ChatGPT Ads: What Actually Launched, and What It Means for Your Business

The short version

ChatGPT sells ads now, and as of May 2026 any US business can buy them without a large minimum commitment. That is the whole headline. Everything else is detail, but the detail is where the decisions live, so here it is in plain language.

What actually happened, and when

• January 16, 2026 — OpenAI publicly confirmed ChatGPT was entering paid advertising.
• February 9, 2026 — Ads began serving to US users on the Free and ChatGPT Go tiers. Plus, Team, and Enterprise subscribers were deliberately excluded to keep paid tiers ad-free.
• May 5, 2026 — The self-serve ChatGPT Ads Manager opened in beta to all US businesses, and the $50,000 minimum spend from the pilot phase was removed entirely.
• May 7, 2026 — OpenAI announced a pilot expansion into the UK, Mexico, Brazil, Japan, and South Korea.

During the pilot, entry required a $50,000 commitment, which in practice meant national brands and the large holding-company agencies. Removing that floor is the part that matters for everyone else: a local multi-location operator can now run a real test on a budget they choose.

How the ads actually work

If you have run Google Ads, the account structure will feel familiar: campaigns contain ad groups, which contain ads. What is genuinely different is the targeting model.

Google Ads is a keyword auction. You bid on strings of text a person types. ChatGPT Ads is built around conversational intent categories. Instead of bidding on "emergency plumber near me," you describe the kind of conversation you want to show up in, and the system matches relevant discussions whether or not the person used your exact phrasing. That is a meaningful shift: people talk to an assistant differently than they type into a search box, usually with more context and fewer keywords.

You can refine with the controls you would expect: geography down to metro level, device type, conversation complexity, time-based bid adjustments, and negative targeting to keep out conversations that waste money. Integration with first-party CRM data for custom audiences has been signalled as a future capability rather than something available today.

The unit itself is modest by design: a short headline, descriptive text, and optional rich media, labeled as sponsored and placed below the relevant conversation. It is not an interstitial and it does not interrupt the answer.

What it costs

There is no minimum spend on the self-serve platform. Bidding is click and engagement based, and early reports describe costs varying widely by intent category and competition — some comparable to mid-funnel Google search, others notably cheaper.

Be careful with any figure you read right now, including the ones in this paragraph. The platform is weeks old at scale. Benchmarks published today are drawn from small samples in an auction with unusually few bidders, and they will move as more advertisers arrive. Anyone quoting you a guaranteed cost per lead on this channel is guessing.

The review process is stricter than you expect

Every campaign is reviewed before it serves. The guidance is explicit that ads must provide genuine value within the conversation, that overly promotional language gets flagged, and that misleading claims result in removal.

This is worth sitting with, because it inverts a habit a lot of advertisers have. The copy that wins in a display auction — urgency, superlatives, pressure — is the copy most likely to be rejected here. Ads that read like a helpful continuation of the conversation are the ones that survive review.

Some categories cannot participate at all: adult content, gambling, weapons, illegal products or services, and misleading financial schemes. Healthcare, pharmaceuticals, and financial services face stricter review and may face limitations, which is worth knowing before you plan a budget around it.

Why anyone should care right now

Two reasons, and neither is that it will replace Google.

First, the behavior is real. People increasingly ask an assistant for a recommendation instead of scanning a page of links, and an assistant returns a short answer rather than ten options. Whether you get named in that answer, earned or paid, is becoming a real distribution question.

Second, early auctions are cheap and uncrowded, and the learning compounds. Every advertiser who understood Google Ads in 2004 or Facebook in 2012 had an advantage for years afterwards, not because the channel stayed cheap, but because they understood it before their competitors did.

What we would actually do

Keep your working channels funded. Carve out a contained test budget, not a reallocation of something that is already producing. Build a small number of tightly defined intent categories rather than a broad net. Write ads that would be genuinely useful if a person read them mid-conversation. Instrument it properly, because a conversational channel disappears inside last-click reporting. Then judge it honestly after a real sample, and be willing to stop.

And pair it with the earned side. Paid placement and being recommended organically are different mechanisms, and the businesses that do both will be much harder to displace than the ones renting attention alone.