Our protocol
How we test AI tools
Most "AI tool reviews" are summaries of the vendor's own landing page. Ours are not. Every review on this site follows the same public protocol, so you can check our work — or rerun it yourself.
The 5 steps
| Step | What we do | Why it matters |
|---|---|---|
| 1. Real task | We pick 3 jobs an actual user would pay for — not demo prompts the vendor scripted. | Demo outputs are hand-polished. Yours won't look like them. |
| 2. Same script | All competing tools get the identical input: same text, same length, same languages. | Only controlled inputs make comparison fair. |
| 3. Blind judgment | Outputs are labeled A/B/C and rated on voice realism, pacing, artifacts — listener doesn't know which tool made which. | Brand bias is the #1 failure of AI tool comparisons. |
| 4. Export check | We export the final file on the cheapest paid plan and on the free tier: format, watermark, download limits. | Many tools shine in-app and fall apart at export time. |
| 5. Price math | We compute real cost per finished minute of output, including credits consumed by retries and failed generations. | Sticker price ≠ real price. Retry burn is where budgets die. |
What we don't accept
- Sponsored rankings. Vendors cannot pay to change a score. Affiliate links fund the site; they never touch the verdict.
- Free-trial-only reviews. If we didn't test export limits and plan restrictions, we say so.
- Perfect 10s. Every tool has failure modes. If a review has no "don't buy this if…" section, we didn't test hard enough.
Corrections policy
Found an error or a result you can't reproduce? Email us with your settings and we'll rerun the test and publish the correction with a date stamp.
Affiliate disclosure: some links earn us a commission at no extra cost to you. Scores are set before monetization is considered — see the protocol above.