For about a year, “advertising on ChatGPT” was a thing you read about but couldn’t actually do. The pilot had a $200,000 monthly minimum. Then it dropped to $50,000. Then, on May 5, 2026, OpenAI opened the Ads Manager to every US business with no minimum spend at all — and added CPC bidding on top of the old CPM-only model.
That’s the part worth paying attention to. A $0 minimum and pay-per-click bidding means a small advertiser can now buy ChatGPT inventory the same way they buy Google or Meta. You can fund a campaign with $200 and a credit card. The question stopped being “can I?” and became “should I, and how much?”
I’ve spent the last few weeks reading the docs, the early case reports, and the pricing math. Here’s my honest read on where the first dollar should go.
What actually changed on May 5
The mechanics matter, so let’s be precise about them.
You sign up at ads.openai.com if you’re a US-based business. There are two objectives. Clicks buys on CPC — OpenAI recommends a starting bid of $3–$5 per click, and bids under roughly $3 tend to just not deliver impressions. Reach buys on CPM, with a default max bid around $60 per thousand impressions. The auction is a relevance-weighted second-price model, so your bid isn’t the whole story; ad quality and conversational fit pull weight too.
Measurement is more grown-up than I expected for a v1. You get a Conversions API and a pixel for tracking purchases, sign-ups, and leads, plus UTM support and campaign-level aggregated reporting. What you don’t get is individual user data — reporting stays aggregated, which is consistent with how OpenAI has talked about user trust.
The targeting model is the genuinely new thing. It’s not keyword matching. OpenAI calls it conversational targeting: you supply context hints describing the topics or conversations you want to show up next to, and the system places your ad against relevant chats. If you’ve spent years building exact-match and negative-keyword lists in Google Ads, throw that muscle memory out. This is closer to describing your ideal customer’s problem than bidding on a string.
The one constraint that shapes everything
Ads live below the conversation, in a clearly labeled, visually distinct box. They never appear inside the model’s answer.
This is OpenAI drawing a hard line, and it’s the most important design decision in the whole product. The ad unit is small and rigid — advertiser name, favicon, a 16-character title, 32 characters of copy, an image, and a link. You’re writing for a fortune cookie, not a landing page.
Why does this matter for your budget? Because it changes click intent. On Google Search, someone types a query with commercial intent and your ad sits right where they’re looking for an answer to buy. In ChatGPT, the user came to think through something — compare options, draft a plan, understand a topic — and your ad shows up adjacent to that thinking. The intent is real but earlier in the funnel. Expect more “huh, interesting” clicks and fewer “take my money right now” clicks, at least until the formats mature.
That’s not a knock. Early-funnel, high-consideration research is exactly where a lot of B2B and considered-purchase buying actually starts. It just means you should measure ChatGPT ads on assisted conversions and pipeline, not last-click ROAS, or you’ll talk yourself out of something that’s working.
ChatGPT vs Gemini vs Copilot, honestly
Right now this is less a three-way race than one live platform and two that are still loading.
ChatGPT Ads is the only one a small advertiser can self-serve into today with CPC bidding and no minimum. Reach is enormous — ChatGPT is the front door to AI for most people — and the inventory is brand new, which usually means it’s underpriced. Nobody has a five-year-old, fully-optimized account beating you in the auction. The downside: targeting is young, the creative unit is tiny, and there’s no track record to benchmark against.
Gemini Ads is the one I’d watch most closely. Google has confirmed ads are coming to Gemini in 2026, but they aren’t broadly live inside the assistant yet. What is live is ads in AI Overviews and AI Mode across Google Search — reportedly weaving into around a quarter of AI-generated results. When Gemini proper gets ads, Google brings the thing nobody else has: two decades of intent data, an ad platform your team already knows, and the ability to slot AI inventory into existing campaigns. If you only have time to learn one new surface this year, betting on the Google one is the safe institutional choice. You just can’t spend there yet.
Microsoft Copilot / Microsoft Advertising is the quiet third option. Microsoft already runs an ad network and has been folding AI chat surfaces into it, so for advertisers already buying Microsoft Search inventory, Copilot placements are more of an extension than a new platform to learn. The reach is smaller — Copilot has roughly 15 million paid M365 seats but a much lower share of people who actively choose it over the assistant they were handed — so treat it as a complement, not a primary channel.
And then there’s the platform that said no.
Perplexity walked away — that’s a signal, not a footnote
In February 2026, Perplexity killed its ad program entirely. Not paused — killed, with executives saying they have no plans to bring it back. Their argument: the moment ads appear in answers, users start second-guessing whether the response is honest or quietly bought. One exec’s framing was that ads “butcher trust” in AI.
Anthropic’s Claude has stayed ad-free too. So the industry has split into two camps running opposite experiments. OpenAI and Google are betting that careful, labeled, below-the-answer ads can coexist with trust. Perplexity and Anthropic are betting they can’t, and that an ad-free product is the moat.
For you as an advertiser, the takeaway is twofold. First, the available ad surfaces (ChatGPT, eventually Gemini, Copilot) are precisely the ones that decided the trust hit is manageable — which is also why every one of them keeps ads out of the answer itself. Second, don’t assume this inventory is permanent or unlimited. If user trust metrics wobble, these programs can shrink as fast as Perplexity’s vanished. Early-mover advantage here comes with early-mover risk.
The first-budget playbook
So who should actually open an account this month?
Test it now if you sell something people research before buying — B2B SaaS, professional services, education, high-consideration ecommerce. That’s where ChatGPT’s “user is thinking it through” context is a feature, not a bug. The early-mover-cheap-inventory argument is strongest for you: low competition, novel placement, and a chance to learn the conversational-targeting model before your competitors do.
Wait if you’re in a regulated category. OpenAI itself suggests healthcare and finance advertisers hold off until guidelines firm up, and that’s good advice — the last thing you want is a compliance headache over a 32-character ad. Local services with tight geographic targeting needs should also wait; the targeting isn’t precise enough yet to avoid wasting impressions on the wrong city.
If you do test, here’s the sane setup:
- Start with a Clicks/CPC campaign, not Reach. You want to measure response, and CPC ties spend to action. Save CPM for when you have a brand-awareness reason.
- Bid in the $3–$5 range from the start. Going cheap to “test the waters” mostly buys you no impressions, which teaches you nothing.
- Wire up the Conversions API or pixel before you spend a dollar. Conversational ads will look bad on last-click and fine on assisted conversions — if you can’t see the assist, you’ll kill a working campaign.
- Budget like it’s an experiment: a few hundred to low thousands of dollars over two to four weeks, enough to clear noise. Don’t reallocate your Google budget; fund this from a learning line.
- Write the copy as a hook to one specific problem, not a value-prop dump. You have 32 characters. Pick the single thing your ideal reader is mid-conversation about.
DIY or go through a partner
If you’re a small team, just use the self-serve Ads Manager. It’s built for exactly that, and a partner won’t help much on a $1,000 test.
If you’re already working with one of the big agency holding companies — Dentsu, Omnicom, Publicis, WPP — they all have early access and can fold ChatGPT into a broader media plan. And if your stack runs on ad-tech, OpenAI has integrations with Adobe, Criteo, Kargo, Pacvue, and StackAdapt, so you can manage and measure ChatGPT alongside your other channels rather than in a silo. For a mid-market advertiser already spending through one of those tools, that integration is the strongest reason to move sooner rather than later — the operational cost of adding the channel is low.
So where does the first dollar go?
Here’s how I’d call it by business type, as of June 2026:
- Local service business — wait. Targeting precision isn’t there, and Google still owns local intent.
- Ecommerce, considered purchase — test a small CPC campaign. Measure assisted conversions, not ROAS.
- B2B SaaS — test now, properly. This is the best fit for the surface, and the inventory won’t stay this cheap.
- Lead gen — test, but instrument conversions obsessively before spending. The funnel is early; prove the assist or you’ll misjudge it.
The honest summary is that ChatGPT Ads in mid-2026 is a real, buyable channel with genuinely cheap early inventory and a measurement story that’s better than I expected — wrapped around a creative unit that’s tiny and a targeting model nobody has mastered yet. That’s a good profile for a controlled experiment and a bad one for shifting serious budget.
Gemini is the platform that’ll probably matter most once its ads go live, so the smartest move might be to learn the conversational-ad muscle on ChatGPT now, cheaply, so you’re fluent when Google’s version shows up with the intent data to make it convert.
If you’ve got a spare few hundred dollars and a product people research before they buy, open an account at ads.openai.com, run one tight CPC test, and see what a fortune-cookie-sized ad next to a billion conversations actually does for you.
Sources: WebFX — ChatGPT Ads Manager, WebFX — How Much Do ChatGPT Ads Cost, Search Engine Land — Perplexity stops testing advertising, Campaign US — Perplexity pulls the plug on ads, DigitalApplied — AI Search Advertising: ChatGPT vs Google vs Perplexity.