For years, Sam Altman called advertising in ChatGPT a last resort. That changed on February 9, 2026, when OpenAI began serving sponsored placements to free-tier users in the U.S. What started as a $200,000-minimum enterprise pilot has, in a matter of months, become a self-serve channel any business can test — with no minimum spend at all. If you buy paid search or social, this is the newest surface worth understanding, and the barrier to entry has never been lower.

Moving fast ChatGPT advertising is still officially a beta. Formats, character limits, pricing, and eligible markets are changing month to month. Treat every figure here as a current snapshot and confirm the specifics in OpenAI's Ads Manager or with your platform partner before you launch.

What ChatGPT advertising actually is

ChatGPT ads are sponsored cards that appear below a relevant AI response — never inside the answer itself. OpenAI's design principle is strict on this point: ads are clearly labeled "Sponsored," visually separated from the assistant's text, and they do not influence how ChatGPT answers a question. Think of it less like a Google search ad and more like a knowledgeable friend slipping in a relevant recommendation after they've already helped you.

There is currently one ad format, sometimes called the "chat card" or Sponsored Recommendation Card. It contains six parts: your advertiser name, a favicon or logo, a headline, a short description, a single image, and a destination URL. There is no second headline, no extra description lines, and no creative carousel — one card, a handful of fields, done. That constraint is the whole point: the unit is built to feel native to a conversation.

Who actually sees the ads

This is the single most important thing to understand before you size the opportunity, because the reachable audience is much narrower than ChatGPT's headline user numbers suggest.

  • Free and Go tiers only. Ads show to logged-in adult users on the Free and ChatGPT Go plans. Plus, Pro, Team, Business, Enterprise, and Education subscribers never see ads.
  • Adults only. OpenAI doesn't serve ads to users who indicate they're under 18, or who the system predicts are under 18.
  • Opt-outs exist. Eligible users can choose an "Ads-Free" option on the Free tier in exchange for lower usage limits, which trims inventory further.
  • Markets are expanding. The pilot launched in the U.S. and has since expanded; through partners like StackAdapt, targeting is currently available in the United States, Canada, Australia, New Zealand, the UK, Japan, and South Korea.

During the early pilot, an OpenAI spokesperson noted fewer than 20% of eligible users were being shown ads on a given day. In other words: real reach today is a fraction of the "900 million weekly users" figure you'll see quoted. Plan accordingly.

Ad sizes and creative specifications

The creative brief is refreshingly short. Because the card renders as a small thumbnail next to text, simplicity beats polish. Here are the working specs as of mid-2026 — sourced from OpenAI's help center and StackAdapt's advertiser documentation.

The image

  • Shape: square, 1:1 aspect ratio.
  • Minimum size: 256×256 pixels — but that's a floor, not a target. Export at 512×512 or 1024×1024 so it stays crisp on high-resolution phones, and let ChatGPT scale it down.
  • File formats: JPG, PNG, or WEBP. Keep the file under ~1MB for fast loading.
  • What works: a single product on a clean background, a bold logo mark, or simple iconography with high contrast. The ChatGPT interface has a light background, so bold, saturated visuals with dark outlines cut through.
  • What fails: text baked into the image (illegible at thumbnail scale), busy lifestyle scenes, and low-contrast washes. Let the headline and description carry the message — the image should communicate visually, not textually.

The copy

  • Headline: short — StackAdapt's advertiser guidance lists roughly 30 characters. Some sources report other limits as OpenAI tunes the format, so write tight and confirm the current cap in the platform.
  • Description: roughly 60 characters of body copy.
  • Favicon & advertiser name: a clean, square version of your logo mark, plus your brand name.
  • Destination URL: add UTMs for tracking. Critically, your landing page must be crawlable by OpenAI's OAI-AdsBot and OAI-SearchBot user agents — if your robots.txt, CDN, or firewall blocks them, the ad gets approved but silently fails to serve.
The copy that wins here reads like a helpful recommendation, not a billboard. If it sounds like a sales pitch, it gets ignored.

How targeting works

Forget keyword bidding. ChatGPT advertising is built on contextual matching — OpenAI reads the topic of the live conversation and serves the ad most relevant to what the user is actually researching or comparing. At the ad-group level, you provide "context hints": plain-language descriptions of the questions, needs, and situations your customers bring to ChatGPT. These guide matching but do not behave like exact-match search keywords, and they don't guarantee placement in any specific conversation.

The practical implication: keep each ad group focused on a single product category or use case, and describe scenarios rather than listing standalone keywords. Someone asking how to plan a festival outfit is a better match for "festival-ready styles" than for a generic apparel ad. This is much closer to the discipline behind Generative Engine Optimization than to classic PPC — you're aligning with intent expressed in natural language.

Pricing and minimum spends

This is where the story has moved fastest. Here's the timeline that matters:

  • February 2026 (launch): $60 CPM with a $200,000–$250,000 minimum commitment. Enterprise only.
  • April 2026: minimum cut to $50,000; observed CPMs dropped toward ~$25.
  • May 5, 2026: OpenAI opened its self-serve Ads Manager to all U.S. businesses, added cost-per-click bidding alongside CPM, launched a Conversions API — and removed the platform-level minimum spend entirely.

So where does that leave costs today?

  • CPM: roughly $25 observed, down from the $60 launch default (still varies by topic cluster).
  • CPC: now available, with bid floors reported around $3–$5 by category.
  • Minimum spend (direct): none. OpenAI's self-serve Ads Manager costs nothing to start.

A sensible test budget many practitioners suggest is reallocating ~5% of an existing search budget, or starting around $50/day, and giving the campaign 2–4 weeks to stabilize before judging it. One honest caveat: reporting is aggregated by design (closer to connected-TV measurement than Google Ads), and early click-through rates have been modest. Go in to learn, not to scale on day one.

How to sign up

There are two routes into ChatGPT advertising, and the right one depends on whether ChatGPT is your only channel or one of many.

Route 1 — OpenAI directly (self-serve)

Register your interest and access the self-serve platform through openai.com/advertisers, which routes to OpenAI's Ads Manager. In the Ads Manager Beta you'll build campaigns (objective, budget, dates, target countries), then ad groups organized by theme or intent, then the creative itself. Choose a "Views" objective for CPM reach or a "Clicks" objective for CPC traffic. Access is still gated for some accounts, and approval for verified businesses has been running about 1–2 weeks. This route is best if you only want to spend inside ChatGPT and want to own the relationship directly.

Route 2 — a programmatic platform partner

If ChatGPT is one line in a broader media plan, running it through a demand-side platform lets you buy it alongside display, CTV, video, and more — with a single audience view, unified reporting, and hands-on support while the channel's conventions are still being written. That's the case for most advertisers with meaningful budgets, who are never executing in a single silo.

The programmatic platforms offering ChatGPT ads

When OpenAI opened self-serve access, it did so alongside a slate of technology partners. These are the ones you can work through today:

  • StackAdapt — no minimum spend. The independent DSP started its ChatGPT pilot with a $50,000 minimum in April, then dropped it to $0 in mid-May and opened the channel to roughly all of its ~1,000 advertisers. Its CRO confirmed the no-minimum stance publicly. StackAdapt's pitch is the multi-channel view: run ChatGPT alongside display, CTV, video, DOOH, and email, and see performance in one place. If you want the lowest-friction way to fold ChatGPT into an existing programmatic plan, this is it. Their how-to guide is a solid primer.
  • Criteo — retail & commerce focus. OpenAI's first ad-tech partner cut its ChatGPT minimum from $50,000 to $10,000 in June and lets retail brands pipe existing product feeds straight into OpenAI's ad system — so you can launch without hand-building individual ads. Criteo has reported 1,000+ brands live on ChatGPT through its platform. Best if you have a live product catalog and a commerce use case.
  • Kargo. Joined as a technology partner in the May self-serve expansion.
  • Adobe Advertising. Part of the partner slate, useful if your stack already lives in Adobe's ecosystem.
  • Pacvue. A commerce-advertising platform among the launch partners, oriented toward retail media buyers.

A useful way to choose: Criteo is a retail-and-commerce play, StackAdapt is a multi-channel buying play, and going direct through OpenAI makes sense only if ChatGPT is genuinely your sole channel. The ad-tech scramble to sign advertisers is itself a signal — everyone wants to lock in relationships before the channel matures.

The benefits — and the honest caveats

Why it's worth testing now

  • High-intent moments. Someone asking ChatGPT to compare options or recommend a product is deeper into a decision than someone passively scrolling a feed. Criteo's early (self-reported, unverified) data showed AI-referred traffic converting at roughly 1.5–2× the rate of other referral channels in categories like consumer electronics and home.
  • A brand-new surface with low competition. The advertiser base is still small. Early movers get to learn the creative and bidding conventions before costs and crowding rise.
  • No cost of entry. With OpenAI's minimum gone and StackAdapt at $0, you can run a genuine experiment for the price of the media itself.
  • Native, non-intrusive placement. Ads sit below the answer and read like a recommendation, which tends to protect brand perception.

What to plan around

  • Thin, aggregated reporting. You get views, clicks, spend, CTR, CPC/CPM, and conversions — but no individual conversation data. Set attribution expectations internally before launch.
  • Narrow, opt-out-able inventory. Free/Go adults only, minus the ads-free crowd.
  • Modest CTRs and fast creative fatigue. The same users see you across multiple daily sessions; refresh creative every 7–14 days.
  • Regulated categories are limited. The format best suits retail, e-commerce, CPG, travel, education, and B2B. Health, mental health, politics, and other sensitive topics are excluded.

A practical way to start

  • Map 30–50 real prompts your customers would type into ChatGPT, and turn them into context hints and tightly themed ad groups.
  • Build answer-first landing pages that continue the conversation the ad started — and make sure OpenAI's crawlers aren't blocked.
  • Write conversational copy in multiple headline/description variations so the system has options to match against.
  • Pick your route: OpenAI direct if ChatGPT is your only play, StackAdapt (no minimum) or Criteo (feeds/commerce) if it's part of a wider plan.
  • Fund a real test, give it 2–4 weeks, and judge it on incremental learning, not day-one ROAS.

The takeaway

ChatGPT advertising went from a $200,000 enterprise experiment to a no-minimum, self-serve channel in under four months. The reachable audience is narrower and the reporting thinner than the hype implies, but the intent is genuine and the cost of a disciplined test is now effectively zero. The smart move isn't to bet the budget — it's to run a cheap, well-instrumented experiment now, learn the creative and targeting conventions while competition is thin, and be ready to scale the moment audience matching and richer formats flip on. If you'd like help mapping the prompts, building the landing pages, and choosing between going direct and running through a platform like StackAdapt, that's exactly the kind of work we do.