A vendor emailed you about ChatGPT ads. Somebody on your team forwarded an article. The question you actually have is not what they are. It is whether this is a real channel you are late to, or a thing you can safely ignore for another two quarters.
Here is the honest version, including the parts nobody can answer yet.
What they actually are
OpenAI moved advertising in ChatGPT out of limited beta in early 2026. When someone asks a question that matches an advertiser’s targeting, a brand can appear as part of the answer, labeled as sponsored.
That last detail is the whole story, and it is why this is not simply another place to run the ads you already run.
A search ad sits next to the answer. Someone asks a question, they get a page of results, and your ad occupies a slot on that page alongside everything else. The person knows they are looking at a list of options and they choose from it.
A ChatGPT ad sits inside the answer. There is no list. The assistant produces a response, and your brand appears within it, marked as sponsored.
The person is not comparing a page of options. They asked for a recommendation and they got one.
Why that changes the math
Three things follow from placement inside the answer rather than beside it, and they cut in different directions.
The moment is later in the decision than most paid ads reach. A search ad usually catches someone who is still assembling their options. A placement inside an answer reaches someone who asked a question expecting a resolution. That is closer to the buying decision, which historically means a more valuable impression.
There are far fewer slots. A results page holds many ads and many organic listings. An answer names a handful of things. Scarcity pushes the price up, and it means the gap between being included and being left out is total rather than gradual. There is no page two of an answer.
Being present is not the same as being chosen. If your brand appears in the answer with an old price, a service you stopped offering, or the wrong service area, the paid placement bought you a confident wrong statement about your business. Buyers give assistant answers more credit than they give ads, which cuts both ways.
What nobody can tell you yet, including us
This is the part most articles about ChatGPT ads skip, and skipping it is how owners end up spending on a channel nobody can grade.
- The bidding model is still stabilizing. Anyone quoting you a reliable benchmark cost per click or cost per acquisition for your category is guessing. Early pricing on a new channel moves as inventory and demand find each other, and numbers from three months ago may not describe what you would pay this month.
- Performance data is thin and mostly anecdotal. There is not yet a body of results across enough categories and enough spend to say with confidence which businesses this works for. Anyone claiming otherwise is extrapolating from a handful of accounts.
- Measurement is workable but new. OpenAI provides conversion tracking infrastructure. Connecting it to your existing reporting so that performance sits next to your other channels, attributed to revenue or booked jobs rather than to platform metrics, is a setup task somebody has to actually do. If it does not get done, you will be running a channel you cannot compare to anything.
None of that means do not run them. It means run them as a test with a number attached, not as a line item you renew out of habit.
How to decide, in four questions
These are the same questions worth asking about any new channel, which is the point. The channel is new. The questions are not.
- 1. Are your buyers actually asking about your category there?Not whether people use ChatGPT, which they obviously do. Whether people making your kind of purchase are asking about it in that setting. A high-consideration purchase where buyers research for weeks is a very different case from an emergency service call. You can check this in ten minutes by asking the questions your customers ask and reading what comes back.
- 2. Can you reach them efficiently at the price it currently costs?This needs a number from your side before you look at a rate card: what a customer is worth to you, and what you can afford to pay to get one. Most businesses that get burned on a new channel never set that number, so they have no way to tell an expensive test from a failed one.
- 3. Can you measure whether it worked?If the tracking is not connected to the outcome that runs your business, you will get platform metrics and a feeling. That is not enough to decide on a renewal.
- 4. What happens to the person after they click?A placement inside a recommendation delivers someone who has effectively been referred. If they land on a page that does not immediately confirm what the answer said about you, you paid a premium for the introduction and wasted it.
What to have in place before you spend anything
Two things, and the second one is the one that gets skipped.
A number you would be happy to pay for a customer. Set it before you see any pricing, so the pricing does not set it for you.
Accurate information about your business in the places these systems read. This matters more here than on any other paid channel. If your hours, pricing, service area, or product line disagree with themselves across the web, an assistant summarizing you will pick one version and state it confidently, and you may be paying to amplify the wrong one. Fixing that is unglamorous and it improves how you appear in the answers you are not paying for either.
That second point is the reason we do not treat this as a standalone thing to buy. Paid placement gets you into the answer this month. Accurate, consistent, corroborated information is what determines whether the answer about you is right, and it keeps working when the budget stops. We cover the mechanics of the unpaid side under answer engine optimization and generative engine optimization, and the paid side under ChatGPT Ads.
What this is not
- It is not a replacement for what you are running now. The businesses treating it that way are making a bet on a channel with no track record using money that is currently producing results somewhere measurable.
- It is also not a reason to reorganize around AI. This is a new place to buy attention, evaluated the same way you would evaluate any new place to buy attention. The novelty is real and it does not change the questions.
- And it is not urgent in the way the emails in your inbox suggest. Being early on a channel matters when early inventory is cheap and your competitors are absent. Whether that is true in your category is a question with an answer, and it is worth getting the answer before you spend, rather than after.
More on how we run paid ads and attribution generally.
Want us to tell you whether it is worth running?
We will look at your category, whether your buyers are actually asking about it in AI assistants, what it would cost to reach them, and whether you can measure it. If it is not the right move for your business yet, we will tell you to wait. We run the analysis before any conversation about working together.
👉 See What’s Actually Holding Your Marketing BackWant more like this?
We break down what is actually working in search, Amazon, and paid ads every week, with no filler. Sign up for the newsletter.
👉 Sign Up for the Newsletter