OpenAI began testing ads in ChatGPT on 9 February 2026 and expanded through August to nine markets: the US, Canada, Australia, New Zealand, UK, Mexico, Brazil, Japan and South Korea. Here is the detail that should shape your entire decision about whether to participate: ads serve only to Free and Go tier users. Plus, Pro, Business, Enterprise and Edu are excluded.
That's not a minor implementation note. It's a structural targeting constraint. Every person who has demonstrated willingness to pay for AI tooling, including most of the professional and business audience, is invisible to you on this channel. You can reach the non-paying segment only. For a large number of B2B and premium-consumer advertisers, that is the wrong half of the market.
I don't think that makes the channel worthless. It makes it a specific tool for a specific job, and worth entering with clear eyes rather than as a land-grab.
Key Takeaways
- OpenAI started testing ads in ChatGPT on 9 February 2026 (openai.com).
- Live across nine markets as of August 2026: US, Canada, Australia, New Zealand, UK, Mexico, Brazil, Japan, South Korea.
- Free and Go tiers only: Plus, Pro, Business, Enterprise and Edu users never see ads.
- That means your reachable audience is structurally skewed away from paying users, including most of the professional segment.
- Ads are matched to conversation topic, past chats, and prior ad interactions.
- OpenAI states ads don't influence answers, and advertisers receive only aggregate impression and click data.
- Advertiser signup: openai.com/advertisers.
- There is no established playbook. Treat a first campaign as a learning budget with a defined kill criterion.
The Tier Constraint Is the Story
Most coverage of ChatGPT ads has focused on the novelty of ads appearing in an AI assistant at all. The more consequential fact is who they can reach.
What "Free and Go only" actually filters out
Paid tiers are excluded. That removes, in one stroke:
- Individual professionals who pay for Plus because they use it for work.
- Power users on Pro.
- Every seat inside a company on Business or Enterprise.
- Education deployments on Edu.
What remains is people using ChatGPT without paying for it: a large, real, and commercially meaningful audience, but a specific one. It skews toward casual and exploratory use, toward lower willingness to pay for software, and away from the organisational buyer entirely.
Who this is genuinely good for
If you sell to a broad consumer audience, this is a legitimate reach channel with the usual early-channel pricing advantages. Consumer apps, DTC products, entertainment, education, mass-market services, local businesses in the nine live markets, the audience constraint barely bites.
Who should think hard first
If you sell B2B software, professional services, or anything with a five-figure contract value, your buyer is disproportionately likely to be sitting behind an excluded tier. Not entirely, plenty of professionals use the free tier, but the correlation between "pays for AI tooling" and "buys professional software" is not one you should ignore.
Somebody will run a B2B campaign here and post about the results. Until then, treat B2B as an unproven use of the channel rather than an obvious one.
The honest uncertainty
Tier boundaries are a product decision, and product decisions change. If OpenAI extends ads to paid tiers, the calculus shifts substantially. Check the current position at openai.com before planning around what's written here.
How the Targeting Works
Per OpenAI, ads are matched using three signals: the topic of the current conversation, past chats, and prior interactions with ads.
Conversation topic
The most distinctive signal, and the one with no real precedent. A user discussing a problem in detail is expressing intent with far more context than a search query carries: they've described their situation, their constraints, and often what they've already tried.
Past chats
Behavioural history, familiar in shape from other platforms but built from an unusually rich substrate. What someone discusses with an assistant over months is a more revealing profile than browsing history.
Prior ad interactions
Standard feedback-loop optimisation.
What you don't get to do with any of this
You are not, from the available information, building the audience. The matching is done by the platform. This is closer to contextual placement than to the granular audience construction you'd do in Meta Ads Manager: and if your team's competence is built on audience-building craft, much of that skill doesn't transfer here.
Control and Measurement: The Honest Table
| Dimension | What you control / receive | What you don't |
|---|---|---|
| Audience tier | Nothing: serving is Free and Go only | Cannot reach Plus, Pro, Business, Enterprise or Edu users at all |
| Geography | The nine live markets (US, CA, AU, NZ, UK, MX, BR, JP, KR) | Anywhere else, until OpenAI expands |
| Targeting signals | Platform matches on conversation topic, past chats, prior ad interactions | You aren't constructing granular custom audiences yourself |
| Answer content | Nothing, OpenAI states ads don't influence answers | You cannot buy your way into the assistant's response |
| Creative | Your ad creative and destination | How and where it renders in the conversation flow |
| Reporting | Aggregate impression and click data | User-level data, conversation content, individual identifiers |
| Attribution | Whatever you instrument on your own site (UTMs, landing pages, post-conversion surveys) | Platform-side conversion tracking at the granularity you're used to |
| Benchmarks | Your own test data | Industry benchmarks, the channel is too new for reliable ones |
What you can and can't control or measure in ChatGPT ads, based on OpenAI's published information as of September 2026.
The two constraints that matter most
Ads don't influence answers. OpenAI is explicit about this, and it's important for expectation-setting: this channel does not buy you a mention inside the assistant's response. It buys ad placement. If your actual goal is being recommended by the model, that's earned through the same content and authority work as any other citation, and it's a completely separate programme of work.
Aggregate data only. Impressions and clicks. That's the reporting layer. It is thinner than any mature ad platform, and it means your ability to learn from a campaign depends almost entirely on instrumentation you build yourself.
Designing a First Test
There is no established playbook. Anyone selling you one has made it up. Here's how I'd approach it.
Set a learning budget, not a performance budget
Size it as what you'd pay for a genuine answer to "does this channel work for us." Small enough that a zero return is tolerable, large enough to produce interpretable data. Decide the number before you start, and decide in advance what result would make you stop.
Check the tier constraint against your buyer first
Before spending anything: is your buyer plausibly on a free tier? If you sell enterprise software, write down honestly why you think this reaches your buyer. If you can't answer it, don't run the test, run it on a channel where you can.
Build attribution on your own side
Given aggregate-only reporting, everything depends on what you instrument:
- Dedicated landing pages per campaign, so traffic is unambiguous.
- Full UTM discipline on every destination URL.
- A "how did you hear about us" field on your conversion form, with ChatGPT as an option. Self-reported attribution is imperfect and it is genuinely useful when platform data is thin.
- A clean pre-period baseline, so you can see whether direct and branded-search traffic moves during the flight.
Run long enough to read the result
Short flights in a new channel produce noise. Give it enough time and volume to distinguish a real signal from variance, and resist the urge to draw conclusions in week one.
Pick one hypothesis
Not five. "Does this channel produce qualified traffic for our category at a tolerable cost?" is a question you can answer. "What's our optimal creative, audience, offer and landing page combination?" is not, not in a channel with no baselines and thin reporting.
Write down what you learn, including the negatives
You'll be one of relatively few people with real data on this. That's worth something internally, and a documented negative result is a genuinely valuable output, it stops your organisation relitigating the question every quarter.
What This Means for the Broader AI Marketing Landscape
Zoom out and there's a pattern worth naming. Advertising is arriving inside AI assistants: OpenAI here, Google already running ads in AI Mode. The surface where people ask questions is being monetised, exactly as search was.
Paid presence and model recommendation remain separate
This is consistent across both platforms. OpenAI states ads don't influence answers. SE Ranking's analysis of Google AI Mode, reported by Search Engine Land in July 2026, found only 11.53% of advertisers were cited as sources for keywords they bid on.
The conclusion generalises: you cannot buy your way into being recommended by an AI system. You can buy adjacency to the conversation. Being the thing the model actually suggests is earned, and it's earned through the unglamorous work of producing content with real authority, original data, and specific defensible claims.
Which is why I'd keep the budget split honest
Paid placement in AI surfaces is worth testing on its merits: reach, clicks, conversions. Judge it on those. Don't let it absorb the budget for earned visibility work, and don't let anyone report the two as one programme. They have different mechanisms, different timelines, and different ceilings.
Frequently Asked Questions
When did OpenAI launch ads in ChatGPT?
Testing began 9 February 2026, per openai.com/index/testing-ads-in-chatgpt, and expanded through August 2026.
Which countries have ChatGPT ads?
As of the August 2026 expansion: the US, Canada, Australia, New Zealand, UK, Mexico, Brazil, Japan and South Korea.
Who sees ads in ChatGPT?
Free and Go tier users only. Plus, Pro, Business, Enterprise and Edu users are excluded.
Why does the tier restriction matter for targeting?
Because it means you can only reach people who haven't paid for ChatGPT. That skews your reachable audience away from the segment most willing to spend on software and tools: a real constraint, particularly for B2B and premium products.
How are ChatGPT ads targeted?
OpenAI states ads are matched to the topic of the conversation, past chats, and prior interactions with ads. The platform does the matching; you aren't constructing the audience yourself.
Do ads influence ChatGPT's answers?
OpenAI states that they do not. You cannot buy a mention inside the assistant's response: that's earned through content authority, the same as citation in any AI system.
What data do advertisers get?
Aggregate impression and click data. No user-level data, no conversation content, no individual identifiers.
How do I measure ChatGPT ad performance properly?
Build attribution on your own side: dedicated landing pages, disciplined UTMs, a self-reported "how did you hear about us" field on conversion forms, and a clean pre-campaign baseline so you can see lift in direct and branded traffic.
How do I sign up as an advertiser?
Through openai.com/advertisers.
Should B2B companies advertise in ChatGPT?
Proceed carefully. The Free-and-Go-only constraint likely excludes a meaningful share of professional and enterprise buyers. If you test it, do so with a small learning budget and a written hypothesis about why your buyer sits on a free tier.
What should a first ChatGPT ads test cost?
Whatever you'd pay for a clear answer to whether the channel works for you: with a kill criterion decided before launch. There are no reliable benchmarks yet, so treat the spend as research, not performance marketing.
If you're deciding how to split budget between paid AI placement and earned AI visibility: and want someone who'll tell you plainly which one your situation actually calls for, that's the work I do. I've spent 4+ years in marketing helping edtech and startup brands grow organically, including work with Masai School that took Instagram from 26K to 117K and LinkedIn from 50K to 160K. See the work and get in touch through the contact form at younusfardeen.com.