AIPickrEvery recommendation on this site follows the same method. Here it is in full, so you can judge our verdicts for yourself.
1. Start from a real job
We test tools on work our readers actually do: a 1,500-word blog draft, a YouTube Short from a podcast clip, a bug fix in a real codebase. Feature lists come second.
2. Same task, same prompts
Each tool in a comparison gets an identical brief and identical inputs. We keep one standard prompt set and paste it unchanged into every tool, in a new chat with no custom instructions:
- Writing: a short office email from webinar numbers, a 300–350-word blog section with a price placeholder, and a paragraph with five planted grammar errors to edit.
- Images: a 16:9 blog header, and a YouTube thumbnail that must spell an exact headline.
- Coding: a tiny project with one failing test to fix.
We save the prompts and raw outputs so a retest months later is a fair comparison. We note what we can observe, such as how many attempts a usable result took, what went wrong, and how close the output came to the brief. We don't publish response times unless we actually measured them.
Who runs the tests
Tests are run with AI-assisted browser operation: an AI agent pastes the standard prompts into each tool and captures the screen. The site owner reviews the logs, outputs and screenshots before anything is published. Screenshots are real captures of the tool; we only crop them and black out personal details. First-hand notes in our articles start with "In our test…" and describe that single test session, not long-term use.
We only use access we already have. If a tool needs a sign-in, a paid plan or a verification step we don't have, we skip it and say so rather than signing up or paying just to test.
What we've tested so far (November 2026)
- Tested: ChatGPT (Free plan), Gemini, and Grammarly's free online grammar checker (signed out).
In every article, sections based on our own hands-on testing are marked In our test. Everything else is based on the vendors' official pages and documentation. Where we tested on a free plan but recommend a paid one, the note says so.
3. Prices from official pages only
- We copy prices from the vendor's own pricing page and note the month we checked, shown as Last checked: YYYY-MM.
- If we can't confirm a price or limit on an official page, we write Not listed instead of guessing.
- We recheck prices at least every quarter and whenever a vendor announces a change.
4. What we score
| Criterion | What it means | Weight |
|---|---|---|
| Output quality | Accuracy, usefulness and how much editing the result needs | 35% |
| Ease of use | Time from sign-up to a usable result | 20% |
| Value | What the paid plan gets you versus the alternatives | 20% |
| Features & integrations | Fits into tools you already use | 15% |
| Privacy & controls | Data training opt-outs, team controls, export | 10% |
Scores appear only after a tool has been tested hands-on. Until then the score card stays blank.
5. A verdict, every time
Each guide names a pick and says who should use something else. "It depends" is never the final answer.
6. Corrections
If a vendor changes a plan or you spot a mistake, tell us through the contact form. We update the page and its "Last updated" date.