An ad image is the square visual creative that appears inside a ChatGPT sponsored ad card, sitting alongside the title, copy, and advertiser name that make up a paid placement in ChatGPT's answers. It is the picture a buyer sees in the ad unit, separate from the text fields and the landing page the click sends them to.
Because a ChatGPT ad card is small and intent-matched, your ad image decides in a glance whether a reader trusts the placement enough to act on it. Pick a cluttered or off-message visual and the card gets ignored or rejected at policy review, so the budget behind it earns nothing.
In ChatGPT Ads, the ad image is the creative file an advertiser uploads or pulls from a product feed so the sponsored card has a visual to show next to its text. It is one field in a larger ad unit that also holds the advertiser name, a favicon, a title, ad copy, and a landing URL.
The image has to meet OpenAI's technical rules before it can run. OpenAI asks for a square PNG or JPG hosted at a publicly reachable URL and sized within its published limits, and its guidance tells advertisers to keep the visual simple, legible, and aligned with the ad's message. Because the card renders small, a photo of the product or the person using it reads better than a logo or a busy, text-heavy graphic.
The ad image sits beside the copy fields and the landing page, and each does a different job: the image earns the glance, the copy states the offer, and the page closes it. AirOps helps teams produce and version on-brand ad images at the scale a feed-driven ChatGPT campaign needs.
Resources: See the creative specs and policy checks a ChatGPT ad image must pass
An ad image moves through the same path every ChatGPT ad takes, from your design file to a served card.
Create the visual. Design or generate a clean, square image that shows the product or its use, legible at the small size the card renders.
Meet the specs. Export a PNG or JPG that fits OpenAI's published size limits and host it at a public URL the ad system can open directly.
Attach it. Add the image to an ad through bulk upload, the advertiser API, or a product feed that supplies one image per item.
Pass policy review. OpenAI checks the image, copy, and landing page together, and rejects the ad if the visual is misleading, cluttered, or off-message.
Serve in the card. Once approved, the image renders inside the clearly labeled sponsored card that appears below a relevant ChatGPT answer.
A live ad image tells you the creative cleared policy and is eligible to show. It stays silent on whether buyers find it relevant; only impressions, clicks, and conversions on the card reveal that.
ChatGPT Ads put your brand in front of buyers at the exact moment they ask a question, and the ad image is the first thing they judge. Spend on the placement is wasted if the visual does not earn attention and trust in the half-second before a reader moves on.
It carries the first impression. On a small sponsored card, the image lands before anyone reads the copy, so a weak visual caps the performance of every other field you paid to place.
It gates whether the ad runs at all. OpenAI reviews the image against its ad policies, and a misleading or non-compliant visual gets the whole ad rejected, stalling a launch you planned around.
It separates your brand from the answer around it. The card sits beside an organic ChatGPT answer, so an image that looks generic or stocky blends into the background and surrenders the attention the placement was supposed to buy.
Growth marketers use ad images to test which product visual earns the most clicks on a ChatGPT sponsored card before scaling the winner across a campaign.
Content strategists use ad images to keep the creative in a ChatGPT ad consistent with the landing page and the brand voice a buyer meets after the click.
Demand gen leads use ad images to extend a paid acquisition program into ChatGPT without rebuilding creative for every product in a large feed.
The ad image has to be a square PNG or JPG, hosted at a publicly reachable URL and sized within OpenAI's published limits, so a file that fails those format rules never reaches review no matter how strong the picture itself is.
OpenAI reviews the image together with the copy and the landing page, so a visual that looks misleading, cluttered, or disconnected from the offer can get the entire ad rejected and makes compliance part of the creative brief from the first draft.
In product-feed campaigns the ad image is pulled automatically from each item's image field, so the quality, framing, and consistency of your feed images set the quality of every ad built from that feed, often across hundreds of products at once.
Earns attention inside ChatGPT, where the ad sits beside the answer a buyer is already reading.
Signals trust in the half-second before a reader decides whether the sponsored card is worth a click.
Scales across a product feed so one creative standard covers hundreds of ads at once.
Clears OpenAI's policy review faster when the visual is simple, legible, and aligned with the copy.
Reinforces brand recognition through a consistent look between the card image and the landing page.
Lead with the product or the person using it, because a concrete subject reads faster than a logo at card size.
Export a square image at the top of OpenAI's size range, so the same file clears both the bulk-upload and API rules at once.
Host the image at a public URL that opens directly to the file, because a URL the ad system cannot load stalls the launch.
Keep the frame clean and legible, since a busy or text-heavy graphic loses detail at the small size the card renders.
Match the image to the title and copy, so the card reads as one coherent message a reviewer and a buyer both trust.
Audit feed images before launch, because the ad inherits whatever quality and framing each product image already has.
Avoid baking marketing copy into the image itself. The card already carries a title and description, so text inside the picture competes with those fields, breaks at small sizes, and often trips policy review for cluttered or misleading creative.
AirOps: produces and versions on-brand ad images across a full product feed, so a ChatGPT campaign launches with consistent creative instead of one-off files.
OpenAI Ads Manager: uploads the ad image, attaches it to a campaign, and runs it through policy review before the sponsored card can serve.
Google Analytics 4: tracks the landing-page visits and conversions that follow a click on the ad image, so you can tell which creative actually drives action.
Study the card. Open ChatGPT and find a few live sponsored cards this week to see how small the image renders and what kinds of visuals read well beside an answer.
Set a creative standard. Decide on a square format, a clean subject, and a size at the top of OpenAI's range, then document it so every image you make matches.
Build the first image. Create one square PNG or JPG that shows your product or its use, and host it at a public URL that opens directly to the file.
Run it through review. Attach the image to an ad in OpenAI's Ads Manager, confirm it aligns with your copy and landing page, and submit it for policy review.
Measure and iterate. Once the card serves, watch clicks and conversions by creative, then swap in a new image to test against the one that is live.
An ad image is the square visual in a ChatGPT sponsored card, shown next to the title, copy, and advertiser name.
You create it as a square PNG or JPG, host it at a public URL, and attach it by bulk upload, API, or product feed.
OpenAI's format rules and small card size mean the image must be clean and legible before anything else matters.
A misleading or cluttered visual can fail policy review and block the whole ad from running.
The biggest gains come from a concrete, on-message subject that earns the glance and matches the landing page.
The ad image is the picture in the sponsored card, while the ad copy is the short text, the title and description, that sits beside it. They do separate jobs inside the same unit. The image works first and fast: it is what a reader notices as the card loads, and its job is to earn a second of attention and signal what the ad is about. The copy does the explaining, naming the offer and giving the reader a reason to click. Because the card is small and intent-matched, neither field can carry the placement alone. A strong image with vague copy earns attention and then loses it; sharp copy behind a muddy or off-brand image rarely gets read at all. The practical rule is to design them together, so the subject in the image and the promise in the copy point at the same offer the landing page delivers.
Start with two or three distinct ad image variations per offer, then expand only once you have enough impressions to tell them apart. OpenAI's own guidance asks each variation to introduce a different angle, so meaningful differences matter more than raw volume. Two or three gives you a real test without splitting your budget so thin that no single creative gathers readable data. Make the variations genuinely different: a product shot against a lifestyle shot, a tight crop against a wider frame, one subject against another. Avoid shipping near-identical files that differ only in a filter, because the results will not tell you anything you can act on. As a campaign matures and spend grows, you can run more variations in parallel, especially in feed-driven campaigns where each product already carries its own image. The right number is the most you can test while still giving each one enough exposure to judge.
The same ad image performs differently because the answer it appears beside changes with every query. A ChatGPT ad card is intent-matched, so one impression might sit under a question where your product is the obvious fit and the next under a loosely related answer where it feels out of place. The surrounding context shapes how relevant the image looks, even though the file never changed. Audience, device, and the market where the ad runs add more variation, since a visual that reads clearly on one screen can lose detail on another. Query intent and competition shift from moment to moment, so the same creative can earn a strong click rate one week and a weak one the next. This is why marketers judge an ad image across many impressions over time, and treat any single placement as one noisy data point instead of a verdict.
You control the image itself completely, and you influence where it shows without controlling it outright. What you own is the file: its subject, framing, format, and the public URL it lives at, plus whether it aligns with the copy and passes policy review. What you do not own is placement. ChatGPT decides which query triggers your card and which answer it sits beside, based on your targeting and its own relevance matching. You cannot pin your ad image to a specific answer or guarantee it shows for a given prompt. The useful move is to control everything upstream of the auction: ship a clean, compliant, on-message image, set your targeting carefully, and host the file where the ad system can always reach it. Do that, and you give the image the best possible chance of serving well whenever ChatGPT chooses to surface your card.
A good ad image is clean, legible at small size, and instantly readable as what you sell. Because the sponsored card is tiny and sits beside an answer, the test is whether a reader understands the image in a glance without reading the copy. Strong cards usually lead with the product or the person using it, use a simple background, and avoid baked-in text that breaks at small sizes. Technically, a good image meets OpenAI's format rules the first time, so it clears review without a resubmission. There is no single benchmark number that defines good, since performance depends on your offer and audience, so judge quality by outcomes: a healthy click-through rate against your other channels and a conversion rate on the landing page that justifies the spend. If the image earns attention and the click converts, it is doing its job.