ChatGPT Visual Ads: What Small Businesses Should Make

OpenAI is about to put ads next to the images people make in ChatGPT. Not next to text answers. Next to the moment someone types "show me a living room with a green velvet sofa" and waits for a picture. OpenAI announced the new ChatGPT visu

OpenAI is about to put ads next to the images people make in ChatGPT. Not next to text answers. Next to the moment someone types "show me a living room with a green velvet sofa" and waits for a picture.

OpenAI announced the new ChatGPT visual ads format on October 5, 2026, and it matters for small business for one practical reason: it turns image creative into the entry ticket. If you can't produce a good product photo on demand, this channel isn't open to you in any useful way.

Here's what is confirmed, what is still a test, and what to have ready before you apply.

OpenAI confirmed the format, but only as a US test

The facts first. In OpenAI's own announcement, the company says it is "introducing a new visual ad format in ChatGPT." The ads use images to show product inspiration, product usage, or the experiences a product makes possible.

The first placement is during image generation in ChatGPT, for users on the Free and Go plans. Ads are clearly labeled and stay separate from the image being created. OpenAI also repeats its standing line that advertising does not influence the answers ChatGPT gives.

Then the caveat that decides whether this is for you: testing begins later this month in the US with an initial group of advertisers. That is a closed pilot, not a launch. Businesses can sign up for ChatGPT Ads at ads.openai.com today, but signing up is not the same as getting visual inventory.

Read those numbers the way you'd read any vendor's case studies. OpenAI chose the examples, named brands supplied the data, and partners like DV Rockerbox and Triple Whale did the measuring. The 52.7% figure comes from an early analysis of global campaigns optimized for deeper-funnel outcomes. None of it tells you what a four-person shop with a $20 daily budget will see.

The audience is people mid-creation, which is not the same as people ready to buy

Think about who sees this ad. Someone on a free account who is busy generating a picture. They are in a making mood, not necessarily a buying mood.

OpenAI's pitch is that people use image generation to imagine something, and a product shown in that moment can help them decide. For a sofa, a wedding cake, a hiking boot, a tattoo flash sheet, that is plausible. Visual products with a clear "I could have that" reaction suit it.

For an accountant, a plumber, or anyone selling a service with no visible outcome, it is a hard fit. We'd skip the test and spend the same money where intent is already high. If you're weighing that call, our breakdown of ChatGPT Ads for small business covers who sees ads at all and why Plus, Pro, and Business seats are excluded from the audience.

An ad that sits beside an AI-generated image has one job: don't look like the thing next to it. If your creative reads as generated, it vanishes into the feed it's parked in.

— our take on the placement, not OpenAI's

That is opinion, not data, but it is the opinion we'd act on. Real product in a real hand, real light, real scale. The generated image is the competition for attention, so the ad needs to look like it came from a different universe.

The businesses that fit are the ones whose product looks good before anyone explains it

Here is how we'd sort a list of owners, based on what the format shows rather than what the press coverage hopes for.

A strong fit: a furniture or home-goods store, a bakery or specialty food brand, a boutique selling clothing, jewelry, or gifts, a landscaper or interior designer whose finished work photographs well. The picture does the selling, and a person who just asked an AI for a room or an outfit is already thinking about the thing.

A maybe: a salon or a studio that sells a look, a travel operator selling a place. The image sells, but the booking step is longer, so tracking matters more.

A skip for now: anything where the buyer needs a quote, a consultation, or a phone call before the image means anything. A law firm, an HVAC contractor, a bookkeeper. Their money works harder in channels where someone has already said what they need.

Nothing in OpenAI's announcement says these categories are restricted or favored. This is our read of fit, built on how image ads behave on every other platform we run.

What a first brief looks like when you only get one shot at an invite

If an invitation lands, you will want assets inside a day or two, not three weeks. Here is the brief we would hand a designer, or an AI image workflow with a human checking every frame.

One product, three framings. The product alone, so the shape reads at a glance. The product in use, with a person's hand or a lived-in setting. The outcome, such as the finished room, the plated meal, the gift opened. Same product, same palette, three different jobs.

Keep any text on the image short, and do not hide the product behind a headline. The ad is labeled by ChatGPT already, so the image should not shout. Write the destination page to match the picture: if the ad shows a green velvet sofa, the click lands on that sofa, not a collection grid.

Then decide who is accountable for the result. A channel this young changes its rules every few weeks, as our own coverage of ChatGPT Ads this year shows. Someone has to watch the account weekly. If that someone is you, budget the hours as part of the cost.

Your creative is the part you control, and it has to be ready before the invite

OpenAI says the format helps businesses "tell their story through images." Translate that into a short checklist for an owner.

You need a handful of finished static images per product or offer, not one hero shot. Product alone on a clean background, product in use, and a context shot that shows the result. Three concepts is a sensible start for a test. Fifty variants is what we'd run once something works, the same pace we describe in our post on shipping 50 ad variants a week.

You also need to know your own disclosure position. If any of those images are AI-made, the rules on labeling differ by platform and by country, and whether you have to label AI-generated ads is a question to settle before the creative goes live, not after a complaint.

Measurement got serious, but a small shop needs two pieces of it, not twelve

The second half of the announcement is a long list of measurement partners. Hightouch, Tealium, and LiveRamp help send conversion data into ChatGPT Ads. Attribution partners include AppsFlyer, Triple Whale, Adjust, DV Rockerbox, Northbeam, Branch, Singular, Kochava, Airbridge, and Tenjin. Haus, Measured, and WorkMagic are exploring geo-based experiments to test whether the ads cause extra sales rather than just taking credit for them. OpenAI itself describes that incrementality work as early.

Almost none of that is for a small account. What you need is much shorter.

First, working conversion tracking. OpenAI's pixel and Conversions API are the signal source, and its measurement post says an Event Quality Score helps you judge how good those signals are. If a purchase or a booked call doesn't show up in your own analytics, nothing downstream is trustworthy. If you sell on Shopify, our post on the ChatGPT Ads Shopify app covers the plumbing that already exists.

Second, an attribution window that fits how you sell. OpenAI lets advertisers pick a click-through window and add a one-day view-through window. A same-day purchase and a three-week consideration cycle need different settings, and the choice changes what your report says, though not how the campaign optimizes.

One detail is worth a second look. Because people often come to ChatGPT to work through a task, OpenAI says a click may not be the next step. If you only count clicks, you will undercount. If you count view-throughs without discipline, you will overcount. Decide the rule before you see the dashboard.

Brand safety is a pilot, and Negative Phrases are for qualifying advertisers only

OpenAI says its placement guardrails try to keep ads out of emotionally vulnerable or sensitive conversations, under its advertising policies. For narrow brand-specific needs, Negative Phrases are available to qualifying advertisers. DoubleVerify and Integral Ad Science are running evaluation pilots to check how those safeguards work, without access to private user conversations.

For a small business the useful reading is modest. You get platform-level guardrails, not a custom blocklist, unless you qualify for the extra control. If your brand would be hurt by sitting beside certain topics, ask what controls your account actually has before you spend.

A sensible first test costs less than the creative it takes to look good

If you get access, run it like a lab, not a launch. Pick one product with a strong visual. Prepare three static concepts. Set a daily cap you would lose without comment. Confirm a test conversion appears in your own analytics before the spend ramps. Compare against the channel you already trust, not against zero.

Run it for long enough to see more than one buying cycle, and judge on new-customer share and cost per acquisition, not on impressions. Portland Leather's 93% new visitors is a nice number to chase, but only if the sales are real.

If you'd rather not build all of that yourself, that is exactly what we do. We make the images, wire the tracking, and run the test so you see results instead of a to-do list. Our content and ad creative service is built for new channels at this stage.