How to Make Facebook Ads With ChatGPT and Canva
You need two tools for this and both have free tiers: ChatGPT to generate the words, Canva to build the images. Meta’s own systems handle the targeting and optimisation, which is the part small advertisers used to pay third-party platforms for.
The mistake most people make is producing one carefully polished ad. Ad accounts decline through creative fatigue, not imperfection, so your goal is several genuinely different versions — and the fastest route to that is generating ten hooks, picking four, and letting Meta find the winner.
This is that workflow end to end, including the exact Canva dimensions, the policy constraints to build into your prompt, and which numbers actually tell you whether it’s working.
Step 1: Decide the offer before you touch either tool
One offer, one audience, one action
Most failed ad campaigns fail here rather than in the creative. Write down three things before you open anything: what specifically you’re offering, who it’s for, and what you want someone to do.
“We do plumbing” is not an offer. “Boiler service, £95, booked within a week, for homeowners in [town]” is an offer — it’s specific, it has a price, it has a timeframe, and someone can decide about it. Vague offers produce vague ads that produce nothing.
Also settle what a result is worth to you. If a booked job is worth £200 in profit, you know roughly what you can afford to pay for one, and that number is what tells you later whether the campaign worked.
Step 2: Generate ten hooks in ChatGPT
Volume first, then pick four
The first line decides whether anyone reads the rest, so generate plenty and choose. This prompt has the platform policy constraints built in, which matters — a language model asked to write persuasive ad copy will happily produce exactly the claims that get ads rejected, because those claims are persuasive.
OFFER: [what you’re offering, with the real price]
AUDIENCE: [who specifically, and where]
WHAT THEY’RE WORRIED ABOUT: [the real objection or fear]
WHAT MAKES THIS WORTH IT: [honest answer]
Use a different angle for each: a problem statement, a specific number or price, a question, a common misconception, a direct offer, an objection answered, a comparison, genuine urgency (only if true), a local angle, and one plainly factual.
HARD RULES — these get ads rejected:
– No claims about results or outcomes
– No before-and-after implications
– No wording that implies I know something personal about the reader (“struggling with…”, “worried about your…”)
– No superlatives I can’t evidence
– No exclamation marks
– Under 15 words each
Then tell me which three you’d test first and why.
Pick four that are genuinely different from each other — not four variations on the same idea. If three of your favourites all lead with price, you’re testing one thing four times.
Step 3: Build the creative in Canva
Three sizes, and where the text can safely go
Set up your dimensions correctly once and every future ad is a template fill. These are the sizes worth producing.
Canva dimensions for Meta ads
| 1080 × 1350 | Portrait feed. Usually the best-performing feed size because it occupies more screen. Make this your default. |
| 1080 × 1080 | Square feed. Safe, works everywhere, useful as a second variant. |
| 1080 × 1920 | Stories and Reels. Keep all text and logos clear of roughly the top 250px and bottom 250px, where the interface overlays sit. |
What to actually put on the image. Your hook as large legible text, a photo of your real work, your price if it’s a selling point, and nothing else. Heavy text blocks still underperform even though the old hard limit is gone, so aim for a short line someone reads in under a second.
Use your own photos. Stock imagery in a local business ad is recognisable and undermines the credibility everything else is building. If you have no usable photo, a bold text-only design on a solid colour outperforms generic stock.
Build one master design, then duplicate it three times and swap the hook. Canva’s resize function generates the other two dimensions from each, so four hooks across three sizes is about twenty minutes.
Step 4: Write the primary text and headlines
Short, and the first line does the work
For each of these four hooks — [paste your chosen four] — write:
1. Primary text, two versions: one under 40 words, one under 90 words
2. A headline under 40 characters
3. A description under 30 characters
Rules:
– The first line must be the hook, not a restatement of it
– Include the real price in at least two of the four
– One clear action at the end — no “click the link below to learn more about how we can help”
– Same policy rules as before: no results claims, no personal-attribute implications, no exclamation marks
– British English
Then tell me which single ad you’d expect to perform best and what you’d expect to be wrong about that.
Most of your primary text gets truncated on mobile, so front-load everything. If the offer isn’t clear in the first line, it isn’t clear.
Step 5: Set it up without over-complicating it
One campaign, one ad set, four ads
The commonest small-budget mistake is splitting a modest spend across six ad sets to “test properly”. That starves each one of the data Meta needs to optimise, and you end up with six inconclusive results instead of one clear one.
Structure it simply: one campaign, one ad set, your four creatives inside it. Let Meta’s automated placement and audience features do their job rather than hand-picking placements and narrow interests — its own systems now handle that better than manual setup for most small advertisers.
Budget: set it high enough that the campaign can plausibly produce several results a week. If it can’t, you’ll get noisy data and conclude advertising doesn’t work. If your budget is genuinely small, optimise for a cheaper action — an enquiry or a message rather than a completed sale.
Then leave it alone for several days. Editing an ad set resets its learning, and judging performance after twelve hours tells you nothing.
Step 6: Make the landing page continue the promise
Where most wasted spend actually goes
You’ve paid for the click. If the page it lands on doesn’t obviously continue what the ad said, that money is gone — and this is a far more common failure than bad creative.
Here’s the page it points to: [paste the current page copy, or describe it]
Tell me:
1. Does the opening of this page continue the ad’s promise, or change the subject?
2. What would someone arriving from this ad expect to see immediately, and is it there?
3. What’s missing that would stop them acting — price, availability, what happens next, reassurance?
Then rewrite the top section of the page so it follows on directly from the ad. Keep the same offer and price. Don’t restate the ad word for word.
Two practical points: send traffic to a page about the specific offer rather than your home page, and make the action obvious without scrolling on a phone.
Step 7: Read the right numbers
Cost per result, not clicks
After a few days you’ll have data. The number that matters is cost per result — per enquiry, per booking, per sale — measured against what that result is worth to you from Step 1. Everything else is diagnostic.
Click-through rate tells you whether the creative stopped people. A high CTR with no conversions usually means the ad promised something the page didn’t deliver, which points at Step 6 rather than the creative. Cheap clicks that don’t convert are not a success.
Then feed the winner. When one creative clearly outperforms, produce more variations of that — different hooks on the same visual, different opening lines, a video version — rather than starting fresh. This is how you manage creative fatigue instead of being surprised by it.
Tell me:
1. Which creative is genuinely performing best, and whether the difference is meaningful or likely noise given these numbers
2. What the winning ad and the worst ad differ on — what’s the actual variable?
3. Five new hooks that keep whatever’s working and change one thing
4. Whether any result suggests my landing page rather than my creative is the problem
Be honest if the data is too thin to conclude anything.
Before you publish, check
- No results claims, before-and-afters, or personal-attribute wording.
- The price on the ad matches the price on the page.
- Text clear of the safe zones on the 1080×1920 versions.
- Photos are your own work, not stock.
- The link works — click it yourself, on a phone.
- Four genuinely different angles, not four versions of one.
- The landing page action is visible without scrolling on mobile.
- Category rules checked if you’re in health, finance, employment, housing or credit.
Frequently asked questions
How much should I spend to test properly?
Enough that the campaign can produce several results a week, which is a more useful test than a fixed figure. Meta’s systems need a reasonable volume of conversions to optimise, so a budget too small to generate them produces noisy data that will convince you advertising doesn’t work when actually you never gave it enough signal. Work backwards: if a booked job is worth £200 profit and you’d be happy paying £40 for one, a budget that could plausibly produce two or three a week is a real test. If that’s beyond you, optimise for something cheaper — an enquiry, a message, a form fill — rather than a completed sale, and accept you’re testing interest rather than revenue. What you shouldn’t do is spread a small budget across several ad sets, which starves all of them. And leave it running several days before judging; editing resets the learning.
Why did my ad get rejected?
Usually a claim rather than the image, and usually one the AI wrote because it was persuasive. The frequent causes: claiming or implying a result (“lose weight”, “double your income”, “fix your credit”), before-and-after imagery in health and beauty, wording that implies knowledge of the reader’s personal circumstances (“struggling with debt?”, “worried about your teeth?”), superlatives you can’t evidence, and category-specific breaches in health, finance, employment, housing and credit. The prompts in this guide are written to avoid the general ones, but they can’t cover your specific sector. If you’re rejected, read the stated reason properly rather than resubmitting the same thing — resubmitting repeatedly can flag the account. Fix the specific claim, and if you disagree, the appeal process exists and is often successful when the rejection was automated and wrong.
Do I need video, or will images do?
Images work perfectly well, particularly for offer-led and price-led ads, so don’t let the absence of video stop you starting. That said, video generally gets more reach because the platforms push it harder, and it doesn’t need to be produced — a phone clip of you doing the work, with captions, frequently outperforms a polished asset because it looks like content rather than advertising. If you want to add video without filming anything new, a set of photos assembled into a five-second clip with a slow zoom and on-screen text counts as video as far as the platform is concerned, and free editors do that in minutes. The practical approach for a first campaign: start with images because they’re faster, and add one rough video variant in your second round to see whether it changes your cost per result.
Should I use Meta’s AI suggestions or my own creative?
Both, and understand what each is for. Meta’s generative features will produce variations of what you upload — alternative text, cropped and reformatted versions, sometimes background changes — and there’s no reason to refuse that, since it costs nothing and adds variants for testing. What it can’t do is invent your raw material: it can’t film your product, put your customer on camera, or write a hook that reflects something true about your business, because it doesn’t know any of that. So supply genuinely different creative and let Meta remix and distribute it. One caution worth applying: review what its automatic variations actually produce before they run, because an automated crop can put your text off-frame or a generated background can look wrong for your trade. Approve rather than assume.
The bottom line
Define one specific offer with a real price before you open either tool, because vague offers produce ads that go nowhere. Generate ten hooks in ChatGPT with the policy constraints written into the prompt, pick four that are genuinely different, and build them in Canva at 1080×1350 with your own photos and one short legible line. Then keep the setup simple — one campaign, one ad set, four ads — and let Meta optimise rather than hand-managing placements. Judge on cost per result against what a result is worth to you, make sure the landing page continues the ad’s promise, and when something wins, produce more of that rather than starting again.
Platform advertising policies, campaign features, creative specifications and automated tools change frequently — verify current requirements in Meta’s own documentation before planning a campaign. Category restrictions apply in health, finance, employment, housing and credit advertising. Nothing here is legal or regulatory advice.