We Tested 20 Everyday AI Tasks on ChatGPT, Claude and Gemini — Here’s What Actually Won

Gemini invented a follow-up question the meeting had already answered — the other two got it right. Turns out “summarize this” isn’t as safe as it sounds.

A genuine coin flip. Both Claude and Gemini wrote warm, accurate, ready-to-send emails. ChatGPT’s version was accurate too, but noticeably thinner — no request to actually confirm the new time.

ChatGPT was the only one willing to do real date math — it assumed “today,” then correctly worked out that a Saturday plan lands on a weekend and confirmed it. Claude left every date unresolved, saying it needed an anchor date first. Gemini got partway there.

ChatGPT and Gemini tied — both correctly resolved “next Tuesday” and “the 30th” into real calendar dates. Claude refused to commit to dates at all, then listed them out of order anyway, which somewhat defeats the point of asking for a sorted list.

All three flagged the auto-renewal and the risky move-out clause — good, that’s the part that actually matters. Gemini’s structure mapped most directly onto the five things we asked for; Claude tacked on a legal disclaimer paragraph nobody asked for.

All three fit the 3-line limit. ChatGPT’s formatting was the cleanest and it picked the single most useful reminder (parking) instead of cramming two into one line.

ChatGPT wrote the most rigorously fact-checked itinerary — it actually flagged that the castle and monastery share a closing day, so don’t schedule them back-to-back. That’s the kind of detail that saves an actual trip.

Claude nailed every requirement — route, hour-by-hour plan, three nights of lodging, and a budget that added up to exactly $575 of the $600 cap. ChatGPT was upfront that its plan only covers two nights of lodging, not three — worth knowing before you budget the trip.

Gemini and Claude tied on this one, and for a fun reason: they each caught a different, genuinely useful detail the other missed — Gemini flagged Japan’s “Visit Japan Web” pre-registration, Claude flagged that a common decongestant is restricted at customs. Two different “oh, I didn’t know that” moments.

This one actually matters. We told it about a bad lower back, and Gemini still slotted “dumbbell thrusters” into a fast circuit — exactly the kind of move that stresses the exact thing we said to protect. Claude built a genuinely careful plan instead.

We said “using only these ingredients” — a real constraint, not a suggestion. ChatGPT was the only one that respected it to the letter (using tomato liquid instead of oil to cook). Claude quietly added oil, salt, and pepper anyway.

Claude went the extra step and actually calculated price-per-square-foot for each listing ($2.47 vs. $2.57 vs. $2.75) to back up its recommendation. The other two gave solid tables but skipped the math that makes “best value” a real answer instead of a guess.

Very close call. ChatGPT edged it out mostly on a technicality — Gemini’s comparison table used a raw HTML tag that won’t actually render as a line break if you paste it somewhere plain-text.

Claude and ChatGPT tied. Claude was upfront when a couple of gift ideas slightly missed the stated budget; ChatGPT went further and checked its price estimates against real, current retailer listings. Different kinds of careful.

ChatGPT gave exactly what was asked for — ten clean question-and-answer pairs, nothing extra. Claude’s answers were strong but ran long and paragraph-heavy, which defeats the point of something you’re supposed to rehearse out loud.

ChatGPT stuck to the facts we actually gave it. Gemini invented an achievement — “elevating client satisfaction” — that was never in the background we provided. Not something you want showing up on your actual resume.

ChatGPT delivered exactly 5 bullet points, no throat-clearing, nothing missed. Claude’s last bullet wasted space restating “no decisions made” instead of capturing the one thing that was actually offered (Sam volunteering to help with calls).

Gemini’s explanation was the warmest and most complete. Claude’s answer was solid too, but opened with a bizarre, robotic line — “no code or files needed here” — that has nothing to do with explaining a song. Not the tone you want from a chat about Fleetwood Mac.

Gemini and ChatGPT both fit inside 280 characters. Claude didn’t — we counted, its tweet ran to 289 characters, which simply doesn’t fit. A tweet that’s too long to post isn’t a small miss for this specific task.

All three landed on the same sensible verdict, but Claude’s answer to “what am I missing” was the sharpest — it spotted a subtle bias hiding in the pilot’s own numbers (easy tickets might be crowding out the hard ones), something the other two didn’t catch.

⭐ Want every prompt in one place?

All the prompts from this test (and 130+ more from our daily skills) are free on GitHub — DailySkill Prompt Library. Star it to get new prompts as we publish.