Conversion tracking for ChatGPT ads

You measure ChatGPT ads with the OpenAI pixel on your site, and with the Conversions API for conversions that happen later or outside the browser. UTM parameters are supported, so you also find the same traffic in your own analytics. What you do not measure this way is the largest part of ChatGPT’s influence: the conversations where you are named without anyone clicking.

Measuring is not optional: without conversion data flowing back, you cannot let the auction optimize toward a conversion goal, and you are stuck with manual bidding.

Sources: OpenAI’s help center on ads in ChatGPT and reporting on the self-service Ads Manager. Checked September 5, 2026.

What the OpenAI pixel does

The pixel is a script you place on your site. It connects behavior after the click to the campaign that caused the click: which pages someone viewed, which actions they took, whether they converted.

That is the same role as the pixels you know from other ad platforms. Two things matter more here than usual:

• Place it before you turn the campaign on. A week of advertising without measurement is a week without learning data, and the model cannot optimize without a conversion signal.
• Make sure it actually fires on your landing page. On pages that only build their content after a bundle loads, this goes wrong more often than you would think.

For the exact implementation, follow OpenAI’s documentation. Here we describe what the instrument does and what you use it for, not which field names to fill in.

When you need the Conversions API

The Conversions API is a server-side connection. You send the conversion from your own server to OpenAI instead of from the visitor’s browser. You need it in three situations:

• The conversion happens later. A lead you only mark as qualified after a phone call, or an order you only confirm after a credit check.
• The conversion happens outside the browser. An order completed by phone, or a subscription activated in your back office.
• The browser measures unreliably. Blockers, dropouts and limited cookie lifetimes mean part of the conversions never arrive through the pixel alone.

In practice you usually use both: the pixel for the direct signal, the API for the conversions that really matter.

UTMs and your existing analytics

UTM parameters are supported. Put them on your ad URLs and you see the traffic in your own analytics, next to what the Ads Manager reports. That is useful for two reasons: you can check OpenAI’s numbers against your own source, and you keep paid AI traffic separate from the unpaid AI traffic that comes in by itself.

Note: only use UTMs on external links. On your own internal links they overwrite the original acquisition source and break your acquisition reporting.

Why last click falls short here

This is the core, and it is not a theoretical objection.

A large part of ChatGPT’s influence does not run through a click. Someone asks a question, gets an answer that names your business, does not click, and searches your brand name three days later. Last click gives that conversion to the branded search. ChatGPT does not show up in your reporting, neither paid nor organic.

With paid campaigns you catch part of it with view-through conversions. Set your conversion window to at least seven days, longer for a longer sales cycle. But the part that comes in through a branded search or directly stays out of view of every ad platform.

See tracking ChatGPT traffic in Google Analytics. The broader case is in the ROI of GEO.

A measurement setup that holds up

Four layers, from hard to soft:

1. Direct conversions: pixel and Conversions API. This is your hardest signal and it steers your bidding strategy.
2. Traffic with a recognizable source: UTMs plus your own analytics. Use this to check the platform’s numbers.
3. Traffic without a recognizable source: the part that comes in as direct. Follow the trend, not the absolute number.
4. Mentions without a click: measure whether you are named for the questions that matter. You do that by asking the same questions repeatedly, as the prompt tracker does every day.

Layers 1 and 2 tell you what your campaign brings in. Layers 3 and 4 tell you whether ChatGPT knows your business at all. If you only steer on layer 1, you optimize a channel you only see half of.

Measuring paid AI traffic is half the job

Once you measure paid AI traffic, the next question is unavoidable: how much traffic comes in without an ad? That is the same measurement question without a campaign around it, and OpenAI’s tools do not answer it.

For that you look at two things: which AI crawlers fetch your pages, and which visits can be attributed to an AI platform. See measuring AI visibility: metrics and tools.

The bidding models you can steer with this measurement are in what ChatGPT ads cost. The full explanation of the channel is in advertising in ChatGPT.