ChatGPT Ads get visual formats: what marketers should measure
OpenAI is testing visual ads during image generation and expanding attribution, conversion-data and incrementality measurement. Here’s how marketers should evaluate the channel without over-reading early case studies.
OpenAI is giving ChatGPT Ads a more visual format and a more complete measurement stack. On October 5, the company announced that it will begin testing visual ads during image generation later this month in the US with an initial group of advertisers. It also expanded conversion-data integrations, attribution partnerships, incrementality work and brand-suitability testing.
The visual format is an initial US test, not a broad global launch. OpenAI has not disclosed pricing, auction mechanics, advertiser-group size or a broader rollout timetable.
What changed
- Visual ads can show product inspiration, usage or the experience a product enables, with the first test during image generation.
- Hightouch, Tealium and LiveRamp integrations can help advertisers send conversion data from existing systems.
- OpenAI lists attribution partners including AppsFlyer, Triple Whale, Adjust, DV Rockerbox, Northbeam, Branch, Singular, Kochava, Airbridge and Tenjin.
- OpenAI says it is exploring geo-based incrementality experiments with Haus, Measured and WorkMagic.
- Brand-suitability evaluation pilots are being developed with DoubleVerify and Integral Ad Science.
The measurement stack matters more than the format
The durable change is not simply that ChatGPT can show a visual ad. The channel is becoming measurable through layers marketers already use: conversion signals, attribution, incrementality and suitability partners. That makes it easier to evaluate ChatGPT Ads alongside paid search and paid social rather than as an isolated experiment.
Different metrics answer different questions
| Layer | Useful for | Does not prove alone |
|---|---|---|
| Attribution | Associating conversions with ChatGPT Ads under a model | Those conversions would not have happened without the ads |
| New-customer share | Seeing whether traffic appears incremental to the known customer pool | Causal lift |
| Geo incrementality | Estimating causal impact against a control | That the result generalizes to every brand |
| Suitability evaluation | Checking placement context against brand rules | Profitability |
Read the early numbers as signals, not benchmarks
OpenAI cites several encouraging examples. According to DV Rockerbox, WeightWatchers had an attributed cost per acquisition 15.3% below its blended paid-search benchmark. WorkMagic reported that 67% of incremental purchases for Dose came from net-new customers. Triple Whale reported that 93% of Portland Leather visitors from ChatGPT Ads were new.
These are individual brand examples reported by measurement partners and cited by OpenAI. They use different methods and answer different questions. A lower attributed CPA, an incremental-purchase estimate and a new-visitor share are not interchangeable measures of ROAS.
A practical test plan
- Define one business outcome before launch: qualified lead, first purchase, subscription, booked call or another event that can be reconciled with your source of truth.
- Verify conversion data and event matching before judging performance.
- Compare attributed CPA or ROAS with existing channels while keeping attribution assumptions visible.
- Track new-customer share separately from total conversions.
- When spend is large enough, use a holdout or geo test to estimate causal lift instead of relying only on attributed conversions.
- Evaluate creative response and placement suitability separately from business impact.
- Scale only after both efficiency and incrementality checks support the decision.
What is still unknown
OpenAI has not disclosed broad pricing or auction mechanics for the visual format, the size of the initial advertiser group, or a general rollout timetable. It also has not published a multi-advertiser benchmark that would justify using the early case-study numbers as planning assumptions.
ChatGPT Ads are becoming easier to evaluate like a serious performance channel. The useful approach is not to assume conversational advertising will beat search or social; it is to test with clean conversion data, use attribution for diagnosis and incrementality for proof. Measurement discipline should decide whether budget scales.
Sources & useful resources
- OpenAI announcement— Primary source, Oct. 5, 2026