How to buy ads in ChatGPT: self-serve, the DSP route and how to prepare
The short answer
As of mid-2026 you can, if you are a US business, log into OpenAI’s Ads Manager and buy ChatGPT placement yourself: the self-serve product opened in beta in May, with an approval period, restricted categories and placements limited to certain user tiers. If you are outside that programme, buying for multiple clients, or building ChatGPT into a wider media plan, the practical route is a demand-side platform with early access to AI-assistant inventory. (For what the formats look like in the product, see our early reports piece; for the placement mechanics, see how advertising inside ChatGPT works.)
This guide covers the buying half: the routes, what a flight is actually like, and what to do before committing budget.
The three routes to placement
1. Self-serve, through OpenAI’s Ads Manager. The beta gives US advertisers direct campaign creation on conversational-context targeting. The honest caveats: an approval queue rather than instant launch, category restrictions that bite hardest for regulated and political-adjacent advertisers, placement limited to parts of the user base, and controls that are young. For a single brand wanting to learn the surface first-hand, this is now the obvious starting point.
2. Through a DSP with early access. Demand-side platforms have been securing early access to AI-assistant inventory alongside their conventional programmatic supply. This route earns its place when ChatGPT is one line on a bigger plan: agencies and firms buying for clients, advertisers outside self-serve coverage, and campaigns that want AI-assistant placement planned, bought and measured in the same frame as display, video, CTV and audio. This is the route Morris McLane uses for client programmes.
3. The organic route. Not a purchase at all, but the same surface: ChatGPT cites sources in its answers, and being the cited source on your category’s buyer questions is available to any organisation willing to do the work. It also makes paid placement worth more, because an ad next to an answer that already cites you compounds; an ad next to an answer that contradicts you fights itself.
What buying is actually like
Early AI-assistant buying feels less like mature search and more like early programmatic: promising, coarse, and changing monthly.
- Context, not keywords. Placement rides on the meaning of the conversation. Bidding exists, but the levers you know from search — exact-match terms, bid stacks, quality scores — do not map across cleanly; relevance and creative do more of the work.
- Coarse controls. Expect broader targeting than you are used to, and expect the options to change between flights as the product matures.
- Young measurement. Reporting exists but gives you less to optimise against than a decade-old search platform. Sensible early flights are structured as tests: a hypothesis, a modest budget, a defined read-out.
- Policy in motion. What categories may advertise, and with what claims, is still settling. Regulated and political-adjacent advertisers should expect extra review and build in time for it.
The honest framing for a 2026 budget: an early flight buys learning and first-mover familiarity more than proven performance. That is a legitimate purchase — the advertisers who understood paid search in 2003 and paid social in 2009 banked years of advantage — but it should be priced as R&D, not as a proven channel.
What to do before the first flight
- Baseline your organic presence. On a fixed set of the buyer questions that matter to you, what do ChatGPT, Gemini and Perplexity answer, and whom do they cite? Without this, you cannot tell whether paid activity changed anything. (How to measure it.)
- Fix the machine-readable layer. The pages assistants cite — direct answers, definitions, comparisons, FAQs — are the same pages that make paid placement land. This work is available today at predictable cost through an AI search visibility programme.
- Choose the route deliberately. A US brand testing the surface can start self-serve this week. A firm buying for clients, or an advertiser who needs ChatGPT inside a cross-channel plan, should line up DSP access before the moment it is needed.
Where Morris McLane fits
Morris McLane buys AI-assistant placement through DSP early access, run commercially as a managed ChatGPT advertising service within its AI advertising capability, alongside conventional programmatic and CTV — and runs the organic layer that makes those impressions worth more. For firms and their clients we handle both halves as one programme: measured organic visibility as the foundation, paid placement as the complement, one report covering both. If you want a test flight designed rather than improvised, start here.
Frequently asked questions
Can anyone buy ads in ChatGPT right now?
US businesses can, with caveats. OpenAI opened a self-serve Ads Manager in beta to US advertisers in May 2026: campaigns are created directly, with an approval period, category restrictions, and placements limited to certain user tiers. Outside the US, or for advertisers who need ChatGPT placement bought alongside conventional programmatic channels, access runs through demand-side platforms holding early-access inventory. Either way, this is an early market: expect the rules to change between flights.
What are the routes to buying ChatGPT ad placement?
Three in practice. OpenAI's self-serve Ads Manager, in beta for US advertisers, is the direct route: create the campaign, pass approval, and buy on conversational-context targeting. The DSP route reaches the same emerging inventory through demand-side platforms with early access, bought and measured alongside display, video and CTV. And the organic route costs media nothing: making your pages the sources ChatGPT cites, which pays off whether or not you ever buy an ad.
How is advertising in ChatGPT different from search ads?
The auction logic differs. Placement rides on the meaning of a live conversation rather than a bid on a typed keyword, so relevance to the moment and creative quality carry more of the load than bid mechanics. Targeting controls are coarser than mature search platforms, and measurement is younger: reporting exists, but gives you less to optimise against. Sensible early flights are structured as tests with modest budgets and explicit read-outs.
When does the DSP route beat self-serve?
When ChatGPT is one line on a bigger plan. Agencies and firms buying for clients, advertisers outside the self-serve programme's coverage, and campaigns that need AI-assistant placement alongside programmatic display, video, CTV and audio all benefit from one buying platform, one measurement frame and one billing relationship. Self-serve suits a single US advertiser testing the surface directly; the DSP route suits portfolios and cross-channel programmes.
What should an advertiser do before spending on ChatGPT ads?
Two things. First, baseline your organic presence: know what ChatGPT and the other assistants already say and cite about your category on a fixed set of buyer questions, so paid activity has something to be measured against. Second, fix your machine-readable layer, the pages assistants quote, because paid placement alongside a wrong or absent organic presence wastes the impression. Both are useful regardless of how the ad products evolve.
Should budget go to ChatGPT ads or to AI search visibility?
They answer different problems. Paid placement buys presence in conversations immediately and ends when the spend ends. Organic AI visibility, being the source assistants cite, compounds and persists, but takes longer to build. Most advertisers should fund the organic layer first because it is available at predictable cost, then run capped, well-instrumented paid tests as the complement, managed as one programme rather than two silos.
