A dated, source-linked record of ChatGPT Ads: market availability, objectives, bidding, budgets, context hints, product feeds, measurement, policy and observed performance. We separate what OpenAI documents from what campaigns show and what remains a hypothesis.
This is not a mirror of release notes. We record changes that can alter an advertising decision, link the primary source and keep uncertainty visible when the evidence is not yet strong enough for a conclusion.
Availability. The official Ads Manager page checked today lists 52 countries with self-service access. Spain and the other 30 markets in OpenAI’s European expansion are included, but self-service availability now extends well beyond Europe.
Buying and optimisation. General campaign documentation lists CPM, CPC and oCPC. The latest conversion-optimised campaign documentation also introduces oCPM in beta, with access expanding.
Measurement. OpenAI documents Pixel and Conversions API, the oppref click reference, event-ID deduplication and eligible advanced-matching signals where available.
Retail. Ads Manager supports campaigns built from product feeds. Policy. OpenAI’s current advertising-policy changelog is at v1.6, updated 10 September 2026.
Primary sources: Ads Manager Availability · Conversion-optimized Campaigns · Conversion Measurement · Ad policies.
OpenAI now documents two billing models for conversion-optimised campaigns: oCPC, billed on valid clicks, and oCPM, billed on impressions. oCPM can optimise towards a selected conversion while accounting for eligible signals after both ad clicks and ad views. OpenAI describes access as rapidly expanding rather than universal.
This changes a practical planning decision. Objective, billing model and measurement can no longer be treated as separate setup choices: conversion tracking must be in place, the campaign needs a supported standard conversion event, and an existing campaign cannot simply be switched to a different objective or billing model.
Primary source: Conversion-optimized Campaigns · OpenAI Help Center.
The product-feed documentation checked today supports CSV/TXT uploads, hosted URLs and SFTP connections. Feed items expire after two weeks without refresh, so a large catalogue is not a one-off setup task; it needs a reliable synchronisation process.
For eligible new ad groups, Maximize results can be selected by default and automatically adjust bids towards clicks or conversions. The ads_metadata field can add advertiser-defined attributes when standard product filters are not enough.
OpenAI says advertisers using product feeds have been among the platform’s strongest performers. We record that as a vendor claim, not an independent benchmark: the public documentation does not provide a comparable denominator by market, vertical or spend level. Primary source: Create Campaigns from Product Feeds · OpenAI Help Center.
OpenAI’s v1.6 advertising-policy update clarifies its right to reject ads that conflict with its advertising principles, business interests or competitive position. Placement rules continue to restrict ads around sensitive user contexts and brand-unsafe conversations.
The operational consequence is that policy cannot be treated as a launch-day checkbox. Advertiser eligibility, claims and permitted contexts can change, so policy review belongs in ongoing campaign operations.
Primary source: Ad policies · OpenAI — changelog v1.6, September 2026.
Ads Manager now treats a daily budget as the average amount to spend per day over a seven-day period. On an individual day, billed media spend can reach up to twice the selected daily budget; across the applicable seven-day period, it should not exceed seven times that budget.
Existing campaigns adopt this behaviour automatically. That matters for pacing and alerting: spending above the nominal daily amount on one day is not, by itself, evidence of campaign overspend. The full spending window matters.
Primary source: Daily Budgets · OpenAI Help Center.
OpenAI opened Ads Manager self-service across the 31 European markets it had announced. By 13 September, the official availability page lists 52 self-service countries, making the current footprint materially broader than the original US-and-Europe narrative.
On the same date, OpenAI said ChatGPT Ads had reached a $1 billion annualised revenue run rate and was being used by tens of thousands of advertisers. That is evidence of commercial scale, not evidence that any individual advertiser should expect a particular CPC, CPA or ROAS.
Primary sources: European expansion · current Ads Manager availability · ChatGPT Ads scale milestone.
OpenAI explicitly says ChatGPT Ads does not yet have performance benchmarks across advertisers, industries or campaign types. There is still no defensible public number for a “normal” CPC, CTR, CPA or ROAS across Europe — or globally.
We are also watching four unresolved questions: effective oCPM availability by account and market; the incremental value of context hints versus alternative structures; how much feed performance is driven by catalogue quality, format or user intent; and how much conversion value is truly incremental versus Google Ads or other channels.
None of these becomes a conclusion until there is a sufficiently strong public source or first-party sample with context. Primary source: Frequently asked questions · OpenAI Help Center.
We separate what the platform allows, what performance data shows and how each result was measured. A new product feature is never presented as a performance promise.
Date, source and context before a release, observation or test is promoted into a recommendation.
A change that can alter an advertising decision: market availability, objective, billing model, bidding, budget, ad groups, context hints, creative, product feeds, measurement, reporting or policy. We do not log every interface change if it does not affect operations.
Documented means published by OpenAI and linked to a primary source. Observed means behaviour seen in campaign or reporting data with enough context to interpret it. Hypothesis is an explanation that still needs testing. We do not use the same degree of certainty for all three.
Yes, when the sample is defensible and can be accompanied by objective, market, vertical, period and sample size. Until then, “no benchmark yet” is more useful than a number without a denominator.
Because OpenAI does not yet publish cross-advertiser performance benchmarks, and a small early sample is not enough to define a normal CPC for a market. Recommended bids or bid caps should not be mistaken for the CPC actually paid.
We do not assume so. OpenAI currently labels oCPM as beta and says access is rapidly expanding. The Observatory therefore records it as an expanding capability, not a universal feature across every account and market.
Only in aggregated and anonymised form where permission, sample size and context make the data useful without exposing identifiable commercial information.
Whenever we verify a material product change or new evidence. The visible snapshot date shows the last full verification; individual entries keep their own date, or a verified-on date when the source does not publish an exact change date.
No. Europe is a priority because of the August 2026 expansion, but Ads Manager is already available across multiple regions. We separate markets whenever availability, policy, product behaviour or performance makes aggregation misleading.
We can design a measured pilot from day one, with explicit hypotheses and criteria for what to scale, correct or stop.
What you sell, which markets you work in and what you want to learn from the first pilot. If the scope is not clear yet, we will define it with you.