How to Advertise on ChatGPT and AI Chats

Advertising inside ChatGPT is bought through a self-serve system with no keyword auction. You pay per impression or per click, the ad is labeled as advertising, and it sits separately from the answer the model writes. You define a budget, supply the creative, and the platform meters delivery.

In short:

• ChatGPT ads bill per impression or per click; there is no bidding on keywords.
• The ad is marked as advertising and placed apart from the generated answer.
• Targeting comes from the campaign settings the platform offers, not from a phrase you buy.
• Budget depends on your goal and on how the platform meters delivery, not on a published floor.
• Tracking works like any other paid channel: landing page, analytics, conversion events, cost per acquisition.

How do you advertise on ChatGPT?

You set up a campaign in OpenAI's self-serve advertising system, choose how you want to pay, define a budget, and upload the creative the system serves. Delivery is metered per impression or per click, and the placement is labeled as advertising and kept separate from the answer the model produces. There is no keyword bidding layer between you and the placement, so the setup work is mostly about creative, budget, and the targeting options the campaign settings expose.

Practically, the workflow looks like this:

1. Create the campaign and pick the billing model the platform offers, per impression or per click.
2. Set the budget and decide how the platform should pace delivery against it.
3. Supply the creative: the copy and any assets the system needs to serve the ad.
4. Define targeting using the settings available in the campaign, since there is no keyword list to build.
5. Send traffic to a page that has analytics and conversion tracking installed before you spend anything.
6. Read results by cost per acquisition and compare them against your other paid channels.

Step five is the one people skip. If the landing page has no conversion event, the ad becomes unmeasurable, and you end up judging it on impressions alone. That is not a verdict on the channel, it is a missing measurement.

Why keyword bidding does not exist in ChatGPT advertising

Keyword bidding does not exist in ChatGPT advertising, so you cannot buy a phrase the way you would in Google Search. Targeting works through the campaign settings the platform provides, and the ad is placed next to or apart from the AI answer rather than inside it. That single difference reshapes how you plan the campaign.

In search, you start from intent expressed as a query and you bid on the words that carry it. Here you start from who you want to reach and what the platform lets you select, then you write creative that earns attention in a context where the reader is already reading something else. The ad competes with the answer for attention, not with nine other blue links.

This is why the usual keyword research spreadsheet is not the first artifact you build. The first artifact is a clear statement of who the ad should reach and what a click is worth to you. Everything downstream, budget, creative, and the decision to keep running, follows from that number.

If you are used to auction mechanics elsewhere, the contrast is worth internalizing before you spend. Auction-based delivery is explained in detail in our breakdown of how TikTok ads work: auction and delivery explained, and the same logic applies to LinkedIn's model in how do LinkedIn ads work: mechanics and formats. ChatGPT advertising deliberately drops the auction layer.

ChatGPT ads versus Google Ads

The two systems differ on three things that matter: how the placement is triggered, how you pay, and how long you have to count a conversion. Google Ads runs an auction on keywords and charges by click, with a conversion window of 30 days by default that you can set anywhere from 1 to 90 days. ChatGPT ads charge per impression or per click but have no keyword auction at all. Google serves against a query; ChatGPT shows the ad alongside a generated answer.

Google Ads has no fixed minimum budget, so the absence of a published floor on the ChatGPT side is less unusual than it first looks. What you actually need is a budget large enough to produce a readable result, and that number comes from your own economics: what a click costs you, what share of clicks convert, and what a customer is worth. If a click costs X and one in Y clicks converts, your cost per acquisition is X multiplied by Y, and the budget has to cover enough clicks to tell signal from noise. Those are your numbers, not the platform's.

What budget do you need to advertise in AI chats?

There is no single correct budget for AI chatbot advertising. ChatGPT advertising lets you choose between paying per impression and paying per click, and the amount you need depends on your goal and on how the platform meters delivery. Other channels publish their own floors, which gives you a rough sense of how differently platforms think about entry size: LinkedIn requires a minimum daily budget of 10 USD, and TikTok requires more than 50 USD per campaign and 20 USD per ad group per day. Google Ads has no fixed minimum budget.

Two decisions drive the number:

• Billing model. Per-impression buying puts the risk on your creative and your targeting, because you pay whether or not anyone acts. Per-click buying puts the risk on the click price. Neither is cheaper in the abstract; they fail in different ways.
• Result threshold. You need enough delivery to see whether the campaign produces conversions at a cost you can live with. Below that threshold you are reading noise.

A budget that cannot produce a readable result is not a test, it is a donation. Decide in advance what result would make you scale and what result would make you stop, and write both down before the campaign starts.

How to track results from ads shown in AI chats

You track ads in AI chats the same way you track any other paid channel: send the traffic to a page with your analytics and conversion tracking in place, then compare cost per acquisition across channels. In GA4 you can choose an attribution model of last click or data-driven, and event data is retained for 2 or 14 months depending on your setting.

The mechanics are unglamorous and they decide everything. Tag the destination URL so the traffic is attributable. Fire a conversion event on the action that matters, not on the page view. Pick an attribution model and stay consistent, because switching models mid-flight rewrites history and makes the channel look better or worse than it is.

Which numbers move depends on your offer, your landing page, and how well the placement matches your audience. A campaign with strong creative and a mismatched audience will lose to a plain ad in front of the right people. That is true on every platform, and it is the reason a first test should change one variable at a time.

If you are running several channels at once, the comparison is only fair when the tracking is identical across them. The same conversion event, the same attribution model, the same reporting window. Otherwise you are comparing four different measurement systems and calling it performance. The working habits that keep multi-channel reporting honest are covered in campaign manager: tasks, skills, when to hire one.

Setting up the campaign without wasting the first spend

Most first campaigns fail for reasons that have nothing to do with the platform. The landing page has no conversion event. The creative describes the product instead of the reason to click. The audience is defined so broadly that the ad reaches people who will never buy. None of these are ChatGPT-specific problems, and all of them are cheaper to fix before launch than after.

A setup that has been prepared in advance, with the campaign built but not yet live, lets you review creative, targeting, and budget as a set before money moves. That is the approach behind AI-assisted campaign setup, where the campaign is generated and then held on pause until you confirm the launch and the budget yourself.

Before you switch anything on, check these:

• The conversion event fires correctly on the destination page.
• The creative says what the reader gets, in the reader's words.
• The targeting matches the audience the offer was written for.
• The budget can produce a readable result within the time you are willing to wait.
• You have written down what result would make you scale and what would make you stop.

What to expect in the first weeks

Expect the first weeks to be a measurement exercise, not a performance one. You are buying enough data to decide whether the placement reaches the right people at a cost that works. That decision needs a stable setup, consistent tracking, and a willingness to change one thing at a time.

Two failure modes show up repeatedly. The first is judging the channel after a spend too small to be meaningful, then concluding it does not work. The second is scaling a campaign that produced a handful of conversions without checking whether those conversions were cheaper or more expensive than the same conversions from your other channels.

Neither error is unique to AI chatbot advertising. Both are more expensive here because the placement is new to most buyers, and there is no long history of benchmarks to lean on. Your own cost per acquisition is the only benchmark that matters, and you build it by running the campaign with tracking that works.

Next step

Pick one offer, install conversion tracking on its landing page, and set a budget you are willing to spend to learn whether the placement reaches the right people. Then build the campaign and hold it before launch so you can review the creative, the targeting, and the budget as one decision. Start with the campaign setup workflow at AI-assisted campaign setup.