ChatGPT Ad Images: What to Show in a Chat Card
The image in a ChatGPT chat card is small, sits under an answer, and has to explain your offer at a glance. Here is what works and what gets rejected.
On Meta, the image is the ad. It has to stop a thumb mid-scroll. In a ChatGPT chat card, the job is different. The person has just read an answer about the exact problem your product solves, and the card sits below it, labeled as sponsored. Your image does not need to interrupt anyone. It needs to confirm, in a fraction of a second, that this is the thing they were just reading about.
That shift changes what a good image looks like. This guide covers the role of the image in the chat card, concepts that work for different business types, the policy lines you cannot cross, and a checklist to run before you upload.
Where the image sits in a chat card
A ChatGPT ad unit, the chat card, includes:
- Your advertiser name and favicon
- A headline of 3–50 characters
- A body of up to about 100 characters
- An image
- A landing URL
It appears below ChatGPT's answer, visually separated from it, and it does not change what ChatGPT says. The image is one element among several, shown at a modest size. That has three consequences:
- Detail disappears. Fine text, busy backgrounds and small product features will not read.
- The headline does the arguing. The image supports the headline rather than carrying the message alone.
- Consistency matters. If the image shows something different from the headline, or from the landing page, the card feels off and people skip it.
What a good chat card image does
Think of the image as visual proof of the headline. If the headline says "Weeknight dinners kids actually eat," the image should show a real plate of food a child would plausibly eat, not a stylized brand pattern.
A good image does one of these jobs:
- Shows the product clearly. One item, centered, on a simple background. Best for physical products where recognition matters.
- Shows the product in use. Hands applying the serum, a dog wearing the harness, the app screen on a phone in someone's hand. Best when use is the selling point.
- Shows the outcome honestly. A tidy shelf after using the organizer, a packed lunchbox. Not a before/after comparison (more on that below).
- Shows the experience. The cozy room with the throw blanket, the class in progress. Best for services, courses and lifestyle products.
Image concepts by business type
| Business | Strong concept | Weak concept |
|---|---|---|
| Skincare | Product bottle beside its texture swatch, plain background | Model with flawless skin implying a result |
| Pet products | Dog using the product in a real home | Logo on a colored background |
| Consumer app | Phone showing one key screen, clean surroundings | Collage of six tiny screens |
| Home goods | Item in a styled, uncluttered room | Spec diagram with measurements |
| Online course | Instructor teaching, or a lesson screen | Stock photo of a diploma |
| Local service | Your team at work, real job site | Generic stock handshake |
| Apparel | Garment worn, full item visible | Flat lay with five products |
The pattern: one subject, recognizable at small size, directly tied to what the headline promises.
Policy rules that apply to images
Every image is reviewed against OpenAI's ad policies. Images trip policy more often than people expect, because a picture can imply a claim the copy carefully avoids. Avoid these:
- Before/after transformations. Not allowed. This includes split-frame comparisons of skin, bodies, hair, rooms presented as results, or anything that implies a guaranteed transformation.
- Implied guaranteed outcomes. An image of a pile of cash next to a course, or a dramatically slimmer body next to a supplement, implies a claim you cannot make.
- Impersonation. No images styled to look like ChatGPT's interface, OpenAI branding, public figures, or another company's products and brand assets.
- Other brands' trademarks. Competitor logos, packaging or recognizable products in the frame can get the ad rejected and are a bad idea legally.
- Shocking or graphic content. Gross-out imagery, injuries, or fear-based visuals.
- Text that shouts. All-caps overlays, fake countdown timers and "only 2 left" banners count as clickbait or fake urgency.
- Exploiting insecurity. Images designed to make people feel bad about their appearance or circumstances.
Restricted categories, including health, supplements, wellness, weight management, financial products, alcohol, dating and products for minors, face extra scrutiny. Keep imagery neutral and product-focused in those categories. For the full picture, see our guide to ChatGPT ad policies, and run your copy through the ChatGPT ad policy checker before submitting.
Text in images: keep it minimal
The headline and body already carry your words. Adding text to the image usually hurts more than it helps, because it shrinks to illegibility and can repeat or contradict the copy.
If you must include text, limit it to:
- Your product name, if it is not obvious from the packaging
- A single short label that is true and verifiable, like "Refillable"
Never put prices, discounts or claims in the image that are not also on the landing page. Message match applies to images as much as to copy.
Match the image to the landing page
When someone clicks, the first screen they see should look like a continuation of the card. If the image shows a blue water bottle and the landing page opens on a lineup of eight colors with the blue one below the fold, you have broken the thread.
Simple rules:
- The product in the image should appear above the fold on the landing page.
- The color, variant or bundle shown should be the one the link leads to.
- If the image shows a specific use (a lunchbox, a hiking trail), the landing page should acknowledge that use.
The message match checker can compare your ad and landing page copy, and our guide to landing pages for ChatGPT ad traffic goes deeper.
Sourcing images: what to use
You do not need a studio shoot for every ad. In rough order of preference:
- Your own product photography. Real, owned, and exactly what the customer receives.
- Customer content you have permission to use. Authentic and often more relatable, but get written permission and do not imply results.
- Lifestyle shots from your brand library. Fine if the product is clearly visible.
- Carefully made AI-generated or edited images. Acceptable for concepts and backgrounds, but the product shown must be accurate. An AI image that makes your product look better than it is creates a misleading claim.
- Stock photos. Use sparingly and only when the product is not the focus, such as for a service. Generic stock lowers trust.
Never use images from competitors' ads, even as a "reference." SecondWin reads long-running ads in the Meta Ad Library to understand which messages work in a niche, but every image concept we write is original.
A worked example
Say you sell a $48 insulated kids' lunchbox. Your context hint targets "parent packing school lunches who wants food to stay cold until noon."
- Headline (33 characters): Lunches that stay cold until noon
- Body (79 characters): Insulated, leakproof and dishwasher safe. Sized for real school lunch portions.
- Image concept: An open lunchbox on a kitchen counter, packed with a sandwich, fruit and an ice pack, morning light. Product color matches the default variant on the landing page.
What we avoided: a crying child with a soggy sandwich (exploits distress), a split before/after (prohibited format), and text saying "BEST LUNCHBOX EVER" (unproven superlative, all caps).
Pre-upload checklist
- One clear subject, recognizable at small size
- Directly supports the headline's promise
- No before/after, no implied guaranteed result
- No competitor products, logos or brand styling
- No OpenAI or ChatGPT branding or interface styling
- Little or no text; nothing that is not on the landing page
- Product shown matches the landing page variant
- You own the image or have permission to use it
How SecondWin handles images
Every SecondWin ad includes an image concept written alongside the headline, description and context hints, so the visual and the words make the same promise. Concepts are checked against OpenAI's policies and our own rules (no unproven claims, no competitor names, no made-up facts) before anything goes live in your account.
To see the image concepts and copy we would draft for your product, get your free ChatGPT ads report. It takes your URL and needs no account.
FAQ
What size should a ChatGPT ad image be?
Use the dimensions and file requirements shown in ChatGPT Ads Manager when you upload, since OpenAI may adjust specs as the format evolves. More important than exact pixels is design for small display: one clear subject, a simple background, and little or no text, so the image still reads when shown at chat-card size below an answer.
Can I use before and after images in ChatGPT ads?
No. OpenAI's ad policies do not allow before/after transformation imagery. That includes split comparisons of skin, bodies, hair or similar results presented as outcomes of your product. Show the product itself, the product in use, or an honest depiction of the experience instead, and keep claims in your copy provable.
Should I put text on my ChatGPT ad image?
Usually not. The headline and body already carry your message, and image text becomes hard to read at chat-card size. If you include text, keep it to a product name or one short, verifiable label. Never add prices, discounts or claims that do not also appear on your landing page.
Can I use AI-generated images in ChatGPT ads?
You can use AI-generated or edited images, but the product shown must be accurate. An image that makes your product look different or better than what customers receive creates a misleading claim, which breaks OpenAI's rules against deceptive ads. AI imagery works best for settings and backgrounds around a real product photo.