On October 5, 2026, OpenAI published a post about measuring ChatGPT Ads. It names more than twenty companies, five early results and a handful of settings, and it is the first thing I would read before spending money on the channel. This article says what the post announces, what each number does and does not show, what OpenAI’s Help Center adds, and how I would test ChatGPT Ads in a store.
I am writing this a day after the post, so it is a reading of OpenAI’s own pages, not a report from a campaign. Every fact comes from the pages listed at the end, which I read on October 6, and nothing here has been tested. The four-week plan is a design to try, and the numbers in the one example are made up.
What did OpenAI announce about ChatGPT Ads measurement?#
OpenAI announced more ways to measure ChatGPT Ads: a pixel and a Conversions API for sending conversions, partner integrations for attribution and incrementality, reporting windows for clicks and views, and brand suitability pilots. The post is More ways to measure ChatGPT Ads. The facts, as it gives them:
- Announced: October 5, 2026, on OpenAI’s Ads blog. A second post the same day introduced a new visual ad format.
- Signals: the OpenAI Pixel and the Conversions API (CAPI) send purchases, leads, installs and sign-ups. An Event Quality Score (EQS) rates those signals.
- Partners: Hightouch, Tealium and LiveRamp for sending data; ten named attribution partners (AppsFlyer, Triple Whale, Adjust, DV Rockerbox, Northbeam, Branch, Singular, Kochava, Airbridge and Tenjin); Fospha, Measured and INCRMNTAL for fuller measurement; Haus, Measured and WorkMagic for geo-based incrementality tests; Kantar and Cint for early-stage brand tests; DoubleVerify and Integral Ad Science for brand suitability pilots in development.
- Reporting: a click window that fits your buying cycle, plus an optional one-day view-through window. The post says these settings change reports, not how campaigns optimize.
- Optimization: advanced conversion optimization learns from eligible click-through and view-through conversions for the goal you choose.
- Brand safety: Negative Phrases, for placement rules specific to a brand, are now available to qualifying advertisers.
- Reach: OpenAI says ChatGPT reaches 1.2 billion people each week. It does not say how many of them see ads.
What do the five early results show?#
They show five cases reported by OpenAI’s partners, each measuring something different, so they cannot be added together or read as a benchmark. OpenAI’s own FAQ agrees: it says ChatGPT Ads does not yet have performance benchmarks across advertisers, industries or campaign types.
| Result in the post | Who measured it | What it compares | What the post does not say |
|---|
| Cost per acquisition 15.3 percent lower | DV Rockerbox, for WeightWatchers | Attributed CPA on ChatGPT Ads against the advertiser’s blended paid search benchmark | Spend, dates, number of conversions |
| 67 percent of incremental purchases from net-new customers | WorkMagic, for Dose | The share of extra purchases found by a lift study that came from new customers | Test dates, spend, how net-new was decided |
| 100 percent of campaigns with incremental conversions | INCRMNTAL | Campaigns with statistically significant lift | How many campaigns or advertisers, how large the lift was |
| 93 percent new visitors | Triple Whale, for Portland Leather Goods | The share of visitors from ChatGPT Ads who were new | Orders and revenue: a visitor is not a buyer |
| 83 percent more revenue per 1,000 impressions | Fospha | ChatGPT Ads against the paid social average for the same advertisers | How many advertisers, what the impressions cost |
Two more figures sit in the text. In an early analysis of global campaigns optimized for deeper-funnel outcomes, 52.7 percent of eligible one-day view-through conversions happened within an hour of the matched impression. And WorkMagic’s geo-lift study estimated 2.3 times more incremental orders for Dose than last-click attribution captured. A geo-lift study, as the term is normally used, compares sales in regions where ads ran with regions where they did not, so it does not depend on clicks. Neither figure comes with dates or spend.
Two kinds of number are mixed in the table. The DV Rockerbox, Triple Whale and Fospha results come from attribution tools, which say who gets credit under their own rules. The WorkMagic and INCRMNTAL results are about lift, which tries to say what would not have happened without the ads. OpenAI itself calls its incrementality work early.
Why is revenue per 1,000 impressions not a return?#
Because it leaves out what the impressions cost. Here is an example with invented numbers, to show the arithmetic. They are not prices from ChatGPT Ads or any other platform:
| Measure | Paid social | ChatGPT Ads |
|---|
| Revenue per 1,000 impressions | 10.00 euros | 18.30 euros (83 percent more) |
| Cost per 1,000 impressions (CPM) | 4.00 euros | 9.00 euros |
| Revenue per euro spent | 2.50 | 2.03 |
ChatGPT Ads earns 83 percent more per impression and still returns less per euro, because each impression costs more than twice as much. It would need a CPM below 7.32 euros to match paid social here. Below that it wins, above it it loses. The post gives no impression prices, so the Fospha figure cannot be turned into a return. Ads Manager does report average CPM, and a ROAS column built from order_created events, so for your own campaigns you can do the sum.
Is ChatGPT Ads available in Sweden?#
Yes. OpenAI’s Ads Manager availability page lists Sweden as available for self-service, and its August 18 announcement named Sweden among 31 European markets. Two conditions apply: the legal entity that advertises and is billed must be based in a listed country, and ads appear only to people on the Free and Go plans. The post does not say which partner integrations work in which markets, and the new visual ad format starts as a US test later in October.
How do you set up measurement for ChatGPT Ads?#
Create a data source in Ads Manager, send conversion events with the OpenAI Pixel, the Conversions API or both, and choose your reporting windows. The post announces the tools and the steps are in the Help Center. These matter most for a store:
- Install the Pixel on the pages that matter. A tag manager such as Google Tag Manager works if it loads the snippet reliably and does not reorder the calls.
- Send server-side events, such as confirmed purchases, through the Conversions API. Give a browser event and its server copy the same event ID, and pass the click reference (
oppref) that the Pixel keeps in a first-party cookie. - Turn on automatic advanced matching. The Pixel reads customer details from your forms and hashes them in the browser, and the Help Center says raw details are not sent through it.
- Read Event Quality. Each data source gets a score and warnings, refreshed daily from the last seven complete days. It does not predict performance, and the Help Center says not to add events or change times to lift a score.
- Add landing page query parameters in Ads Manager: your own UTMs, and macros such as
{campaign_id}, so GA4 and your store can see the visits. Some partners ask for their own parameters. - Choose reporting windows: 7, 14 or 30 days after a click, and 0 or 1 day after a view. Reporting uses last-touch attribution, not multi-touch.
- Connect a partner if you already pay for one. Most connect with an Ads Manager API key.
Allow 24 to 48 hours for attributed conversions to appear.
Do the reporting windows change how a campaign optimizes?#
No. The post and the Help Center both say the 7, 14 and 30-day click windows and the one-day view window change reporting only. What changes optimization is the billing choice, which the post does not mention. With click billing, a conversion optimization campaign learns from conversions that follow a click. With impression billing it learns from a broader set that includes conversions after a view, and the Help Center says choosing clicks will lower your conversions. A campaign uses one standard event, and custom events are not supported for it yet. My reading: a report set to clicks only can sit beside a campaign that is learning from views.
How would I test ChatGPT Ads in a store?#
I would run one small campaign for four weeks, with one conversion event, tagged visits and the platform, GA4 and the store’s orders read side by side. It is a design to try, not a result:
- Decide what decides. Pick MER and the cost of a new customer, and write down the line they must clear before you start.
- Cap the loss. A daily budget is an average, and the Help Center says one day can spend up to twice it.
- Set up the signals first: the Pixel, server events and a look at the Event Quality warnings before the campaign starts.
- Create one conversion optimization campaign on the purchase event. I would start with click billing, because a click leaves a visit that GA4 and the store can check and a view never does. The trade-off is fewer conversions.
- Tag every ad with your own UTMs, then confirm in GA4 that the first clicks arrive and see where they are grouped. Organic visits from ChatGPT are a separate matter, covered in the article on AI shopping assistants.
- Set a click window that fits the store’s buying cycle and read the view-through line separately. Compare by conversion time, because the default report dates a conversion by the click or view that earned the credit.
- After four weeks, read the platform with clicks only, the platform with views, GA4 and the store’s orders side by side, the way the reconciliation article does. Expect gaps. The Help Center lists the usual reasons: windows, time zones, consent and storage, deduplication and modeled conversions.
- If spending grows enough to matter, ask a partner about a geo test. The post names Haus, Measured and WorkMagic. A geo test needs enough orders in each region to see a difference, which a small store may not have.
What should you watch out for?#
Five things:
- It is a beta. OpenAI’s FAQ calls ChatGPT Ads a beta in the early stages of scaling and says delivery and spend can fluctuate. Settings change, so check how recently each Help Center page was updated.
- The results are cases, not benchmarks. Ask for the denominators: how many advertisers, what period, what spend, attributed or incremental.
- Totals can include modeled conversions. The Help Center says reported conversion totals may include modeled conversions where available.
- The partner list is wider than the setup pages. On October 6 the setup pages did not list Tealium, Measured or Haus, which the post names. Check your partner’s page before you plan around it.
- Consent is yours to settle. The Help Center says to send conversion data only after giving users clear information and getting any consent the law requires. The Pixel’s cookie and automatic advanced matching are both questions for whoever owns privacy for your store. What happens to measurement when a visitor says no is in my GA4 and Piwik PRO article.
Where this leaves me#
What I like about the post is that it names names: partners you can check, windows you can set and a score for your own data. What I would not do is plan a budget around five numbers with no denominators. The most useful sentence is on the Help Center’s FAQ page: there are no benchmarks yet. So the work is the usual work: set up the signals, tag the visits, keep the test small and read it next to the store’s own orders. A launch post leads with its best numbers, which is its job. Reading them is mine.
Sources: OpenAI Ads, More ways to measure ChatGPT Ads (October 5, 2026). OpenAI, Building advertising for the way people use AI (October 5, 2026) and ChatGPT Ads expands across Europe (August 18, 2026, updated August 31). OpenAI Help Center, read on October 6, 2026: Measure Results, Conversion Measurement, Conversion-optimized Campaigns, Understand and improve event quality, Set up Measurement Partner Integrations, Set up Mobile Measurement Partner Integrations, Frequently asked questions and Ads Manager Availability. ChatGPT Ads is in beta and its documentation changes often, so some details here may have moved on.