Can ChatGPT make Facebook ads? Where it helps and where it fails

By Spencer Tribe · Last updated September 22, 2026

Yes. ChatGPT can write Facebook ad copy and make an ad image from a prompt or from photos you upload. For a business it falls short in three places: it redraws your logo rather than placing your file, so details drift; it does not know which ad layouts work in your category unless you tell it; and every ad starts from a blank prompt. Use it for copy and ideas. For the image, a tool that places your real logo and starts from a proven ad, such as Showfolk, gets to a usable ad faster.

What is ChatGPT good at for Facebook ads?

Copy. Ten versions of the primary text, headlines under a word limit, and the same message in a different tone.
Ideas. Angles to test, offers to try, and objections your ad should answer.
A first picture. A concept image to show someone what you have in mind.

An ad made with Showfolk: an app on a phone with the headline Your GLP-1 Progress, Simplified — The four parts of a Facebook feed ad
**Primary text.** Above the image. Only the first line or two show before “See more”, so the point goes first.
**Image.** The ad itself. It does most of the work, and should carry only a few words.
**Headline.** Under the image. A few words that repeat the offer.
**Button.** The action you chose: Shop now, Book now, Send message.

Where does it go wrong?

Problem	Why it happens	What it costs you
Your logo comes back slightly different	Image models draw a logo as a picture of a logo; they do not place your file.	An ad you cannot run, or one that looks almost like you.
Words need checking	Recent models handle short text far better, but letters are still drawn, not typed.	A read-through of every ad, and retries.
It does not know what works in your category	It has not seen which ads brands in your industry are paying to run.	Generic layouts, unless you describe one yourself.
Every ad starts from scratch	Your brand is whatever the prompt says this time.	Colors and style drift from one ad to the next.

Why image models garble text and logos.

What prompt works better?

If you use ChatGPT for the image, give it the things it cannot guess:

Upload a photo of your product, and describe the ad you want in layout terms: “the product centered on a solid color background, a short headline at the top.” The eight layouts give you the words.
Give it your exact colors as hex codes and your exact headline in quotation marks, and ask it to leave space for the logo rather than drawing one.
Ask for the size you need, 4:5 for the feed or 9:16 for Stories, and check the file before you upload it.
Add your real logo yourself afterwards, in any editor.

How does ChatGPT compare with Showfolk?

	ChatGPT	Showfolk
Where the ad’s design comes from	Your prompt	A real ad a major brand is running, which you pick or we pick for you
Your logo	Redrawn by the model	Your file, placed as a real layer
Your headline	Drawn by the model	Typed and placed as a real layer, then read back for spelling
Your colors and fonts	Whatever the prompt says each time	Taken from your website once, used in every ad
Sizes	Ask, then check	1:1, 4:5 and 9:16 exports
Changing one spot	Describe the change; other parts can shift too	Click the spot; everything else stays
Copy for the primary text	Very good	Headline only; write the post text anywhere

Showfolk uses an image model for the picture too. The difference is that it starts from a proven ad and puts your logo and words on afterwards, so they are exactly right.

The reference: an ad a major brand is running.

What Showfolk made from it for an app.

Same structure, a product alone on a soft gradient with a short headline, rebuilt for a different business. Nothing is copied: the reference’s product and brand are kept out.

Can I use Showfolk inside ChatGPT?

Yes. Showfolk works as a connector in ChatGPT, Claude and other assistants that support MCP, so you can ask for an ad in the chat and get the finished file back. Setup steps.