ChatGPT, Claude or Gemini — which for what?

The rankings change every few months. What stays is the kind of task. ChatGPT, Claude and Gemini are all fine for drafts. They are not equally fine for the same work. This is the split we start with, and a test for the cases that fall in between.

ChatGPT: the fast first pass

Use ChatGPT when you need several variants quickly from a short brief: subject lines, social hooks, an outline, a reply to a standard email. It is the tool most people already have an account for, and it is good at changing direction on command. “Shorter. Less sales language. Three versions.” works reliably there.

It gets weaker once the job is a long document and you need nothing invented. Then you have to put the check in the prompt explicitly, or it will smooth gaps over with plausible sentences.

Claude: long texts and a tight brief

Claude takes longer source material, contracts in plain language, interview notes and style guides seriously. It stays closer to the structure you set, and it writes less “website” when you asked for an internal note. Good for: cutting a 12-page transcript into a decision memo, pulling a text into a fixed tone, turning a brief into steps.

Less useful as the only tool when you want ten short variants in two minutes. The back-and-forth pace of ChatGPT is often faster for that.

Gemini: research next to the files you already have

Gemini is the sensible first choice when the work lives in Google docs, sheets, or search. Summarising what a doc already says, building a table from notes, checking a question against current search context: you do not have to copy the material into another window first. For a free advert with no sources, the advantage is small.

The 20-minute test when you are unsure

Rankings do not replace this test. Take a real task from this week, the same prompt, three windows. Score each answer with four questions, 0 or 1 each:

• Does it contain only what you supplied?
• Does it keep the length and the shape?
• Would you send it with one correction at most?
• Was the path shorter than writing it yourself?

The model with the most points becomes your default for that kind of task. Not for everything. A second kind of task — a long transcript instead of a short email — can go the other way. Two tests are enough for a personal rule.

Summarise the note below as an internal decision memo.

Shape: exactly five bullets. Each bullet at most 20 words.
Use only what is in the note. No recommendation that is not already there.
Order: situation, open question, two options, next step.
If something is missing, write [missing] instead of filling it in.

Note:
The client wants to move the newsletter from monthly to every two weeks. Editorial is one person, four hours per issue. Open rates of the last three issues: 28%, 31%, 27%. No new list, no budget for a second person. Decision by Friday. Options so far: a shorter issue, or one issue built by reusing older pieces.

The starting points are in ChatGPT for Beginners, Claude for Beginners and Gemini Every Day. How the prompt itself has to be built is in 5 prompt mistakes every beginner makes.