ChatGPT, Claude, and Gemini are not three versions of the same intern

Most people I know have one AI tab they actually use, and two they opened once because a colleague said “you should try Claude.” They paste the same sentence into all three — “help me with this email” — and then pick whichever answer sounds nicest. That is a perfectly reasonable way to choose a restaurant. It is a terrible way to choose a colleague.

ChatGPT, Claude, and Gemini are not three skins on one brain. They were trained differently, they were rewarded for different habits, and they get confused by different kinds of vagueness. If you talk to all of them like ChatGPT, two of them will politely underperform and you will decide they are “worse,” which is like deciding a sous-chef is worse than a waiter because you asked both of them for the bill.

You do not need a framework. You need three small habits, one per model, and the honesty to notice when you are copy-pasting out of laziness rather than judgement.

ChatGPT: the eager generalist

If you have used only one model, it is this one. ChatGPT is the intern who has read everything, wants to please you, and will invent a confident answer rather than sit in silence. That is wonderful for first drafts, outlines, and “give me five options.” It is dangerous when you needed the answer, not an answer.

Casual users get the best out of ChatGPT by being bossy about shape and picky about what “done” looks like. Tell it the format. Tell it who it is talking to. Tell it what to leave out. If you do not, it will give you a tidy, generic page that could have been written for anyone, which is another way of saying it was written for no one.

• Ask for options, then pick, then tighten. ChatGPT is happier iterating than nailing it first time.
• Say “do not invent names, numbers, or sources.” It will still try. Check anyway.
• Give it a scrap of your actual writing if tone matters. “Make it professional” is how every LinkedIn post started to sound the same.

Claude: the careful reader

Claude is the colleague who actually read the brief. It is better when you paste the documents, the messy notes, the “here is what we already tried.” Casual users underuse it because they treat the box like a search bar. Claude is not a search bar. It is a person you have handed a folder.

It also tends to hedge, which feels like weakness if you wanted swagger and is a gift if you wanted something you could send to a client without fact-checking every adjective. Give it structure: headings, XML-ish labels, “answer only from the text I pasted.” It likes being told where the walls are.

• Paste the source. A PDF of the brand guide beats “write in our voice.”
• Ask it to quote or cite the bits it used. If it cannot, you asked the wrong model or gave it nothing to stand on.
• Use it for rewriting, not for trivia. It is a better editor than it is a magician.

Gemini: the one already in your stuff

Gemini’s unfair advantage, for a normal person, is not “smarter.” It is nearby. If your life is Gmail, Docs, and a phone, the best prompt in the world is the one you do not have to copy into another tab. People try it once from a clean chat window, get a bland paragraph, and go back to ChatGPT. They never gave it the only thing that makes it special: the document sitting underneath.

When you are in a blank chat, treat Gemini like a fast, slightly literal intern. Short jobs, clear outputs, less philosophy. “Ten subject lines, max 50 characters, no puns” will beat “help me with marketing.” It is also the model people forget they already pay for inside Google One, which is how you end up with three subscriptions and one habit.

• Use it where the file already is — summarise this Doc, draft from this email thread — before you open a new chat.
• Specify length and format as if you were texting. Gemini does fine with constraints; it does less fine with vibes.
• Do not ask it to be Claude. Long, delicate, “only use this source” jobs belong elsewhere.

The same job, three prompts

Here is a task everyone has: “turn my messy notes into something I can send.” That sentence is not a prompt. It is a wish. Each model wants a slightly different wish.

• ChatGPT: tell it the audience, the length, and what to avoid, then ask for two versions.
• Claude: paste the notes, forbid invented facts, give it section headings.
• Gemini: if the notes are already in a Doc, stay there. If not, ask for bullets first, prose second.

None of this is secret knowledge. It is the difference between talking to a stranger at a party and talking to someone who has just read your notebook. Casual users do not fail at AI because they lack a certification. They fail because they use one sentence in three products and call that a fair test.

If rewriting the prompt three times sounds like a chore, that is the point. It is a chore. That is why a prompt enhancer exists — not to make you sound clever, but to stop you doing the boring bit by hand, badly, in three tabs.

LaunchSphere lets you say which model you will actually run — ChatGPT, Claude, or Gemini — and rewrites the brief for that listener. You still paste the result into the tool you already use. You just stop pretending they are the same intern in different jumpers.