You pasted "write me a Facebook ad for my product" into ChatGPT and got back the same thing everyone gets: a headline with an exclamation mark, three benefits nobody asked about, and a closing line that says "Don't miss out!" It's grammatically perfect and completely dead. You can't run it, because it sounds like an ad about ads.

The problem isn't ChatGPT. It's the brief. You handed it a blank, and it filled the blank with an average of every ad it has ever seen — which is exactly what "generic" means.

What makes a ChatGPT ad copy prompt work

The best ChatGPT ad copy prompts don't ask for "an ad." They hand the model the four things a real copywriter needs before writing a single word: who the customer is and the moment they see this, the one problem they're trying to solve, the specific outcome they want, and the single action the ad should drive. Then they force one angle per output — pain-led, proof-led, or curiosity-led — instead of a mushy blend of all three, and ban the clichés the model reaches for by default. Feed it that and you get copy you can test. An ad prompt is a briefing format, not a magic phrase.

Why "write me an ad" fails every time

A language model predicts the most likely next words. Ask for "a Facebook ad" with no context and the most likely words are the most common ones — the beige middle of a million mediocre ads. Vague in, average out.

Left alone, ChatGPT writes ad copy the way a nervous founder does: it lists what the product is. Features, adjectives, a call to action bolted on the end. "Introducing the smartest way to manage your team's workflow." Nobody stops scrolling for that, because it's about you — and the reader is not thinking about you. They're thinking about their own problem.

Specificity is the fix, but not random specificity. You need the specificity a buyer feels. "Busy" is filler. "Opens the app at 11pm, sees 40 unread Slack messages, closes it again without replying" is a person. ChatGPT can write sharp copy. It just can't invent the details of your customer's life. That part is your job, and it's the whole job.

The four inputs every ad prompt needs

Get these straight before you write any prompt. Skip one and the copy sags.

  • Audience — not a demographic, a moment. Who are they, and what were they doing five seconds before they saw this?
  • Problem — the one pain this ad speaks to, in their words, not your marketing deck's.
  • Outcome — the specific after-state they're buying ("close the month in an afternoon"), not the feature that delivers it ("automated reconciliation").
  • Action — the single next click. One ad, one ask.

Then pick an angle, because the same four inputs can be sold three ways: lead with the pain, lead with proof, or lead with curiosity. Good advertisers don't guess which wins — they run all three and let the numbers talk.

The anatomy of an ad that converts

Strip a high-performing ad down and you find four parts, in this order:

  1. Hook — names the reader's problem or desire in their own words, so they feel seen in the first line.
  2. Tension — why it matters, what it's costing them, what they're missing.
  3. Turn — your product as the resolution, stated as an outcome, not a feature.
  4. Ask — one action, framed as easy.

Most AI ad copy has parts three and four and skips one and two entirely. That's why it reads like a brochure. The prompts below force all four in order, because the order is the mechanism — lead with the product and you've lost the reader before they reach the good part.

The prompts

Copy the one that fits your problem. Each is followed by one line on why it's built that way and what to swap in. All of these work in ChatGPT, Claude, Gemini and Grok.

1. The core brief (Meta primary text)

You're a direct-response copywriter. Write 3 versions of Facebook primary
text for the product below. Each version leads with a DIFFERENT angle:
(1) pain-led, (2) proof-led, (3) curiosity-led. Under 100 words each,
one idea per sentence. No hype words ("unlock", "introducing",
"revolutionary", "game-changer"), no exclamation marks. End each with
the exact CTA.

Product: [what it is]
Audience: [who they are + the moment they see this ad]
Problem: [the one pain, in their words]
Outcome: [the specific after-state]
CTA: [the single action]

Why it's built this way: it forces three testable angles instead of one blend, and the banned-word list kills the tells that make copy sound generated. Swap in: your four inputs — and keep "one idea per sentence," which is what makes copy scannable in-feed.

Before (from "write me a Facebook ad for my bookkeeping app"):

Introducing LedgerFlow — the smartest way to manage your small business finances. Save time and stay organized. Try it today!

After (from the prompt above):

It's the 4th and you still haven't closed last month's books. Every day you put it off, the pile gets scarier and the numbers get fuzzier. LedgerFlow closes your month in an afternoon. See your first month reconciled free →

2. The hook generator

Give me 10 scroll-stopping first lines for a [platform] ad. Each must do
ONE of these: call out the exact audience, name a specific pain, or make
a claim too specific to sound fake. No questions, no "Are you tired of".
Under 12 words each.

Audience: [who]
Problem: [pain]

Why it's built this way: the first line is 80% of the job — if the hook doesn't stop the scroll, nothing else gets read. Generating hooks in isolation beats settling for whatever opener a full-ad prompt hands you. Banning "Are you tired of" alone removes the most overused opener on the internet. Swap in: your audience, named as narrowly as you can.

Before: "Elevate your morning routine and unlock a more productive you."

After: "Most morning routines are just Instagram with extra steps."

3. The angle multiplier

Same product, 5 completely different sales angles. For each: name the
angle, the emotion it targets, and a 1-sentence hook. Angles must be
genuinely different (status, fear of loss, saving time, belonging,
identity) — not 5 rewordings of the same benefit.

Product + audience: [context]

Why it's built this way: most people test the same message in different fonts and call it A/B testing. The biggest lever in ad testing is the angle, not the wording. Swap in: your product — and keep the "genuinely different" constraint or it'll hand you five synonyms.

4. The rewrite (diagnose first)

Here's an ad that's underperforming. First, tell me in 3 bullets exactly
why it's weak — be blunt. Then rewrite it 3 ways, each fixing a different
weakness. Keep my product facts accurate; don't invent features.

Current ad: [paste it]
Audience + problem: [context]

Why it's built this way: making the model diagnose before it rewrites stops it reshuffling your words, and you learn the failure pattern instead of just getting a patch. The diagnosis is almost always "it opens with the product instead of the reader." Swap in: your actual losing ad, plus real audience context — without it the fixes come out generic.

Before: "Introducing GlowSerum — our revolutionary vitamin C formula for radiant skin! Packed with antioxidants. Shop now!"

After: "You've tried four serums. Your skin still looks tired by 3pm. This one's a single drop of stabilized vitamin C — the kind that doesn't oxidize into nothing by lunch. See it in two weeks or send it back."

The rewrite works because it names a real frustration, drops one concrete mechanism, and removes the risk. No adjectives doing the heavy lifting.

5. The Google RSA generator

Write a Google Responsive Search Ad. 15 headlines (max 30 characters each)
and 4 descriptions (max 90 characters each). Include the keyword "[keyword]"
in at least 5 headlines. Mix benefit headlines, feature headlines, and CTA
headlines. Count characters and stay under the limits.

Product: [what it is]
Keyword: [target keyword]
Offer: [any promo]

Why it's built this way: search ads match intent rather than persuade, so the character caps and keyword anchor keep the model literal instead of clever. Google rejects anything over the limit, and stating the spec up front saves a round of manual trimming. Swap in: the actual query people type — the closer your headline mirrors it, the higher it clicks. Always recount before uploading; ChatGPT's character counting is good, not perfect.

6. The objection crusher

List the top 5 reasons someone in my audience would NOT buy this. Then
write one short ad that dismantles the single biggest objection without
sounding defensive. Lead with the objection — don't hide from it.

Product + audience + price: [context]

Why it's built this way: the fastest-converting ads often name the objection out loud ("Too expensive? Here's the math."). Surfacing objections first means the ad answers a real doubt instead of an imaginary one. Swap in: your price — objections change completely at $9 versus $900.

7. The short-form video hook

Write 5 opening lines (first 3 seconds) for a short-form video ad, plus
one line of on-screen text for each. The spoken line and the text must
say DIFFERENT things so both earn attention. Native and casual, not
salesy. No "POV:" openers.

Product + audience: [context]

Why it's built this way: on TikTok and Reels the first three seconds decide everything, and the spoken hook plus on-screen text are two separate weapons — making them differ doubles the information in second one. Swap in: your product, and ban whatever opener is already burnt out in your niche.

8. The one-detail rule

Rewrite this ad using exactly ONE concrete, specific detail that only
someone who really knows this customer would know. Remove every adjective
that isn't doing real work. Keep it under 50 words.

Ad: [paste it]
Customer: [who they are]

Why it's built this way: one true detail beats five adjectives. This is the polish pass you run after the angle is settled — it's what separates copy that sounds researched from copy that sounds generated. Swap in: nothing. Keep it as a finishing step and run it on whatever won your test.

Before: "Discover the all-in-one platform that revolutionizes how your team works."

After: "You reopened the same three tabs to send one invoice. Again."

Which prompt when

Your problem Use Platform What it fixes
Starting from scratch #1 Core brief Meta / Facebook Gives you three angles to test, not one guess
Impressions high, clicks low #2 Hook generator Meta / TikTok A first line that stops the scroll
You've only ever tested one message #3 Angle multiplier Any Different reasons to buy, not different fonts
An ad that used to work is dying #4 Rewrite Any Diagnoses the weakness before patching it
Running search ads #5 Google RSA Google Copy that fits the spec and matches intent
Clicks are fine, sales aren't #6 Objection crusher Meta / Google The doubt that's stopping the purchase
Running Reels or UGC #7 Video hook TikTok / Reels The three seconds that decide everything
Copy is close but sounds generated #8 One-detail rule Any Replaces adjectives with something true

Rule of thumb: if people aren't clicking, your hook is broken — start with #2. If they click but don't buy, your angle or offer is broken — start with #6.

Before you paste anything into your ad account

ChatGPT writes the draft; it doesn't run your account. Three habits worth keeping:

Fact-check every claim. The model will happily invent a feature to make a sentence land. If your product doesn't do it, cut it.

Read it out loud. If it sounds like a robot doing a TV-announcer voice, kill it. Your ear catches the AI tells faster than your eye does.

Never ship one version. The entire point of these prompts is to give you angles to test. Upload at least three and let real clicks decide which one you were wrong about.

Related guides

Run three angles by tonight

Pick the prompt that matches your actual problem, then spend your effort on one input: the reader's problem, written the way they'd say it. That single line is what separates an ad from a description, and it's the one thing ChatGPT can't invent for you — because it doesn't know your customer. You do.

→ ChatGPT Prompts for Ad Copy — the full hand-curated set, ready to copy.