Claude vs ChatGPT for Productivity: Prompts That Work in Both

This page may contain affiliate links.

If you use both Claude and ChatGPT, you've probably noticed something annoying: a prompt that sings in one falls flat in the other. ChatGPT might give you a bulleted list when you wanted prose. Claude might refuse to give you the short answer you asked for. You end up maintaining two sets of prompts, or worse, rewriting them every time you switch tabs.

Here's the good news. The differences between the two models are real, but they're small enough that a well-written prompt works in both. The trick isn't learning two prompt languages — it's learning the handful of habits that make a prompt portable. Below are the patterns I use daily, with copy-paste examples you can drop into either tool right now.

Why the same prompt behaves differently in each tool

Claude and ChatGPT were trained differently, tuned differently, and have different default personalities. ChatGPT tends to be eager — it will pad an answer with structure, headers and caveats unless you tell it not to. Claude tends to be more conversational and will happily write you three paragraphs when one would do, because it's trying to be thorough.

Neither is wrong. But if your prompt is vague, each model fills the gap with its own habits. That's why the fix is almost always the same: remove the ambiguity, and the outputs converge.

A quick example. Ask either tool "write an email about the project delay" and you'll get something generic and slightly embarrassing. Ask "write a 120-word email to a client explaining our project is delayed by two weeks because of a supplier issue, apologise once, offer a revised date, no corporate waffle" and both tools produce something you'd actually send.

The five-part prompt structure that travels between models

Every prompt I reuse across both tools has the same skeleton. It's not clever, but it's reliable:

1. Role — who the AI is pretending to be
2. Task — the one thing you want done
3. Context — the facts it can't guess
4. Format — length, structure, tone
5. Constraints — what to avoid

Here's that as a reusable template:

You are an experienced [role]. [Task in one sentence]. Context: [relevant facts, audience, background]. Format: [length, structure, tone]. Do not: [things to avoid — clichés, filler, emoji, bullet points].

That template works in Claude and ChatGPT without modification. The constraints line is the secret weapon — it's what stops ChatGPT adding a cheerful summary and Claude adding a philosophical reflection.

Prompt patterns for the four jobs you actually do

Most knowledge work falls into a few buckets. Here are portable prompts for each.

Email triage and replies. The problem with AI email drafts is they sound like AI. Force specificity:

You are a busy UK freelancer. Draft a reply to this email. Keep it under 100 words, friendly but direct, British English. Do not start with 'I hope this email finds you well'. Do not add a subject line. Here is the email: [paste]

Both tools handle this well. The 'do not' lines matter more than anything else.

Planning and prioritisation. Vague requests get vague plans. Give the model your actual constraints:

You are a project manager. I have these tasks: [list]. I have 6 focused hours today and a hard deadline on [task] at 4pm. Build me an hour-by-hour plan. Flag anything I should drop or delegate. Output as a simple table, no preamble.

ChatGPT loves tables; Claude will produce them when asked. The 'no preamble' instruction saves you scrolling past two paragraphs of encouragement every single time.

Meeting prep and notes. This is where a portable prompt earns its keep:

You are a chief of staff. From these rough notes, produce: (1) a 5-bullet agenda, (2) three questions I should ask, (3) a one-line desired outcome. Keep it under 200 words total. Here are the notes: [paste]

Focus and thinking. Both models are genuinely useful as a thinking partner if you stop them lecturing you:

Act as a sceptical colleague. I'm about to [decision]. Ask me the three hardest questions you can about it. One at a time, wait for my answer before the next. Do not give advice yet.

That 'one at a time, wait for my answer' instruction is the single most underused prompt trick in both tools. It turns a monologue into an actual conversation.

Where the two tools genuinely differ (and how to handle it)

Three differences are worth knowing about.

Length discipline. ChatGPT will usually respect a word count. Claude sometimes overshoots because it's trying to be complete. If length matters, give Claude a hard ceiling and a reason: "Maximum 150 words — this is going in a Slack message."

Long documents. Claude handles very long pasted text gracefully and tends to summarise faithfully. ChatGPT is excellent too but can drift on huge inputs. If you're feeding in a 20-page report, both work — just ask for a summary with page references so you can spot-check the accuracy yourself.

Tone defaults. ChatGPT leans upbeat and slightly corporate. Claude leans warm and essayistic. If you want dry and factual, say so explicitly in both: "Neutral tone. No enthusiasm. No exclamation marks."

None of these require separate prompts. They require one extra sentence in the constraints line.

If you'd rather skip the trial and error entirely, I've bundled the prompts I actually use into an AI Productivity Pack — 90 tested ChatGPT and Claude prompts covering planning, email, meetings and focus, from £9. It's the done-for-you version of everything above, if that's more useful than building your own library.

Building your own portable prompt library

The real productivity win isn't any single prompt — it's not rewriting the same instructions every week. A few habits make that easy:

Keep one document, not two. If a prompt needs a Claude-specific tweak, add it as a bracketed note rather than forking the prompt. Two libraries rot; one library gets maintained.

Version your constraints. When a prompt produces something slightly off, don't rewrite it — add a line to the 'do not' section. That's where 90% of quality gains come from.

Test new prompts in both tools once. Paste the same prompt into Claude and ChatGPT side by side. If the outputs are broadly similar, you've got a keeper. If they diverge wildly, your prompt is still too vague.

Store the good outputs, not just the prompts. A folder of example outputs is the fastest way to show the model what 'good' looks like — paste one in as a style reference and both tools will match it closely.

One practical note: if you're doing a lot of this, a decent keyboard and a second monitor genuinely help more than another AI subscription. You can find compact wireless keyboards on Amazon UK if yours is the bottleneck.

The bottom line

You don't need to choose between Claude and ChatGPT, and you don't need two prompt libraries. Write prompts with a clear role, a single task, real context, a defined format and a short list of things to avoid — and both models will give you something usable. The constraints line does more work than any clever technique.

Start with one prompt from this article. Adapt it to a job you do every week. Run it in both tools, keep the better output as a style reference, and add constraints until it's reliable. Do that for a fortnight and you'll have a small library that works everywhere — which is worth far more than being loyal to either tool.