How to Use ChatGPT for Copywriting: The Brief-It-Like-a-Junior-Writer Workflow

You typed "write me a sales email for my course" into ChatGPT and got back three paragraphs that could be selling literally anything. That's not a ChatGPT problem, it's a briefing problem. Use ChatGPT for copywriting by handing it the same four things you'd hand a new copywriter on day one: the avatar, the offer, your voice, and a framework to fill.

To use ChatGPT for copywriting well, brief it like a new hire instead of asking it to "write an ad." Give it the avatar (a specific reader, not a demographic), the offer (product, price, mechanism, real proof), your voice (a sample it can match), and a named framework to fill in. Skip any one of the four and you get a generic draft, no matter how good the model is.

This matters because most operators blame the model when the real issue is the input. ChatGPT was built to be safely, broadly helpful across millions of different requests, so its default is the statistical middle of every sales email, ad, and landing page it's ever seen: hedged, generic, and nobody's voice in particular. What decides whether you get a usable draft or filler isn't which version of ChatGPT you're on, it's whether you show up with the same four things you'd hand a junior writer before their first assignment. Everything below builds that brief piece by piece, then hands you the full prompt.

Why does ChatGPT need a brief instead of a request?

Because without one, it fills every gap you leave with the safest, most average version of whatever you asked for, the same way a brand-new hire would if you handed them a sticky note that just said "write something."

ChatGPT has no idea who your reader is, what your offer actually does, what you sound like, or what structure you want, unless you tell it. Left to fill those blanks itself, it fills them with averages: a rhetorical-question opener, three benefits marching in a tidy row, a CTA that says "don't miss out." None of that is wrong exactly. It's just nobody's voice and nobody's specific offer, because you never told it whose voice or which offer.

You wouldn't hand a new copywriter at an agency a sticky note and expect a usable draft back. You'd hand them a brief: who we're talking to, what we're selling, how we sound, what structure the client likes. ChatGPT isn't smarter than that new hire about your business. It's faster, and it never gets tired of a ninth revision, but it starts from zero every time you open a blank chat unless you've set up the pieces that carry the brief forward, which is what the voice section below covers.

That brief has four parts, and they're the order this whole piece follows: the avatar, the offer, your voice, and a framework. Get those four right and the model has almost nowhere left to guess. Leave one out and it guesses there instead, every time.

What does ChatGPT need to know about the reader before it can write a real opening line?

A specific person in a specific moment, described in language they'd actually use, not a demographic label.

"Freelance consultants" is a demographic. "A consultant who just lost his biggest client and is staring at an empty pipeline for next month" is a person mid-crisis. ChatGPT can write to the second one. It can only guess at the first, and the guess reads exactly like a guess.

Three things turn a demographic into a real avatar:

• Who they are, specifically. Not the job title, the situation. What just happened, what they're staring at right now, not what category of person they fall into.
• What they've already tried. Copy that ignores a reader's prior failed attempts reads like it doesn't know them. Naming what didn't work is instant credibility, and ChatGPT will only name it if you feed it in.
• Their own words for the problem, pulled from a DM, a support message, or a comment, not the cleaned-up version you'd put on a testimonial page.

Here's what that looks like worked through. Say you're selling a course that teaches freelance consultants how to land retainer clients instead of chasing one-off gigs, and a DM in your inbox reads: "I keep landing one-off gigs and then scrambling every single month to find the next one." Hand ChatGPT that line as part of the avatar, and the opener stops being "Are you tired of inconsistent income?" and turns into something closer to what that person actually said about their own week.

Run the swap test on whatever ChatGPT gives you back. If the opening line could be sent to a reader of a completely different offer in a completely different niche without changing a word, the avatar wasn't specific enough. Add one more real detail and run it again.

How much offer detail does ChatGPT actually need before it can pitch the product?

Enough that it never has to guess, because every gap you leave, it fills with something plausible-sounding that isn't necessarily true.

Hand it five things: the product name, the price, the mechanism (what it actually does, in one sentence), the transformation (the reader's before state to their after state), and real proof (an actual testimonial, a guarantee, a number you can back up). Leave any of those out and ChatGPT doesn't leave a blank. It writes something that sounds like it belongs there, which is a worse outcome than a blank, because a blank gets caught in editing and an invented detail sometimes doesn't.

The rule underneath all of it: never let ChatGPT invent proof. It will, confidently, and the invented version often sounds more impressive than the real one, which is exactly the problem.

Back to the retainer example. Product: a $197 cohort course. Mechanism: four weeks, live pricing-call practice each week. Transformation: from re-pitching a new client every single month to one retainer conversation that pays out for a year. Proof: a guarantee that refunds the course if the reader hasn't booked one real retainer conversation within 60 days. Feed ChatGPT exactly that. Don't feed it "helps freelancers make more money" and hope the specifics show up on their own, because they won't, not the true ones anyway.

This is also where a lot of drafts fall apart even after the avatar's nailed. Operators get specific about the reader and go vague on the product, and ChatGPT can only be as specific as the thinner of the two inputs. If you haven't landed your own price and structure yet, how to price a cohort course is worth working through before you write a single prompt, because a fuzzy offer produces fuzzy copy no matter how sharp the avatar is.

How do you give ChatGPT your voice without re-explaining it every single chat?

Set it once, somewhere ChatGPT actually holds onto between conversations, instead of pasting a voice sample into every new chat window.

ChatGPT gives you three places to park a standing voice brief instead of retyping it. Custom Instructions is a standing note the model reads before every new chat, good for a quick, permanent rule set you want everywhere. A Project is a named workspace where you can drop instructions and reference files that every chat inside it inherits, useful when you're running an entire launch's worth of copy through one place. A Custom GPT is a saved, purpose-built assistant with its own instructions baked in, worth setting up if "copywriter for my brand" is a role you reach for every week. Pick based on how often you're doing this, not based on which sounds fancier.

Memory is a different thing entirely: it's ChatGPT quietly noting facts it infers you'd want remembered across chats. It's useful, but don't lean on it to hold your full voice brief. It's built to remember that you run a cohort course, not to remember the eleven specific words you've banned from your copy.

Even saved instructions decay inside one long thread. Draft one respects your banned-word list. Draft twenty has quietly wandered back toward the safest, most generic phrasing, because by then there's far more competing context between the model and the instruction you set at the top. If a thread's gone long and the voice is sliding, don't fight it. Start a fresh chat inside the same Project and let the standing instructions do their job from a clean slate.

Which copywriting framework should you hand ChatGPT to fill in?

Whichever one fits the format, named explicitly, because ChatGPT already knows a dozen frameworks and blends them into something generic the moment you don't pick one.

Ask for "persuasive copy" and the model free-associates across every framework it's ever seen, landing somewhere in the middle of all of them. Name one and define what each stage means for this specific offer, not just the acronym, and it has rails instead of vibes.

Define the stages yourself, don't just name the acronym. A bare "use PAS" produces a shallow pass through three labels. A defined stage produces an actual paragraph.

Here's Before-After-Bridge worked through with the retainer example:

• Before: the reader's situation right now, in their own words. That DM line from earlier: scrambling every single month to land the next one-off gig.
• After: the specific outcome. One retainer client signed, income that no longer resets to zero on the first of the month.
• Bridge: the course as the mechanism that closes the gap, backed by the guarantee, not by a vague promise of "more consistent income."

Before-After-Bridge isn't the only option, and it isn't always the right one. A longer sales page that needs to sit in the reader's pain before it pivots often wants the problem-agitate-solve structure instead, and an ad fighting for a scroll-stopping first line leans more on AIDA's attention step. For the fuller menu of which framework fits which situation, this breakdown of copywriting frameworks covers the tradeoffs.

The format changes how much you expand each stage, not which four inputs you're feeding in. A subject line compresses the whole Before line into a few words and a hook. A full landing page expands each stage into its own section. An ad needs five or six short variants off the same Before line instead of one polished pass. Same brief, scaled to what you're actually writing.

How do you edit ChatGPT's first draft instead of accepting it?

Treat it like a junior writer's first pass: mark up the specific lines that are wrong and ask for a fix, instead of regenerating from scratch or shipping it as-is.

A new copywriter's first draft is a start, not a final. You wouldn't fire them over it, and you wouldn't publish it either. You'd circle the weak paragraph and say exactly what's wrong with it. ChatGPT responds the same way: a specific note gets a specific fix, a vague note gets a different flavor of the same generic draft.

A short list of tells worth hunting for in what comes back, because they show up often enough to name directly:

• An inflated verb standing in for a plain description of what the thing does.
• A rhetorical-question opener instead of just stating the point.
• Three benefits marching in a row whether or not three is actually the right number for this offer.
• A hedge stacked in front of a claim that should just be stated plainly.
• A closing line that summarizes the whole email again, like it's worried you weren't paying attention the first time.

When you see one, don't say "make it better." Say "cut the second paragraph, it just restates the first one" or "the CTA hedges, make it one direct sentence." Then read the whole thing out loud before you touch anything else. Silent reading skips over the exact rhythm problems and stiff phrasing that reading aloud catches in seconds.

This is a big enough problem on its own that it's worth its own read: how to write copy with AI without sounding like AI goes deeper on the specific tells and the edit pass that catches what slips through even after a good brief.

What's the exact ChatGPT prompt that briefs it like a junior copywriter?

Here's the full brief in one prompt, avatar, offer, voice, and framework, in the order this piece walked through, with your own details dropped into the brackets.

You are a direct-response copywriter working for [BRAND NAME]. Plain, human,
operator voice. Contractions. No hype, no fake urgency, no claim you can't
back up.

THE AVATAR (who this is for)
- Who they are, specifically, not a demographic: [ONE SENTENCE ON THE
  SPECIFIC PERSON AND THEIR SITUATION RIGHT NOW]
- What they've already tried that didn't work: [WHAT THEY TRIED]
- Their own words for the problem, pulled from a DM, review, or support
  message: "[PASTE THE REAL QUOTE]"

THE OFFER (what we're selling)
- Product: [PRODUCT NAME]
- Price: [PRICE]
- Mechanism, one sentence: [WHAT IT ACTUALLY DOES]
- Transformation: [BEFORE STATE] to [AFTER STATE]
- Proof, nothing invented: [REAL TESTIMONIAL, GUARANTEE, OR NUMBER]

MY VOICE
[PASTE 3 TO 5 SENTENCES OF SOMETHING YOU ACTUALLY WROTE, OR YOUR SAVED
VOICE RULES IF YOU'VE ALREADY BUILT THEM]

THE FRAMEWORK
Write this as a [BEFORE-AFTER-BRIDGE / PAS / AIDA] [EMAIL / LANDING PAGE
SECTION / AD]:
1. [FIRST STAGE]: [WHAT THIS STAGE SHOULD DO, POINTING AT WHICH INPUT
   ABOVE IT PULLS FROM]
2. [SECOND STAGE]: [SAME]
3. [THIRD STAGE]: [SAME]

RULES
- No em dashes. No rhetorical questions. No "unlock," "elevate,"
  "transform," or "supercharge."
- Vary sentence length. Nothing longer than 20 words in a row.
- Use at least one phrase from the avatar's own words above, verbatim.
- One clear call to action: [SPECIFIC CTA].

Target length: [WORD COUNT].

After the draft, list anything you had to guess or invent that I didn't
give you above, so I can check it before this goes anywhere.

Run it once, then treat what comes back as a new hire's first draft, not a finished page. Work through the fact-check list it gives you before you touch the edit pass covered above.

Where does this workflow break down?

It doesn't know your funnel. ChatGPT can write a strong individual email or landing page section, but it has no idea what your last email promised or what page comes after this one unless you paste that context in yourself. Even with Memory turned on, it isn't reliably holding funnel-level continuity across separate chats. Feed it that context explicitly or the piece reads disconnected even when each part is well written on its own.

It invents proof the moment you leave room for it. Skip pasting a real number, testimonial, or guarantee, and ChatGPT fills the gap with something plausible-sounding that may not be true. Never publish a specific claim from a draft you haven't traced back to something real yourself. That's not optional in a regulated niche like health, finance, or anything with an income claim attached, where an invented number is a liability, not just an embarrassing edit.

The brief decays inside a long thread no matter how well you set it up. Custom Instructions and Project-level rules help, but forty messages into one conversation, the early context has far more competing material to get lost in than it did at message one. For anything that matters, start a fresh chat and re-paste the brief rather than trusting a long thread to still be holding it.

And no prompt replaces a real test. This workflow gets you a fast, specific first draft, not a winning ad on the first try. Real traffic against real variations is still the only thing that tells you what actually converts. ChatGPT drafts. It doesn't decide.

Frequently Asked Questions

Is ChatGPT actually good enough to write copy that converts, or do you still need a copywriter?

Yes, with the brief covered above, ChatGPT gets you a genuinely usable first draft, not just a starting point you delete and rewrite. What it doesn't replace is the judgment call on what to test next, how to read a result, or whether the offer itself is the real problem, because no draft fixes a weak offer. Treat it as a fast junior writer, not a strategist.

Do you need to know a copywriting framework before you use ChatGPT?

No, but naming one, Before-After-Bridge, PAS, AIDA, whichever fits the format, produces a noticeably tighter draft than asking ChatGPT to "write persuasively." Pick one, define what each stage means for your specific offer, and keep using it so both you and the model get faster with reps.

How is using ChatGPT for copywriting different from using Claude?

Mechanically, the four-part brief works the same in both. The practical difference shows up on longer drafts: ChatGPT tends to drift back toward generic phrasing sooner in a long thread than Claude does, which is exactly why the Custom Instructions and Projects setup above matters more if ChatGPT is your main tool. See Claude vs ChatGPT for copywriting for the fuller task-by-task breakdown.

How much should you edit what ChatGPT gives you before it's ready to publish?

Plan on at least one real editing pass, every time. Read the draft aloud, cut the tells covered above, and check the fact-check list the prompt asks for against what's actually true. A brief this specific gets you closer to a clean draft than a bare request would, but closer isn't the same as done.

Will copy drafted with ChatGPT get penalized by Google?

Not for being AI-assisted on its own. What actually hurts rankings and conversions alike is copy that reads generic, unverified, or templated, which is exactly what a bare request produces and a real brief avoids. For the fuller answer on how AI-assisted content actually gets treated, see will Google penalize AI content.

Where to take this next

The brief above gets you a strong first draft tonight. What separates operators shipping copy that converts from the ones still typing "write me an ad" into a blank chat is reps, running this same four-part brief across enough emails, pages, and ads that naming the avatar and picking a framework stops feeling like a step and starts feeling like how you think. If you want to build that next to people running real copy against real offers every day, come work through it inside the Asset Academy Skool community.