How to Measure ChatGPT Ads: Conversion Tracking, Attribution and Lift

To measure ChatGPT ads, install the OpenAI Pixel on your site, send the same conversions through the Conversions API, and choose an attribution window in Ads Manager. Check those numbers against your own analytics. If you need to prove the ads created sales that wouldn't have happened anyway, run a geo-lift test with a measurement partner.

Measurement is the question we hear most from brands and agencies weighing ads in AI chat. It has also changed fast this year. The pixel and Conversions API arrived with self-serve Ads Manager in May, and on October 5, 2026, OpenAI announced a wider set of partners for attribution, incrementality and brand lift.

What does ChatGPT Ads Manager report?

Ads Manager reports impressions, clicks, spend, CTR, average CPC, average CPM and conversions at the campaign, ad group and ad level. Sales and ROAS columns are calculated only from attributed order_created events, so other conversion types don't count toward them. You can break results down by device, country and platform, and export them as CSV.

Numbers arrive at different speeds. Impressions, clicks and CTR refresh about every 15 minutes. Spend lags by 7 to 8 hours, and attributed conversions can take 24 to 48 hours to appear, so yesterday's CPA will still move.

How do you set up conversion tracking for ChatGPT ads?

Create a data source in Ads Manager, then send conversion events with the OpenAI Pixel, the Conversions API or both. Using both is "more resilient," OpenAI says.

1. Install the Pixel on every page where a conversion can happen. Google Tag Manager works, as long as it loads the snippet reliably.
2. Keep the click reference. OpenAI appends an oppref parameter to your landing page URL. The Pixel stores it in a first-party cookie. Redirects and checkout handoffs must not drop it.
3. Send server-side events through the Conversions API, with the oppref value when you have it.
4. Use the same event ID when one conversion goes through both the Pixel and the Conversions API, so OpenAI counts it once.
5. Match the campaign's conversion event exactly. An event that fires but doesn't match the event configured for the campaign shows as 0, and fixing it later won't backfill earlier events.

The Pixel also runs automatic advanced matching. It picks up details people type into forms, hashes them in the browser and sends the hash with the conversion, which helps when the click reference is missing. Where available, OpenAI may also model conversions it can't match, and reported totals "may include modeled conversions."

Mobile apps send events through a mobile measurement partner, set up with a Pixel ID and Conversions API key from Ads Manager. Adjust, Airbridge, AppsFlyer, Branch, Kochava, Singular and Tenjin are supported.

Which attribution windows can you choose?

Ads Manager offers a 7-, 14- or 30-day click-through window and an optional one-day view-through window, in any combination. The window changes reporting only. OpenAI says it "does not change campaign optimization, bidding, or billing," and the model is last-touch, not multi-touch.

View-through counts a conversion within a day of someone seeing an ad they didn't click. If both a click and an impression qualify, the click gets the credit. OpenAI reports that in an early analysis, "52.7% of eligible one-day view-through conversions occurred within an hour of the matched ad impression." OpenAI's explanation is that people come to ChatGPT to work through a decision, and clicking an ad "may not be their immediate next step."

Why don't ChatGPT ads numbers match GA4?

Because the two systems count different things. Ads Manager credits a conversion to an ad click or view under OpenAI's rules. GA4 credits its own sessions under its own model, and its counts shift with consent settings, browser blocking, redirects and time zones. OpenAI's help center is blunt about it: "A difference does not necessarily indicate an error."

Make the comparison fair. Add UTM tags to every ad URL so GA4 can separate paid clicks from ChatGPT's organic referrals. Ads Manager can fill in IDs for you with the {campaign_id}, {ad_group_id} and {ad_id} macros. Then compare the same dates in the same time zone, and expect a gap rather than a match.

Which measurement partners work with ChatGPT ads?

As of October 5, 2026, OpenAI lists these partners. Setup steps differ by partner, and most start with an API key from Ads Manager settings.

• Sending conversion data: Hightouch, Tealium and LiveRamp, including offline sales through LiveRamp.
• Mobile app attribution: Adjust, Airbridge, AppsFlyer, Branch, Kochava, Singular and Tenjin.
• Web attribution: Triple Whale, Northbeam and DV Rockerbox.
• Full-funnel measurement: Fospha, Measured and INCRMNTAL.
• Incrementality: Haus, Measured and WorkMagic, through geo-based experiments.
• Brand lift: Kantar and Cint, in early testing.
• Brand suitability: DoubleVerify and Integral Ad Science, in pilots that test OpenAI's safeguards "without accessing private user conversations."

How do you prove ChatGPT ads drove extra sales?

With an incrementality test. In a geo-lift test, you run ads in some regions and hold them back in matched ones, then compare sales. It doesn't depend on clicks or cookies, so it also catches people who saw an ad and bought later. It does need regions you can switch off and enough sales in each to show a difference.

OpenAI calls its own incrementality work "still in its early stages." The first results are partner studies OpenAI published with the announcement. WorkMagic's geo-lift study for wellness brand Dose estimated 2.3 times more incremental orders than last-click attribution captured. DV Rockerbox measured WeightWatchers' attributed CPA at 15.3% below its blended paid search benchmark. Treat them as signs the method works, not as a forecast for your account. Our cost guide collects independent figures.

How do you know your conversion data is good?

Check the Event Quality Score for each pixel or data source. It refreshes daily over the last seven days and flags common problems:

• Conversions arrive without the ad click reference. Check that redirects keep oppref.
• Few events carry an email address or a stable customer ID.
• Server events arrive more than an hour after the action.
• The same action fires more than once, from reloads, duplicate tags or retries.

The score describes your data, not your campaign's performance. OpenAI warns against gaming it: "Do not send extra events, invent identifiers, or change event times to improve a score."

Can you measure ads in Google AI Overviews and Copilot?

Not separately. "Google Ads currently doesn't offer segmented reporting when ads show within Search AI Overviews," according to Google. Microsoft says it "does not currently support specific metrics for ads served in Copilot." In both, the ads come from campaigns you already run, and their results are mixed into those campaigns' totals.

To judge them, measure the whole campaign, or use a geo-lift test. Our guide to which AI assistants sell ads covers how to buy each one, and the buyer's guide covers setting up a first ChatGPT campaign.

Measuring across more than one AI app

Each platform counts with its own windows and rules, so adding their reports together can double-count. Pick one source of truth, usually your own analytics or an attribution partner, and judge every AI channel against it.

Thrad runs ads across AI apps with one budget and one report.