Short answer: if you sell a product or service in the €50–€5,000 range, are already comfortable testing new channels with a few hundred euros, and can live with “we’re not entirely sure why this worked” for a quarter, yes — put a small test budget behind ChatGPT Ads before the end of 2026. If you need precise targeting, a proven ROAS model, or you’re outside the United States, wait. The platform is real, it’s spending real ad dollars, but it is not yet a channel you can plan a Q4 budget around with confidence.
We’ve been watching OpenAI’s ad rollout since the invite-only launch in February and the self-serve opening in May, and we’ve now run small tests for two clients who asked us to. Here’s what we’d tell a business owner asking whether this is worth their time this quarter.
What are ChatGPT Ads, exactly?
They’re sponsored cards that appear below an AI-generated answer inside a ChatGPT conversation — clearly labelled “Sponsored,” visually separated from the response itself. A card currently carries a headline (about 50 characters), a short description (about 100 characters), a small image or favicon, and a link. There’s no banner, no video yet, and no in-feed placement the way you’d expect from Meta or TikTok — the format is closer to a very restrained search ad than a social ad.
Matching works by relevance, not demographics. OpenAI reads the context of the conversation — what you’re asking about, what you’ve asked about earlier in the thread, and (for logged-in users who haven’t opted out) some history — and runs a second-price auction against advertisers who’ve described the kind of conversation their product fits. There’s no age, gender, or interest targeting to set up, which is the single biggest workflow change if you’re coming from Meta Ads Manager.
How much do ChatGPT Ads actually cost right now?
Pricing has moved fast in seven months. At the February invite-only launch, CPMs reportedly opened around $60 with a six-figure minimum spend. By May, the minimum was gone and self-serve access opened to any US business. Here’s roughly where the market sits now, based on what advertiser reports and early ad-tech trackers are showing:
| Metric | Approximate range (2026) | For comparison |
|---|---|---|
| CPM | $25–$60 | Meta: roughly $8–$15 |
| CPC, general | $3–$5 | Google Search: often $8–$12+ |
| CPC, SaaS/software | $8–$18 | Google Search SaaS: often $15–$30+ |
| CPC, finance | $15–$25 | Google Search finance: often $30–$60+ |
| Minimum daily spend | None required; ~$25/day is a realistic operational floor | — |
The number that matters most for a small business: click costs are currently running well below equivalent Google Search terms in most categories. That’s a classic early-platform discount — the kind Meta Ads offered in 2012 and TikTok offered in 2020 — and it won’t last once more advertisers pile in.
Who can actually buy ChatGPT Ads today?
This is where a lot of the hype gets ahead of reality. As of now:
- Access is limited to businesses registered in the United States, with self-serve availability expanding to more countries through the rest of the year.
- Ads only show to people on ChatGPT’s free tier and the lower-cost “Go” subscription — Plus, Pro, Business, and Enterprise users see no ads at all, which caps the addressable audience well below “all of ChatGPT.”
- Health, financial services, political, dating, and adult categories are currently excluded from buying ads at all.
- You’ll go through an identity-verification step before you can spend, which takes longer than setting up a Meta or Google account.
If you’re a Finnish or EU-based business, that first point matters most: you can’t reliably buy ChatGPT Ads yet, though it’s clearly on OpenAI’s roadmap. What you can do today is prepare — get your product feed and landing pages in order, since eligibility tends to expand faster than most advertisers expect.
Is it actually worth testing?
We’d split this by business type rather than give a blanket answer.
Good fit right now: US-eligible household and consumer goods brands, local services, travel and entertainment, and digital products or education — the categories OpenAI currently allows — especially if your product benefits from being recommended mid-conversation rather than searched for by name. A meal-kit brand that shows up when someone asks ChatGPT “what should I cook this week that’s quick” is in a much better position than a brand waiting to be typed into a search box.
Weak fit right now: anything in the excluded categories, anyone who needs granular audience targeting to make the math work, and anyone who can’t tolerate thin attribution data. Conversion tracking exists via UTM parameters and a conversions API, but it’s early, and “why did this convert” is harder to answer than on Meta or Google right now.
How we’d structure a first test
If a client asks us to test this, we run it the same way we’d test any new platform — the same discipline we use in our pre-launch checklist for a new ad platform: small budget, one clear hypothesis, and a fixed decision date.
- Set a two-to-four week test window with a daily budget you’d be comfortable losing entirely — $25–$50/day is enough to see a pattern without real financial risk.
- Write two or three context-hint variations that describe the situations where your product fits, rather than keywords. Think “planning a weekend trip with two kids under 6,” not “family travel.”
- Send traffic to a page built for someone arriving mid-research, not mid-purchase. Resource-style landing pages are outperforming hard sales pitches in early advertiser reports — unsurprising, given the conversational context people are in.
- Track everything you can, but treat the numbers as directional. We’d rather see “this drove 40 site visits and 3 leads we can trace” than trust a platform-reported ROAS this early.
We covered how to measure a new channel honestly, without letting a platform’s own dashboard tell the whole story, in our piece on why Facebook ROAS numbers lie — the same scepticism applies here, arguably more so, given how new the attribution tooling is.
Will running ads hurt your chances of being cited organically?
This is the question we get most from clients who’ve already invested in getting cited in AI answers organically. Right now, the honest answer is: nobody outside OpenAI knows for certain, but there’s no evidence of a direct penalty. Sponsored cards and organic AI answers appear to be produced by separate systems — the ad auction doesn’t feed into what ChatGPT chooses to cite or recommend in its actual response. Paying for a placement in February didn’t seem to change whether a brand got mentioned organically in March. Treat the two as separate budgets with separate goals: paid buys you a guaranteed appearance today, organic visibility buys you a mention whenever someone happens to ask the right question, indefinitely, for free.
What this means for your Q4 budget
Our advice for most of the small businesses we work with: don’t move existing Meta or Google budget into ChatGPT Ads yet. Instead, if you’re eligible and curious, carve out a small, separate test budget — money you’ve already mentally written off as “learning cost” — and give it a real month. If it works, you’ll have a two-to-three quarter head start before the CPMs catch up to Google’s, which is exactly the window early Meta and TikTok advertisers got and exactly the window most businesses missed by waiting for “proof” that never arrives before the discount does.
And if you’re not eligible yet because you’re outside the US, don’t ignore this. Read up, get your creative and landing pages ready, and keep an eye on how ChatGPT’s organic search traffic is already behaving compared to Google AI Overviews — the paid rollout tends to follow the same geographic path as the free product did.