Signal — paid AI advertising
ChatGPT Ads has more ways to credit a sale to an ad. Credit still does not prove the ad caused it.
On 5 October 2026 OpenAI gave advertisers more ways to send their sales data to ChatGPT Ads and more partners that credit those sales to an ad. The report gets richer. It still cannot tell you which customers would have come anyway — and OpenAI calls its own work on that question early.
Published For ChatGPT Ads advertisers Measurement tools announced; causal testing early
The short version
Keep three numbers apart. Sales your own till or booking system recorded. Sales a measurement tool credited to the ad. Sales that happened only because the ad ran. OpenAI’s announcement mostly improves the second. Only a fair comparison — some customers or areas with the ad, some without — answers the third.
The ordinary problem
The ad report and the till rarely tell the same story
The bike shop below is an illustrative example written for this page — not a customer, a case study or a measured result.
Take a bike shop with two locations and an online store. The owner runs a small ChatGPT Ads campaign for a month, and the campaign report credits a set of sales to it.
- Some of those sales are real and new: somebody saw the ad, had never heard of the shop, and bought a bike.
- Some are real but not new: a regular who already had the shop bookmarked saw the ad on the way and bought what they were going to buy that week anyway.
- Some sales the ad did help never show up in the report: somebody saw the ad without clicking, then walked into the shop the following week.
All three groups end up in the same till. The first and the third may include extra business, even though only the first shows up in the report. The report alone cannot tell you which sales would have happened anyway.
The mental model
Three different things get called “conversions”
Each one answers a different question. Read your results in this order, and do not let a number from one row stand in for another.
Recorded conversion
- The question it answers
- Did a sale or booking actually happen?
- Where the answer comes from
- Your own till, booking system or online checkout.
The sale is in your own records. This is the only one of the three that is a plain fact about your business, and every other number should reconcile to it.
Attributed conversion
- The question it answers
- Which ad gets the credit for that sale?
- Where the answer comes from
- A measurement rule applied by the ad platform or a partner.
A recorded sale is matched to an ad: after a click within the window you chose, or — if you switch it on — within one day of somebody seeing the ad without clicking. Attribution is a rule for handing out credit. It can give an ad too much credit (the regular who was coming anyway) or too little (the customer who saw the ad and came in later).
Incremental demand
- The question it answers
- Would that sale have happened without the ad?
- Where the answer comes from
- A comparison between people or areas that saw the ad and similar ones that did not.
The sales that exist only because the ad ran. You cannot read this off any attribution report; it takes a test with a fair comparison group. OpenAI says its own work here is still in its early stages.
A recorded sale is not an attributed sale. An attributed sale is not an extra sale.
What changed
What OpenAI announced on 5 October 2026
OpenAI’s announcement and its companion Ads blog post describe new measurement tools. They are not all at the same stage, so here they are in three groups.
Announced as available now
- Integrations with Hightouch, Tealium and LiveRamp, so advertisers can send sales and other conversion data from systems they already use to ChatGPT Ads.
- Support for attribution partners across web and app — AppsFlyer, Triple Whale, Adjust, DV Rockerbox, Northbeam, Branch, Singular, Kochava, Airbridge and Tenjin — plus full-funnel measurement partners Fospha, Measured and INCRMNTAL.
- A reporting setting for the click-through attribution window, and an optional one-day view-through window. OpenAI says these settings change what the report credits, not how the campaign itself is optimized.
- Negative Phrases, which let a brand keep its ads out of placements for narrow reasons specific to its own policies — available to qualifying advertisers.
Early work or pilots
- Work with Haus, Measured and WorkMagic exploring geo-based experiments, to help advertisers understand the causal impact of ChatGPT Ads. OpenAI says its incrementality work is still in its early stages.
- Brand-suitability evaluation pilots OpenAI is developing with DoubleVerify and Integral Ad Science, designed to assess its safeguards in a controlled testing environment without access to private user conversations.
- Early-stage testing with Kantar and Cint on brand measures.
Not launched yet
- A visual ad format shown during image generation in ChatGPT. Testing begins later in October 2026 in the US with an initial group of advertisers.
OpenAI also says advertising does not influence ChatGPT’s answers. Being named in an answer and paying for an ad remain two separate things, and belong in two separate rows of your reporting.
Partner results
The published results are partners’ claims, not your forecast
OpenAI’s announcement and its Ads blog quote early results from measurement partners. They are reported here as claims, with who made them. Each card says what kind of figure it is: an attributed figure, a comparison test, or a share of visitors.
Reported claimDV Rockerbox Attributed figure
15.3% lower
According to DV Rockerbox, WeightWatchers’ attributed cost per acquisition on ChatGPT Ads was 15.3% lower than its blended paid-search benchmark. This is an attributed figure — credit, not cause.
Reported claimWorkMagic Comparison (lift) test
67%
WorkMagic reported statistically significant lift for the wellness brand Dose, with 67% of the incremental purchases coming from net-new customers.
Reported claimWorkMagic Comparison (geo-lift) test
2.3×
WorkMagic’s geo-lift study for Dose estimated 2.3 times more incremental orders than last-click attribution captured — an example of last-click reporting giving an ad too little credit.
Reported claimTriple Whale Share of visitors
93%
According to Triple Whale, 93% of Portland Leather’s visitors from ChatGPT Ads were new. New visitors are not the same as new sales.
We have not seen the underlying data, and none of these is presented as a local-business result. Treat them as evidence that measurement partners are active on the platform, not a forecast for your business.
Plain words
Four terms, once each
- Conversion data
- A record of an action you care about — a purchase, a booking, a sign-up — sent from your systems to the ad platform so it can be matched to ads.
- Attribution window
- How long after a click (or, for view-through, after somebody sees the ad) a sale can still be credited to that ad.
- View-through conversion
- A sale credited to an ad that the customer saw but did not click.
- Holdout or geo test
- A comparison where some customers or areas are deliberately not shown the ad, so the difference in sales can be measured.
Evidence boundary
What this does not show
- It does not show that ChatGPT Ads bring extra customers to any local business.
- The partner results are figures reported by OpenAI and its partners. We have not checked them independently.
- The announcement does not say in which countries each measurement tool is offered. Negative Phrases are limited to qualifying advertisers.
- The visual ad format has not launched; its test begins later in October 2026 in the US with an initial group of advertisers.
- The brand-suitability evaluations are pilots in a controlled testing environment, not an audit of live conversations.
- We did not open an advertiser account or run a campaign to check any of this.
- Nothing on this page describes an Obenan product, integration or customer result.
What to do next
Four things to settle before you trust an ad report
None of these require ChatGPT Ads. They make any paid campaign easier to judge.
Count from your own records first
Your till, booking system and online checkout are the ledger. An ad report is a second opinion about which of those sales an ad should get credit for.
Write down the attribution settings
Note the click window you chose and whether view-through is switched on before you read the results. A longer window or view-through can credit more sales to the same ad, never fewer.
Ask who would have come anyway
Check how many credited sales came from existing customers. A campaign that mostly reaches your regulars can look strong in an attributed report and add little.
Plan a fair comparison before a bigger budget
Only a comparison — an area or group of customers without the ad, against a similar one with it, over the same weeks — speaks to extra demand. For a single shop the numbers will be noisy, so treat the result as a direction, not proof.
Sources
Where each claim on this page comes from
Every provider claim above is attributed below, with the date it was published and the date we last checked it.
- Building advertising for the way people use AI
OpenAI Published Checked
Conversion-data integrations, attribution and measurement partners, early geo-based incrementality work, Negative Phrases, brand-suitability pilots, the visual-format test, and the partner results quoted above.
- More ways to measure ChatGPT Ads
OpenAI Published Checked
The click-through and one-day view-through attribution windows, that those settings change reporting rather than optimization, Kantar and Cint brand-measurement testing, and the WorkMagic geo-lift comparison with last-click attribution.