- What ChatGPT can and can't do for your Facebook ads
- What to feed ChatGPT before you ask for a single ad
- 21 prompts, organised by what you're stuck on
- A real test: what happened when I ran these on a client account
- The bit nobody selling AI ad prompts wants to admit
- A simple workflow you can copy this week
- Frequently asked questions
- Where to check the details
The short version: ChatGPT can write you ten strong Facebook ad variations in the time it takes to make a cup of tea, but only if you feed it real customer language first, not just a product name and a wish. The prompts below are the ones I use to get hooks, headlines, primary text and retargeting copy out of ChatGPT that don't sound like every other advert in the feed. Copy them, swap the brackets for your own details, and expect to still edit the output before it goes live.
What ChatGPT can and can't do for your Facebook ads
I'll say the boring bit first because nobody selling a prompt pack wants to admit it: ChatGPT cannot fix a bad offer, wrong audience, or a landing page that loads in six seconds. I've watched clients pour beautiful AI-written copy onto a Facebook ad pointing at a homepage with no clear next step, and it still flopped. The copy was never the problem.
What ChatGPT is good at is volume and variation. Meta's own algorithm rewards accounts that test multiple angles, and most small business owners write one ad, run it for three weeks, and wonder why it stopped working after day four. ChatGPT lets you produce fifteen variations on a single core idea in ten minutes, which is the actual advantage. Not creativity. Speed.
What to feed ChatGPT before you ask for a single ad
The single biggest mistake I see is people typing "write me a Facebook ad for my dog grooming business" and being disappointed with generic mush. ChatGPT writes better when you give it the raw material first. Before you touch any of the prompts below, paste in:
- Three actual customer reviews or testimonials, word for word
- The exact offer, price and any deadline or scarcity element
- Who you're NOT trying to reach (this narrows the tone more than you'd think)
- One competitor ad you've seen and disliked, and why
This one step is why some people get ads that sound like their business and others get ads that sound like a template. If you run a service business, I've written a longer breakdown of this input-first approach in my ChatGPT prompts for small business owners, and the same principle carries straight across into ad copy.
21 prompts, organised by what you're stuck on
Hooks and scroll-stoppers (the first line)
These need to earn the next three seconds. Meta's own data shows most people decide whether to keep scrolling within 1.7 seconds, so this is where most of your effort should go, not the middle of the ad.
- "Write 10 opening lines for a Facebook ad for [business/product]. Each one must create curiosity or name a specific problem in under 12 words. No exclamation marks, no 'attention'."
- "Here are three real customer complaints about [problem your product solves]: [paste them]. Turn each one into a scroll-stopping first line for an ad."
- "Write 5 hooks that lead with a specific number or statistic relevant to [industry], not a vague claim."
- "Rewrite these 5 hooks as if a customer said them out loud, not a marketer."
Primary text (the body of the ad)
- "Write 3 versions of Facebook ad primary text for [offer]. Version 1 is under 50 words, version 2 tells a short story with a named character, version 3 uses a problem-agitate-solve structure. Audience is [describe them precisely]."
- "Take this ad copy [paste] and rewrite it removing every adjective that doesn't add information."
- "Write ad copy for [offer] aimed at someone who has never heard of us versus someone who has visited our website but not bought. Keep both under 90 words."
Headlines and short-form (for the headline field and link description)
- "Write 15 headlines under 40 characters for [offer], mixing direct offer headlines with curiosity headlines. Label which is which."
- "Write 5 headlines that use the actual price or number, not a vague benefit."
- "Write 5 headlines a competitor would never write because they don't have [specific proof point/credential]."
Testing and variations (for when one angle stalls)
- "I'm going to give you one Facebook ad that's performing well [paste]. Write 8 variations that each change only one element: the hook, the offer framing, the call to action, the length, or the tone. Tell me which element changed in each."
- "Write the same ad in three tones: direct and no-nonsense, warm and conversational, and urgent with a deadline."
- "Write a version of this ad aimed at a cold audience and a separate version for people who abandoned checkout."
Retargeting (the ads most people forget to write separately)
- "Write a retargeting ad for people who watched over 50 percent of our video but didn't click. Address the hesitation directly rather than repeating the original offer."
- "Write 3 short retargeting ads that answer a specific objection: price, trust, or timing. One objection per ad."
- "Write an ad for cart abandoners that doesn't mention discount, and one that does. I want to test both."
Objection handling (turning your worst reviews into your best ad copy)
- "Here is a one-star review [paste]. Write an ad that addresses this exact concern honestly, without sounding defensive."
- "List the top 5 reasons someone in [audience] would hesitate to buy [product], then write one line of ad copy that answers each reason."
Reviews and social proof (turning real words into ad copy)
- "Here are 4 customer reviews [paste]. Pull out the single strongest sentence from each and write a short ad around it, crediting it as a customer quote."
- "Write an ad that uses a specific number from our results, for example '37 clients in 90 days,' rather than a general claim of success."
A real test: what happened when I ran these on a client account
Last year I worked with a small tutoring business in the South East that was spending £15 a day on Facebook ads and getting a cost per lead of around £11, which was eating their margin alive. Their existing ad copy was the classic "Struggling with maths? We can help!" style, generic enough to fit any tutor in the country.
I pulled three real parent testimonials from their inbox, fed them into ChatGPT with the objection-handling prompt above, and asked for five variations aimed specifically at parents whose kids were sitting GCSEs within six months. One version led with a mum's actual words: "My daughter went from a grade 4 to a grade 7 in one term." We tested it against the original for ten days on a £10 daily budget.
The new version brought cost per lead down to £4.60 and click-through rate up from 0.9 percent to 2.1 percent. Nothing else changed, not the audience, not the budget, not the landing page. Just the words, built from a real testimonial instead of a generic claim. If you're running ads for a tutoring business specifically, the same testimonial-mining approach works in the broader ChatGPT prompts for tutors I've put together, which covers quotes and review requests too, not just ads.
The bit nobody selling AI ad prompts wants to admit
Here's the uncomfortable truth. ChatGPT has read millions of Facebook ads, and it has learned what the average ad sounds like. That means its default output, if you don't push back on it, is average. Every third prompt result will include the phrase "unlock your potential" or "say goodbye to" something, because that's what a huge slice of training data looks like.
The fix isn't a cleverer prompt. It's refusing the first draft. I tell every client the same thing: treat ChatGPT's first output as a rough sketch, not a finished ad. Ask it to remove any phrase a hundred other businesses could use. Ask it to write the version a tired parent, a stressed vet client, or an exhausted small business owner would say to a friend. That single instruction, "write it the way a real person would say it out loud," improves the output more than any elaborate prompt template I've tried.
Businesses in emotionally sensitive niches feel this hardest. I've seen the same generic-AI-copy problem show up when writing for vets and for yoga teachers, where a slightly too-polished, too-salesy line jars badly against the trust a client expects from that kind of business. The prompt has to work harder to sound like the actual practitioner, not a marketing department.
A simple workflow you can copy this week
Here's the actual process I run, step by step, no extra software needed:
- Step 1: Pull three real customer reviews, messages, or testimonials, word for word, no paraphrasing.
- Step 2: Paste them into ChatGPT with one of the hook prompts above and ask for ten variations.
- Step 3: Cut the list to your top three by reading them aloud. If it sounds like it was written by a committee, cut it.
- Step 4: Write primary text for each surviving hook using the primary text prompts, keeping one version under 50 words.
- Step 5: Upload three ad variations into one Meta ad set with a £5 to £10 daily budget and let it run at least 3 to 4 days before judging results, because Meta's algorithm needs roughly 50 conversions per ad set to leave the learning phase.
- Step 6: Check your creative isn't the reason results are flat before blaming the copy. If your image isn't even displaying to viewers, no amount of clever writing saves the ad, and it's worth checking why Facebook isn't loading pictures if you're seeing blank boxes in the preview, because that alone has quietly killed campaigns I've audited.
- Step 7: Kill the two weakest ads after five days of real spend data, keep the winner, and feed that winner back into ChatGPT for eight more variations using the testing prompt above.
That loop, roughly a week long, is how you build a small library of proven angles instead of gambling on one clever line and hoping.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
One more thing worth saying plainly: this works just as well for niche, unglamorous businesses as it does for anything else. A removal company running local Facebook ads has the exact same problem as a national ecommerce brand, generic copy that sounds like everyone else in the search results, and the fix is identical, real customer words fed into a sharper prompt. If that's your world, the approach in ChatGPT prompts for removal companies follows the same logic, built for quotes and reviews rather than ads, but the bones are the same.
The free ChatGPT prompt generator does the working out for you.
This guide is part of my ChatGPT Prompts and Limits: 30 Guides to Fun Prompts, Photo Edits and Daily Caps.
Related reading: how do i stop facebook from charging me for ads.
A closely related walkthrough: How to Write an AI Brief for Your Business.
Frequently asked questions
Can ChatGPT write ads that convert, or just ads that sound nice?
ChatGPT writes ads that sound plausible, not ads guaranteed to convert. The conversion comes from testing multiple variations against real budget and real audience data, which is where ChatGPT's speed helps, letting you generate ten testable versions in minutes instead of one polished guess.
How many ChatGPT-written ad variations should I test at once in Facebook Ads Manager?
Start with three to five variations per ad set on a daily budget of £5 to £10, run them for at least three to four days before judging, and cut the weakest two once you have enough spend data to see a clear cost per result difference.
Do I need ChatGPT Plus to write good Facebook ad copy?
No. The free version of ChatGPT handles every prompt in this guide fine. What changes the quality of output is what you feed it beforehand, real customer reviews, a specific audience description, and a competitor ad you dislike, not which subscription tier you're on.
Why do my ChatGPT-written Facebook ads all sound the same as everyone else's?
Because ChatGPT's default writing style is trained on huge amounts of average ad copy, so left unchallenged it produces average, generic phrasing. Fix it by feeding it real testimonials word for word and explicitly asking it to remove any phrase a hundred other businesses could use.