The same ten jobs, given to Claude, GPT and Gemini: what came back
Ten real freelance tasks, one prompt, one rubric. The answer is “it depends on the task” — but the gaps are wider than expected.
We took ten jobs we had actually been paid for, gave the same prompt to all three, and had two editors grade the output blind. Three grades only: ship as is, needs fixing, unusable.
The totals were close. The distribution was not, which means “which one is best” is the wrong question and “which one for this job” is the right one.
Three clear differences
Long-form rewriting and tone: Claude shipped as is eight times out of ten, and needed fixing twice. It is the least likely to flatten the original voice.
Anything needing verification: Gemini made the fewest factual errors, because it can search. Twice the sources it cited would not open, so you still click through.
Code and spreadsheets: GPT had the highest ship-as-is rate, especially for structured output.
{"q":"We kept all three in the end. Sixty dollars a month, and the time it saves is worth more than sixty dollars.","by":"A freelancer, four years in"}
If you must pick one: Claude for content work, GPT for technical work, Gemini if you look things up all day. Those are three different people, not one answer.
