the flaws.
Rule 5: "last tested" dates and quarterly retests
AI tools change constantly — features ship, pricing changes, quality drifts. Every review carries a "last tested" date, and we retest our top-reviewed tools every quarter. When a retest changes the verdict, we say what changed and when.
What we don't do
- No paid placements. A vendor cannot pay to be reviewed, to be ranked higher, or to have criticism removed.
- No ranking by commission. We don't know or care what a tool pays affiliates when we score it. The rubric is fixed before testing starts.
- No reviews without access. If we can't use it, we don't cover it — even if readers ask.
- No retests skipped. A review more than a quarter old without a "last tested" refresh gets flagged on the page until it's retested or retired.
How we handle free access
Sometimes vendors give us extended trials or review credits so we can test paid tiers. We accept that — it lets us test the plan you'd actually buy — but it never affects the score, and we disclose it in the review when it happens. What we don't accept: payment for coverage, approval rights over our verdicts, or embargoes on criticism.
Corrections
If we get something wrong — a price, a feature limit, a test result — tell us through the [contact page](contact). We fix errors, note the correction, and update the "last tested" date.