Claude vs ChatGPT for Writing, Three Real Briefs and the Actual Copy Each One Produced

By Frederik Baun, conversion copywriter and founder of Words Sounds Visuals

I've run client briefs through Claude, ChatGPT, and Gemini side by side since Gemini added a real writing mode, and I edit somewhere north of a hundred client drafts a month doing it.

Testing Claude vs ChatGPT vs Gemini for Writing on Three Real Briefs

I ran this test on 20 September 2026, using three briefs pulled from my own queue instead of a made-up prompt, because a made-up prompt produces a made-up answer.

Brief one was an about page paragraph for a boutique client, 120 words, built to sound like a specific founder rather than a template.

Brief two was a LinkedIn hook for a bootstrapped SaaS founder announcing a product pivot, one line, built to stop a scroll on a topic that could easily read as bad news.

Brief three was a product description for a candle brand, 40 words, tight, benefit-led, no room for filler.

I gave all three models the identical brief text, the identical word count, and the same two or three reference lines of the client’s existing copy where one existed.

Then I graded each draft the way I’d grade a junior copywriter’s first pass. Did it sound like the client? Did it need a full rewrite, a light edit, or nothing at all?

This is a narrower version of the question I get asked most often about AI copywriting in general: which tool should you use.

The honest answer keeps turning out to be “it depends on the brief,” so I ran three different shapes of brief instead of one, to show exactly where that answer changes.

Which AI Writes the Best About Page?

Claude, on this brief, by a clear margin.

I gave all three models the client’s existing homepage copy as a voice reference: warm, plain-spoken, a little self-deprecating about how long the studio’s been in the same building.

Claude’s opening line: “The building’s been leaning slightly to the left since before any of us worked here, and so have our opinions on open shelving.” That’s the exact register of the client’s own homepage copy, self-deprecating and specific.

Claude carried that same register through all 120 words. I moved two sentences and cut one adjective. That was the whole edit.

ChatGPT’s opening line: “Founded on a passion for thoughtful design, our studio has served the local community for over a decade.” It’s accurate, but it’s a sentence any design studio could run.

It reads like a well-organized stranger wrote it. The warmth wasn’t there until I added it back in by hand.

Gemini’s first draft opened with “We believe great design starts with listening,” competent but interchangeable with any other design studio’s about page.

I fed it two more lines of the client’s copy as a second pass, and the rewrite picked up the building detail on its own. It still needed more editing than Claude’s first draft did.

The shape of this result matches what I found on a narrower Claude vs ChatGPT copywriting test I ran the month before.

Give Claude a real voice reference, and it holds that reference across the length of a longer piece better than the other two manage on a first pass.

Which AI Writes the Best LinkedIn Hook?

ChatGPT, on this brief, because the job called for volume rather than one finished line.

I wasn’t holding a strict voice document for this one, so I asked all three models for ten hook variants each, built around the same announcement: a product pivot the founder needed to frame as progress rather than a failed first attempt.

ChatGPT returned ten distinct angles in one response, each built on a different mechanic. A direct question, a number, a confession that the first version hadn’t worked, and a contrarian claim about pivots being a good sign rather than a bad one.

The one I sent to the client was ChatGPT’s contrarian version: “Everyone told us the pivot was a red flag. Here’s the metric that told us it wasn’t.”

Claude gave me six variants instead of ten, four of which were strong enough to send to the client without editing.

The strongest read: “We killed the feature our first ten customers asked for. Revenue went up the next month.” The other two repeated a sentence structure Claude had already used in variant one.

Gemini returned ten variants like ChatGPT did, but two of them were close enough in phrasing to count as the same hook twice, which left effectively nine distinct options instead of ten.

When the job is fishing for one good line out of many rough ones, raw distinct volume wins.

ChatGPT gave me the most of it on this brief.

Which AI Writes the Best Product Description?

Gemini, on this brief, with the least editing of the three.

Forty words is a tight enough brief that word choice matters more than structure.

Gemini’s first draft opened, “Burns clean, burns slow, and still smells like the candle you’d steal from someone else’s coffee table.” That used two benefit words from the client’s existing product page, clean and slow, without me feeding them in as a reference.

Claude’s draft was well-written but ran a little poetic for a 40-word product blurb: “A small fire for a loud week.” The metaphor worked, but it read as more literary than the client’s usual, blunter copy.

ChatGPT’s first draft hit the word count exactly, but it opened with “an exquisite, artisanal fragrance experience,” and both of those words were on the client’s do-not-use list from an earlier brief. That cost it the edit despite otherwise solid structure.

This was the one brief of the three where the model I’d have guessed going in wasn’t the one that won.

How Each Brief Came Back

Each model won exactly one brief, and lost on a different weakness each time. That’s the real result of this test, not a leaderboard.

Which AI Is Best for Writing?

None of the three outright, since each one won a different brief above for a different reason.

Claude is the one I reach for when a piece has to sound like one specific person or brand across more than a paragraph.

It held a voice reference longest in every test I’ve run on it, including the narrower Claude vs ChatGPT copywriting comparison I published in August.

ChatGPT is the one I reach for when I need volume, ten headline options or ten hook angles to choose from, rather than one near-finished draft.

Gemini is the one I’m still learning the edges of, since it produced the best single draft of the three tests here and the weakest, on two different briefs, in the same afternoon.

Is Claude Better Than ChatGPT for Business Writing?

For anything that needs to sound like a specific business rather than a generic one, yes, based on both this test and the copywriting-only comparison I ran the month before.

The pattern holds across both tests: give Claude a written voice reference or a handful of approved lines, and it carries that reference through a longer piece with less editing on the other end.

ChatGPT closes that gap on shorter pieces, a headline, a hook, a single sentence, where there’s less room for tone to drift over the length of the draft.

So the honest version of “Claude vs ChatGPT for business writing” depends on how long the piece runs and how much personality it has to carry, not on which model tests better on a general benchmark.

The pattern is consistent enough across the three briefs above that I’d expect it to hold on a longer piece too, an email sequence or a full landing page.

The mechanism is the same each time: the longer a draft runs without a voice reference re-anchoring it, the more it drifts toward the generic middle.

Where Does Gemini Fit in a Writing Workflow?

As the wildcard worth testing on any brief with a tight, specific benefit to communicate, based on what it did with the candle brand’s product description above.

Gemini’s strength in my three briefs was picking up on-brand language from a short reference without me spelling it out in the prompt, which is exactly what happened with the two benefit words it used unprompted.

Its weakness was consistency across a longer piece and across multiple variants of the same request, shown by the generic first pass on the about page and the two near-duplicate hooks on the LinkedIn brief.

I’ve started running short, tightly scoped briefs through Gemini first and saving Claude for anything that needs to hold a voice across more than one paragraph.

How Do You Tell Which Draft Still Needs Editing?

Read it against the client’s own copy, not against a general sense of whether it “sounds good,” since a draft can be well-written and still be off-voice.

I keep a running list of the specific AI words I cut on sight, and I ran that check against all nine drafts in this test.

ChatGPT’s about page draft was the only one that tripped it, with “passion for thoughtful design” landing on the list.

Beyond vocabulary, I look for the broader patterns that mark a draft as AI-written: evenly-paced sentences, no specific detail, a structure that could belong to any brand in the category.

That’s what caught Gemini’s first about page pass and ChatGPT’s about page draft both, even though neither one used a single flagged word.

If you want a second opinion beyond your own ear, I’ve also tested how reliable AI content detectors are at catching this kind of draft.

Your own ear, checked against real client copy, still beats the detector.

Which Claude Model Should You Use for Writing?

Sonnet 5, by default, for the vast majority of writing work: cold email openers, landing page copy, about pages, LinkedIn hooks, full sequences. It’s what generated every Claude draft in the three briefs above.

I only move up to Opus 5 when the brief turns into a strategy problem rather than a drafting one.

A positioning angle built from scratch or a full offer rebuild, where the job is weighing five or more real trade-offs instead of writing a finished sentence, is the kind that earns it.

I keep client writing off Haiku entirely. That tier is built for fast, cheap classification work, and I don’t want anything a client will read coming out of a model tuned for speed over quality.

What Is the Best AI Model for Copywriting?

There isn’t a single best model for copywriting, based on all four briefs I’ve now run across two separate tests. There’s a best model for the specific brief in front of you.

Voice-heavy work, client emails, about pages, anything that has to sound like one particular business, points toward Claude, especially once you’ve packaged a real voice reference for it to read.

High-volume idea generation, ten headline options, twenty hook angles to test against each other, points toward ChatGPT.

Tight, benefit-dense copy under a strict word count is the one shape where Gemini beat both of the other two in my testing, and it’s worth a test run on that specific kind of brief before you default to whichever tool you already have open.

That’s a different answer than the one I gave in my earlier Claude vs ChatGPT copywriting comparison, since that test only ran two models, so Gemini’s product-description win had nowhere to show up.

Add a third model to the test and a third answer becomes possible. That’s the whole argument for running your own version of this before you commit to one tool.

How Do You Choose Between Claude, ChatGPT, and Gemini for a Specific Brief?

Start with the shape of the brief, not the model’s reputation.

A long piece that has to hold one voice points to Claude. A short piece where you want many angles points to ChatGPT. A tight, benefit-led piece is worth a Gemini pass before you commit to either of the other two.

Then paste in three to five lines of copy the client has already approved, since every result in this test changed based on whether a real voice reference was in the prompt or not.

Grade the draft the way I did above: does it sound like the client, and does it need a full rewrite, a light edit, or nothing at all. That question tells you more than any general leaderboard will.

Pasting the same reference lines into a fresh chat every time gets old by the fourth client, so I built a copywriting skill for Claude out of my own voice rules and client swipe files.

Claude reads the file before it drafts a single line, instead of me typing the same three paragraphs into every new conversation.

How to Set Up Your Writing Workflow

If you’re standardizing on one tool, pick based on the shape of the writing you do most often, not a general benchmark score.

Voice-heavy, client-facing writing points to Claude. High-volume idea generation points to ChatGPT. Tight, benefit-dense copy under a strict word count is worth a Gemini test before you assume it’s the weaker option.

I use all three on purpose, and which one wins changes with the brief in front of me, exactly like it did across the three briefs above.

Keep a short record of which model won which shape of brief, the way the table above tracks it for mine. After a dozen briefs, the pattern for your own writing will be more reliable than any general comparison, including this one.

Whichever model drafts the piece, run the output through the same editing pass afterward.

I’ve written separately about how to make AI writing sound more human and about the specific AI words worth cutting on sight. That pass matters more than which model produced the first draft.

Run your own version of this test on one brief from your own queue before you settle on a tool for good. The brief that flips the result for you won’t be the same one that flipped it for me.