ChatGPT Prompts for Marketing, From Brief to Post

The lists of chatgpt prompts for marketing are sorted by channel. Five for content, six for email, six for SEO, six for social, four for video, and every one of them starts from a blank chat with brackets to fill in. OpenAI's own Academy page for marketing teams does the same thing, with "Write a launch email for our new product." as a complete prompt. That's a sentence, not a brief, and the model will happily supply the product.

Marketing prompts that produce usable copy run as a chain. You write the brief first, with every claim you're allowed to make listed next to its evidence. Every later prompt says "use only the brief". The post comes out of the messaging, the messaging comes out of the brief, and the last prompt reads the finished copy back against the brief to catch what slipped in. Seven prompts below do that in order, shown in full, and the one I'd copy even if you ignore the chain is the brief, because it's the thing none of the lists ask you to write.

Two guards run through all seven. The model isn't allowed to invent a number about your product, and it isn't allowed to invent a person who used it. The second one became a federal rule in the US in 2024, which is further down.

What Marketing Copy Invents By Default

Ask for a launch email with no brief and you get a launch email. It'll have a saving in it, a percentage, a happy customer with a first name and a job title, a deadline, and a phrase like "trusted by thousands of teams". None of those came from you. The model has read a great deal of marketing copy and knows what shape it has, and the shape includes claims.

Here's the list as I'd write it, with the line that stops each one. These lines are in the prompts below; the table is so you can add them to prompts of your own.

The customer row is the one with teeth. In August 2024 the US Federal Trade Commission announced its final rule on the use of consumer reviews and testimonials, and the press release names "AI-generated fake reviews" in the first prohibition. It covers reviews and testimonials that misrepresent who wrote them or whether they had the experience described, and it applies to a business that buys or publishes such content when it "knew or should have known" the content was fake. The rule also requires insiders who give testimonials to disclose their connection to the company, and it gives the agency civil penalty authority. It's a US rule, and I'm not a lawyer. What I can say is that "write three customer testimonials for our landing page" is a prompt whose output now has a name in a regulation. Don't run it.

The Chain

Seven stages. Each one takes the previous stage's output as its input, and the right-hand column is what it's forbidden from adding.

The reason it's a chain and not a list is stage 7. A standalone prompt has nothing to be audited against. A chain has the brief, and the audit is a diff between what you allowed and what the model wrote. If I could keep only two of the seven, I'd keep the first and the last.

The Reddit thread at the top of the results page makes the same mistake one level up. Its first prompt asks the model to analyse your business model from a bracketed description of the offer, pricing, customer path and average customer value, and return a top ten of where to push. Fill the brackets thinly and the model analyses a business it imagined.

Stage 1, The Brief

Nothing gets written until this exists. The prompt is an interview, the role is a note-taker, and the output has a claims register at the bottom that every later prompt cites by number.

You are helping me write a marketing brief. Do not write any copy. Interview me one question at a time, in this order: what the product or offer is, in my words; who it is for, described only as I describe them; what they are doing today instead of using it; the one change it makes for them; the channel this campaign runs on and the single action we want; the tone, with two examples of copy I like and one I don't; anything we must not say, for legal, brand, or accuracy reasons.

Then build a claims register. Ask me for every factual claim I am willing to make about the product: features, numbers, prices, comparisons, customer quotes, counts. For each one record: a number, the claim in my words, and the evidence (a document, a dashboard, a named person who said it, a published source, or "none"). Do not upgrade my wording. If I give you a claim with no evidence, record it with "none" and flag it.

Stop after 20 claims or when I say "done". Print the brief in seven short sections followed by the numbered claims register.

"Described only as I describe them" is the audience guard. Left to itself the model writes an audience out of stock photos. The evidence column is where the honesty lives, and "none" is an allowed value on purpose, because the point isn't to stop you making a claim, it's to make you look at it before a prompt turns it into a headline. Every register entry with "none" beside it is a decision you'll make once, here, instead of arguing with a draft later. My rule is that a "none" claim doesn't go in an ad, whatever else it ends up in.

The two examples of copy you like and the one you don't are the fastest way I know to set tone without adjectives. If you want to go further and give the model a description of your brand voice built from your own samples, the method is in chatgpt prompts for writing, and the voice block it produces slots in at the top of every prompt below.

Stage 2, Messaging

Using only the brief below, write three messaging angles for this campaign. Each angle is: a one-line promise in the customer's words, the register claims that support it by number, and the objection a skeptical reader would raise in one sentence. An angle that cannot cite at least two register claims is not allowed; say so instead of writing it. Do not write headlines or copy. Do not add claims.
Brief:
[PASTE]

Three angles because the first one the model produces is the obvious one and, in my judgment, the third is usually the most specific. The "at least two register claims" rule is what stops an angle being a vibe. And the objection line is there because you'll pick the angle by its objection as often as by its promise. An angle with an objection you can't answer from the register is an angle you can't run.

Stage 3, The Draft

Using only the brief and the angle below, write a blog post of [N] words for the channel named in the brief. After every factual statement, put the register number in brackets, like [CLAIM 4]. Where the draft wants to make a claim that is not in the register, write [NO CLAIM] in its place and continue. Do not write a testimonial, quote, or customer story. Do not add numbers. Use the tone examples in the brief and none of the words in its must-not-say list. No closing summary paragraph.
Brief:
[PASTE]
Angle:
[PASTE]

The markers are the whole point and they come out before publishing. A draft that comes back with [NO CLAIM] four times is telling you the angle needs claims you don't have. A draft with no markers at all, on the other hand, is a draft that stayed vague to avoid citing anything, and that's a different problem with the same cause. The results page's version of this prompt is the WBS one, "Write a 700-word blog post for our company on [BLOG TITLE]" after a line of USPs, and it's a reasonable prompt that can't be audited, because the USPs were never separated from the evidence.

Stage 4, Ad Copy Cut To The Fields

Google's help page for responsive search ads gives the fields. Between 3 and 15 headlines of up to 30 characters, between 2 and 4 descriptions of up to 90 characters each, and the advice to write the first three headlines "as if they'll appear together in your ad". The same page says Headline 1 and Description 1 usually show, while the second and third headlines and the second description aren't guaranteed to appear. Every ad copy prompt on the results page asks for "5 ad copy variations". The fields don't want variations, they want assets.

Using only the brief and the angle below, write assets for a Google responsive search ad. Produce 15 headlines of at most 30 characters each and 4 descriptions of at most 90 characters each. Show the character count after each one. Headlines 1 to 3 must read as a set, since they may appear together; the rest must each be distinct in meaning, not rewordings. Each headline and description cites a register number in brackets after the count. No urgency, no superlatives, no counts of customers unless they are in the register. Do not pin anything or suggest pinning.
Brief:
[PASTE]
Angle:
[PASTE]

The character counts are there because I don't trust the model's counting and you'll check anyway, and it's faster to check a number than to count. "Distinct in meaning" comes from Google's own advice to provide as many unique headlines as you can, which the model will otherwise satisfy with fifteen ways to say the same thing. The pinning line exists because the help page says pinning isn't recommended for most advertisers and can affect ad strength, and left alone the model likes to suggest it.

Stage 5, The Launch Email

Using only the brief and the angle below, write a launch email. Subject line, preview text, and a body of four short paragraphs. Paragraph one states what changed for the reader, citing one register claim. Paragraph two states what the product does, citing one. Paragraph three answers the objection from the angle, citing one. Paragraph four is the single action from the brief and nothing else. No deadline unless the brief gives one with a date. No "trusted by", no customer names, no exclamation marks. Put the register number in brackets after each paragraph.
Brief:
[PASTE]
Angle:
[PASTE]

One claim per paragraph is a constraint that makes the email shorter and more checkable at the same time. The Smart Insights page on the results page teaches a three-step email method, set the context, define the task, define the role, and that's fine as far as it goes. The step it's missing is the one where the claims come from somewhere. OpenAI's own resource lists "Write a launch email for our new product." as a prompt in full, which tells you how much of the work the model is expected to guess.

Stage 6, The Social Series

Using only the brief and the draft below, write a three-post series for the channel in the brief. Each post is built on a different register claim and names it in brackets at the end. Post one states the change for the reader. Post two shows the feature that makes it. Post three answers the objection. No questions to the audience, no "you won't believe", no invented scenarios about the reader's day, no hashtags unless the brief lists them. Plain sentences in the brief's tone.
Brief:
[PASTE]
Draft:
[PASTE]

Social copy is where the audience-fact invention lives, the "we all know the feeling when..." opener that describes a person the brief never mentioned. "No invented scenarios about the reader's day" is the specific line for it. OpenAI's resource suggests "Draft a 3-post social media series promoting [event, product, or milestone]", and the difference between that and this is the word "only" and the register numbers.

Stage 7, The Audit

Below is finished marketing copy and the brief it was written from. List every factual statement in the copy, quoted, one per line. Next to each, write the register number that supports it, or NOT IN REGISTER if none does, or CONTRADICTS if the register says otherwise. Then list any sentence that implies a claim without stating one (a comparison, a superlative, a count, an implied deadline) and mark it IMPLIED. Do not fix anything. Do not comment on quality. Report only.
Copy:
[PASTE]
Brief:
[PASTE]

Run this on everything, including copy you wrote yourself. The IMPLIED category is the useful one. "The simplest way to..." is a comparison with no register entry, and it reads as harmless until someone asks you to show the comparison. In my judgment this prompt catches more than the writing prompts prevent, because the writing prompts can only guard against inventions you thought to name.

Questions From The Results Page

What are some good ChatGPT marketing prompts? The brief prompt, then the audit prompt. Those two make every other prompt honest, and they work with the lists on the results page as well as with the chain here, since any prompt can be told "use only the brief" and any output can be audited.

What are the top 10 prompts for ChatGPT, and what are some good prompts generally? The six slots that make any prompt work are in chatgpt prompts, and the reusable versions with the variable slots marked are in chat gpt prompts.

Can I use ChatGPT for marketing? Yes, and OpenAI publishes a marketing resource for its work users that recommends reasoning models for strategy, web search or deep research for competitive work, canvas for editing copy, and image generation for moodboards. What that resource doesn't say, anywhere I could find, is that the model invents claims, which is the only reason this article exists. Use it with a brief and a claims register and it's a fast, tireless junior who never makes anything up. Use it without one and it's a fast, tireless junior who does.

What are some good business prompts for ChatGPT? Owner-level work, pricing, operations, hiring, is a different set of prompts with different guards, and it's in chatgpt prompts for business.

What are some effective prompts for ChatGPT to make money? I sell seven digital toolkits priced between $29 and $49, and the first one sold roughly two months after launch. No prompt did that. The prompts wrote listing copy faster, and the copy was audited against what the toolkits actually contain, which is the register idea applied to a product page. A prompt is a way to write faster. What you're writing about still has to exist.

Is there a PDF, and are the prompts free? Everything above is free and copies cleanly into a notes file, which is where I'd keep it. A PDF of prompts goes stale the first time you edit one.

What I'd Do First

Run the brief interview for one product tonight. Answer honestly, especially in the evidence column, and count how many claims end up with "none" beside them.

That count is the most useful number you'll get from this. It's how much of your current marketing is a sentence the model would have been happy to write for you, and it's the list of things to go and get evidence for before the next campaign, in whichever order the audit prompt puts them.