ChatGPT Ads: How They Work and Who Should Run Them

For years, “advertising in ChatGPT” was a rumour OpenAI’s own leadership talked down. Then it became one of the fastest ad-platform rollouts in software history. OpenAI began testing ads in ChatGPT on February 9, 2026 for logged-in adult users on its Free and Go tiers in the US, and within roughly three months had moved from an invite-only pilot with six-figure spend commitments to an open self-serve Ads Manager with no minimum spend.

We manage AI-platform advertising as a dedicated service, so this guide is written from the operator’s side of the dashboard: what ChatGPT ads actually are, how the targeting works (it is genuinely different from search), what they cost now versus the pilot days, who should run them, who should wait, and how paid placement fits alongside the organic AI visibility work we covered in our GEO guide. One honesty note up front, because this platform changes monthly: every date and mechanic below was verified at publish time, and we review this post quarterly.

What a ChatGPT Ad Actually Is

Strip away the hype and the core unit is modest and deliberate: a clearly labeled “Sponsored” card that appears below the assistant’s answer when a conversation shows commercial intent. The card carries an advertiser name, headline, short description, image, and destination link, visually separated from the response. Two design rules define the platform, and OpenAI has repeated them at every stage: the ad never changes what the assistant says, and advertisers never see the content of anyone’s conversation. Since launch the surface has grown, notably with product feed ads for eCommerce advertisers added in May 2026, letting stores surface catalogue items against shopping-intent conversations, but the sponsored-card-below-the-answer remains the anchor format.

Who sees them matters as much as what they are: ads show only to logged-in adults on the Free and Go tiers. Every paid tier (Plus, Pro, Business, Enterprise, Education) is ad-free. You are buying the largest slice of ChatGPT’s audience, its free users, at the moment they are asking for exactly the kind of help your business sells.

Why This Channel Exists: The Audience Math

The inventory is enormous and the intent is unusual. ChatGPT reached roughly 900 million weekly active users by early 2026, per OpenAI’s own figures, with Semrush ranking it among the most visited sites on the web. Independent tracking through mid-2026 found sponsored cards appearing in as much as half of eligible US free-tier replies during peak periods, a volatile number, but a signal of how much inventory OpenAI is comfortable serving. And the traffic behaves differently: Semrush’s 2026 cross-industry research measured AI-driven visitors converting at about 4.4 times the rate of standard organic visitors, which matches the logic of the surface, since people ask assistants for help when they are actively deciding. An ad that arrives inside a decision conversation is closer to a referral than an interruption, when it is matched well, and matching is the platform’s entire bet.

How Targeting Works: No Keywords, No Audiences

This is the part that rewires media-buyer instincts. There is no keyword auction and no prebuilt audience segments. Ads are matched contextually to the live conversation, informed by the topic of the current thread, the user’s past chats and memory, and their past ad interactions, with OpenAI explicit that there is no third-party cross-web tracking and no sale of chat data to advertisers. Practically, that means you target concepts and intents, not queries: an accounting firm is matched into conversations about tax filing stress; a mattress brand into sleep-problem threads; a CRM into “how do I keep track of my customers” conversations. Your levers are the offer, the creative, the landing experience, and the intent categories you campaign against, which makes this feel less like Google Ads and more like being the well-timed suggestion inside a consultation. It also means conversational intent mapping, understanding how your customers phrase their problems to an assistant, is the new keyword research, and it is where most of the strategic work in our campaigns happens.

What ChatGPT Ads Cost

The pricing story is a compressed history of the whole rollout:

For small and mid-size advertisers, the May shift was the door opening: no minimums, self-serve, and auction dynamics still young. That last part is the honest early-mover argument: competition on this inventory is thinner than any mature channel, quality signals and campaign history are being established now, and the advertisers building them will hold a structural cost advantage as the auction crowds. The counterweight is volatility: pricing, formats, and delivery rates have all moved repeatedly in the platform’s first six months, so budgets here should be experimental capital with weekly review, not set-and-forget spend.

A note for Canadian advertisers: the rollout began US-only and has expanded market by market, with availability differing between self-serve and managed-partner routes. Check current country eligibility directly in Ads Manager before planning; we track availability for clients as part of our ChatGPT Ads management, and Canadian brands with US customers can, of course, already buy the US audience.

Who Should Run ChatGPT Ads

The strongest fits we see, in order:

1. eCommerce brands with clean product feeds. The product-feed format plus shopping-intent conversations is the most natural match on the platform, and stores get the fastest feedback loops.
2. Businesses in consideration-heavy categories. Software, financial services (where permitted), education, home services, B2B tools: anywhere customers “talk through” a decision, the conversation is the targeting.
3. Brands already investing in AI visibility. Paid cards and organic citations compound: a business the assistant already mentions organically, per the playbook in our ChatGPT citation guide, gets a credibility echo when its sponsored card appears, while an unknown name leans on the card alone.
4. Early movers in competitive verticals. If your Google Ads auction is brutal, this is the rare adjacent channel where attention is cheap relative to intent, for now.

Who should wait: advertisers in restricted categories (OpenAI blocks health, mental health, and political advertising at the platform level, with the usual regulated-category rules beyond); businesses whose measurement cannot isolate a new channel cleanly; anyone whose landing pages are not ready for conversational traffic (these visitors arrive mid-decision and expect the page to continue the conversation, not restart it); and businesses that have not yet fixed their organic AI presence, because the free layer of this channel, being cited in the answers themselves, costs content instead of media budget and should not be skipped to buy placement below answers that recommend your competitor.

Running It Well: What We Have Learned Operating the Channel

Practices from managing AI-platform campaigns that transfer to anyone starting now. Map conversational intents before touching the dashboard: list the twenty problem-statements your customers would actually type to an assistant, and build campaigns against those intents, not your product taxonomy. Write cards like advice, not billboards: the winning creative on this surface reads as a natural next step to the conversation above it (“Compare Canadian SEO pricing” outperforms brand slogans). Land the click in continuity: the page should answer the same conversation the user was having, which is why we often pair campaigns with dedicated landing pages. Measure like a skeptic: UTM everything, watch assisted conversions and branded-search lift alongside last-click, and judge at 30-day windows because delivery volatility makes daily reads meaningless. And cap the experiment honestly: a defined monthly budget, a written hypothesis, and a kill-or-scale review beats enthusiasm every time, the same discipline we apply to every paid channel we run.

Paid and Organic: The Two-Track AI Strategy

The strategic frame we give clients: AI surfaces now have an earned lane and a paid lane, exactly like search did in 2005, and the businesses that win treat them as one program. The earned lane is GEO: entity consistency, evidence-dense content, and third-party corroboration that gets you cited inside answers, work that compounds and cannot be switched off by a budget cut. The paid lane is the sponsored card: instant presence against intents you have not earned yet, a testing ground for which conversations convert, and insurance during the months the organic footprint builds. The data flows both ways, since the intents that convert as ads are the queries worth building citation content for, and the pages that earn citations make the ads they sit near more credible. Running one lane while ignoring the other is leaving half the surface to competitors, and this early, most Canadian categories have no competitor running both.