ChatGPT Ads Just Got a Face: New Visual Format Changes for Niche Products

A few months ago I ran ChatGPT Ads for one of our products. Getting to the point of launching took about a week on its own — OpenAI had to verify our account before we could put anything live. Once the campaign started, the impressions came in fast but the clicks did not.

Eventually the campaign was discontinued over health-related claims. That was the stated reason and I take it at face value, but it arrived on a campaign that had already spent weeks collecting impressions nobody acted on.

Looking back, I think the deeper problem was fit. What we were selling is niche, and ChatGPT Ads may simply not have been the right platform for it at that stage. If we’d had room to show the product properly, I think we’d have gotten better clicks.

A logo and a banner is not a creative strategy

When we ran our campaign, the creative we could supply was a logo and a banner — and that was the whole canvas. If you sell something people already recognise, that is enough. A familiar logo carries trust, category and price expectation all at once, so the ad only has to remind.

A niche product has none of that stored up. The logo means nothing yet, which leaves the banner to explain, convince and differentiate in a strip of space, beneath an answer the person actually came for.

So here is what I think was happening long before the policy flag arrived.

OpenAI announced a new visual format

OpenAI has announced a new visual ad format in ChatGPT, built around images rather than a logo and a line of text.

The framing is the part I keep rereading. The format is meant to help people imagine how a product could fit into their lives, through images showing product inspiration, product usage, or the experiences it makes possible.

Read that as a media buyer and it is a different brief entirely. The old unit asked what your brand mark was. The new one asks what someone’s life looks like with this thing in it.

Four details are worth holding onto before anyone gets carried away, me included:

• It is a test, not a general rollout. It began later in October in the US, with an initial group of advertisers.
• It appears during image generation. For users on the Free and Go plans — a specific moment, not every conversation.
• Ads are clearly labelled. They are kept separate from the image being created.
• Advertising does not influence ChatGPT’s answers. OpenAI repeats this throughout the post.

OpenAI also expanded the plumbing around the format. Conversion data now flows in through Hightouch, Tealium and LiveRamp; attribution runs through partners including AppsFlyer, Triple Whale, Northbeam, DV Rockerbox and Adjust; and Haus, Measured and WorkMagic are exploring geo-based incrementality tests.

On brand safety, placement guardrails keep ads out of emotionally vulnerable or sensitive conversations, and evaluation pilots with DoubleVerify and Integral Ad Science let independent partners assess those standards without reading private conversations. OpenAI puts the platform’s reach at 1.2 billion people each week.

Why a picture changes the maths for a niche product

For a niche product, the first job of an ad is not persuasion. It is comprehension.

Mass-market advertising skips that step. Everyone knows what a sneaker or a meal kit is, so the whole budget goes on preference: why this one, why now, why at this price.

Niche products start a step behind. Before anyone can want it, they have to grasp what it is and who it is for. A logo cannot carry that and a banner barely can, while an image of the thing in use answers both in the same glance.

There is a second effect that matters more for a small budget. Low click-through on a narrow product is often not weak demand but weak qualification: the right person never recognised themselves in the ad, and the wrong people had no reason to click either. When an optimisation system reads that as disinterest and switches the campaign off, the format has made the decision for you. A richer creative unit is not just a prettier ad. It is a fairer test.

The placement helps too. Serving these ads during image generation reaches someone already in a visual, exploratory frame of mind, rather than someone racing to an answer.

What I’d do differently next time

1. Lead with the product in use, not the product. The old creative showed what we are. The new format rewards showing what changes for the person. In our case that means the moment of relief, not the interface.
2. Write the image for one person, not the category. Niche audiences disqualify themselves quickly, which is fine. The creative should make the right person feel addressed and let everyone else scroll past.
3. Decide the success metric before launch, and make it conversions. Our last campaign was effectively judged on click-through, the least informative number available for a niche product. With attribution now running through Triple Whale, Northbeam and AppsFlyer, there is no excuse.
4. Get the claims right before the creative. Ours were flagged as health-related. Reading the ad policies closely is a far cheaper way to find that out than a mid-flight review, and it is worth budgeting time for approval — account verification alone took us about a week before we could launch at all.
5. Keep it as test budget. This is still a channel being built in public. It deserves a line item, not a reallocation.