How to Advertise on ChatGPT: Self-Serve Ads Manager Guide (2026)

Seven months into the ChatGPT ads program, the buying model has changed. The pilot was enterprise-only, with $200K direct-sales minimums and no dashboard; Ads Manager now lets any eligible business launch a campaign from a $25 daily budget. The pilot’s advertiser pool widened and rotated more than once. cloro’s May 26 measurement saw the pool jump from the narrow 16-brand set of April-May into a broad consumer-commerce mix; the July 3 measurement shows that mix has rotated again — e.l.f., Pottery Barn, IKEA, Ralph Lauren, Macy’s are gone, replaced by a heavy Japanese layer plus a widened travel and telco/connectivity buyer set.

OpenAI launched paid ads on February 9, 2026 through direct enterprise sales. The launch minimum was $200,000; some categories have since seen the minimum cut to $50,000. First-wave advertisers included Target, Ford, Adobe, Mrs. Meyer’s, Expedia, Best Buy, and Enterprise. The July 2026 top-20 (by placement volume) is: Carsensor, Booking.com, Lovable, amaysim/Alinta, Top10.com, Zoho, Coterie Baby, Onamae.com, Flip TV Australia, Monday.com, Inference R&D, Samsung, Expedia, Pivotree, BISSELL, iHerb, Luxury Escapes, DSW, ZoomInfo, Best Buy. For context, our penetration study measures ads in 51% of US responses, 54% Canadian, 50% Australian, and 19% Japanese responses over the 7 days ending 2026-07-03.

This post covers how to buy a placement in Ads Manager, what bids and formats look like, and what the pilot data tells a new advertiser. The rollout context is in our ChatGPT ads status post; the on-platform penetration rate is in the penetration study.

How to advertise on ChatGPT with the self-serve Ads Manager

Most of this guide describes the direct-sales pilot that ran through July 2026. That is no longer the only way in. OpenAI now lets any eligible business sign up at ads.openai.com and build campaigns in Ads Manager, and it keeps sales support for larger advertisers. Search Engine Land’s step-by-step guide (August 10, 2026) puts the minimum at a $25 daily budget, against the $200,000 commitments of the pilot.

The setup, per that guide and OpenAI’s Help Center:

1. Confirm eligibility. Ads show only to Free and Go users; Plus, Pro, Business, Enterprise and Edu accounts see none. Search Engine Land lists the US, Australia, Canada, Japan, New Zealand, South Korea and the UK as targetable, and restricted categories include alcohol, gambling and financial services.
2. Create the account. Have a legal business name, tax ID, card and favicon ready. Agencies are added as users on the advertiser’s account rather than creating it.
3. Structure the campaign. Campaign, ad group, ad. Pick Clicks (CPC), Reach (CPM) or Conversions, then set location, budget and a max bid.
4. Write context hints. There are no keywords. Each ad group takes a free-text hint describing the conversations where the ad fits.
5. Build the ad. Headline up to 50 characters, description up to 100, a square image of at least 256 x 256 px, and a destination URL.
6. Install the OpenAI pixel and add UTM parameters before launch, so conversions show in Ads Manager and in GA4.

Best practices for advertising on ChatGPT

Search Engine Land sees average CPCs of $2 to $5, and a max CPC below $3 triggers a warning that the ad may not deliver, so start at $3 or above. In its test, a keyword-style context hint won more impressions at a lower CPC than a list of full questions, so test both formats in separate ad groups. LYFE Marketing adds that the auction is second-price and relevance-weighted, so a landing page that answers the comparison the user is making lowers cost more than a higher bid does.

Who could advertise during the direct-sales pilot

The current advertiser pool sits at the intersection of four conditions:

1. A real budget commitment. Originally a $200,000 minimum, now reportedly cut to $50,000 for some categories as OpenAI works to widen the buyer pool. Most placements we see in the wild are from advertisers spending well above either floor.
2. A direct relationship with OpenAI’s enterprise sales team. No self-serve sign-up existed. You contacted OpenAI through business channels and negotiated.
3. Category fit. OpenAI is now actively recruiting across consumer commerce, mid-market SaaS, affiliate aggregators, travel, telco/connectivity, and (new since May) Japanese consumer brands. Less interest in commoditized gambling, supplements, or low-margin verticals. The first-wave brand mix (Target, Ford, Adobe, Expedia, Best Buy, Enterprise, Mrs. Meyer’s) signals the original US profile; the May 26 expansion (e.l.f., Pottery Barn, Cursor, Jotform, Top10.com, BestMoney) has largely rotated out; the current July 3 profile (Carsensor, Booking.com, Lovable, Zoho, Coterie Baby, Onamae.com, Monday.com, Samsung, Expedia, BISSELL, iHerb, DSW, ZoomInfo, Best Buy) is where the pool is now.
4. Often: an existing OpenAI API or enterprise contract. Not strictly required, but it shortens the conversation. The ads team uses API spend as a signal of advertiser fit.

The buyer pool is widening, but the entry friction is still high. Direct sales remains the primary path. The programmatic onramp opened in May: StackAdapt joined the ChatGPT ad pilot as the first DSP with access to inventory, and Adthena’s ChatGPT Ads Intelligence Platform launched 2026-05-11. Both are early signals that programmatic and competitive-intelligence layers are forming around the surface — but for most advertisers in July 2026 it was still direct sales or nothing.

Who that still excludes: most SMBs without enterprise-scale ad budget, and the agency-led half of the ad-buying world. The category list has clearly widened from the original B2B-SaaS focus into consumer and mid-market commerce, though, so if you were dismissed earlier on category fit, it’s worth re-checking.

How ChatGPT ads were sold during the pilot

Before Ads Manager opened, every placement went through a human sales process instead of an instant campaign launch. OpenAI still offers sales support to larger advertisers. The sales process as of May 2026:

1. Initial contact. Reach OpenAI’s enterprise team through openai.com/business. Existing API customers with enterprise contracts can route through their account team.
2. Qualification call. Sales validates budget, category fit, and brand suitability. This is where the $200K minimum gets confirmed and category bid ranges are shared. API customers move faster.
3. Creative brief. You provide brand name, favicon, a ~30-character headline, ~100-120 character body, and a destination URL. OpenAI reviews against its published ad policies, which exclude political ads, regulated verticals like gambling and certain financial products, and any creative that degrades the conversation. The policy also defines “sensitive user contexts” (mental health, vulnerability, emotional distress) where ads are blocked entirely regardless of bid. OpenAI’s framing is built around four principles: answer independence, conversation privacy, choice and control, and exclusion from sensitive contexts.
4. Campaign setup. OpenAI configured the campaign on internal tooling, with targeting (category, query keywords, country) set on their side from your brief. There was no dashboard.
5. Reporting. Performance reports came through the account team weekly or biweekly, with no live spend tracking.

That manual model worked while the buyer count was small. It could not scale to the mid-market, and in August 2026 OpenAI opened Ads Manager.

Pricing: from $60 CPM and a $200K minimum to a $25 daily budget

Before the rates, the inventory question: ads are now common rather than experimental, and we track the penetration directly. Over the seven days ending 2026-07-03, our ChatGPT ads penetration study measured a paid placement on 51.0% of US responses, 53.6% Canadian, 49.8% Australian and 18.8% Japanese, with most other markets at or near zero. For scale, that US rate is roughly 49 times Google’s classic SERP ad rate (1.05%) and about 200 times its in-AI-Overview rate (0.24%) on the same query cohort.

The surface has also been unstable in a way a media plan should account for: the rate spiked to 26.5% overall in late May, collapsed below 1% by mid-June, then re-ramped and held. Treat the current level as a plateau that has moved twice already.

The pricing has moved fast through the rollout. Here’s where it sits today.

Dimension	February 2026 launch	May 2026	August 2026 (Ads Manager)
Pricing model	CPM (fixed)	CPC (auction)	CPC, CPM or conversions
Headline rate	$60 CPM	Effective $25 CPM	Average CPC $2 – $5
Advertiser minimum	$200,000	$200,000	$25 daily budget
Software / finance bids	—	$8 – $18 CPC	—
Ecommerce / retail bids	—	$3 – $5 CPC	—
General categories	—	$3 – $5 CPC	—

Source: The Next Web, ALM Corp and Search Engine Land.

The takeaway: the $200K gate is gone. What decides your cost now is the bid and how well the landing page matches the conversation.

The flip from CPM to CPC inside the first 90 days matters. It says two things at once: demand at $60 CPM wasn’t deep enough to clear inventory at full price, so OpenAI let the auction find the real rate; and CPC moves risk off the advertiser, since you only pay on clicks. Advertisers with weak creative pay less than they would on CPM, and advertisers with strong creative pay roughly the same rate while getting more impressions.

For budget planning, the pilot’s category bid ranges are still the best guide to competition. A B2B SaaS bidding $12 CPC against a category competitor spends much faster than an ecommerce brand bidding $4 on long-tail product queries. Rough monthly spend = target monthly clicks × category mid-range bid.

During the pilot the $200K minimum disqualified most accounts: at $12 software CPC it bought about 17,000 clicks. At the $25 daily floor, a $3 CPC buys about 8 clicks a day, which is enough to test context hints before scaling.

Ad format and creative: one sponsored card, no visuals, built for relevance over saturation

ChatGPT ads render as a Sponsored card under the response sources. The format is uniform: no choice of ad unit size, no choice of placement within the response. You provide the creative, OpenAI handles placement.

What you provide:

• Brand name as it should appear in the card
• A favicon-sized brand mark (typically pulled from your domain’s favicon)
• ~30-character headline (the one-line pitch)
• ~100-120 character body description
• Destination URL (OpenAI appends utm_source=chatgpt.com&utm_medium=src automatically)

What you don’t get: image creative in the ad card, video, carousels, or any A/B testing tooling inside OpenAI’s product. A/B has to run as separate campaigns.

The constraints are tight, which is probably good news. ChatGPT ads don’t convert through visual saturation. They convert through relevance to a conversation in progress. A user asking “what’s the best HR system for a 500-person company” is in deep-funnel intent, and a one-line pitch from Rippling lands differently than the same pitch on a banner. The format constraints reinforce that.

For the back-end mechanics, how the placement actually shows up in the response payload, see the technical monitoring guide. It walks through response.result.ads[] and the bzrcdn.openai.com creative CDN.

Targeting: category and query intent work; audience targeting doesn’t yet

This is what most advertisers ask first. The picture below is inferred from pilot placements in cloro’s monitoring corpus. In Ads Manager you set location and context hints yourself, and there is no keyword targeting.

Category targeting: yes. Software advertisers target software conversations, finance targets finance. OpenAI does the category mapping based on conversation classification.

Query keyword targeting: yes, loosely. “Best X for Y” comparison queries surface ads at much higher rates than open-ended definitional queries.

Country targeting: yes. 94% of ChatGPT ads in our measurement window appeared on US-locale prompts (per the penetration study). Other countries showed isolated placements. Works as you’d expect for any geo-aware platform.

Audience or demographic targeting: not yet. No audience cohort targeting has shipped. ChatGPT obviously has the user signal to do it (conversation history, professional context inferred from prompts), but the privacy implications are real and OpenAI is moving carefully on this one.

Conversation context targeting: yes, implicitly. The placement decision uses recent conversation context, not just the current query. A user three turns into a CRM conversation surfaces CRM ads more readily than a user dropping the same single query cold.

Frequency capping per user: unclear, but probably yes at the session level. We can’t see the cap logic directly, but per-response rates stay below 1% even on high-demand queries, which suggests something is throttling.

So in practice: today’s ChatGPT ads work like classic search ads with category targeting. They don’t work like Facebook ads with audience cohorts. Plan around query intent and category fit, not demographics.

Before you spend: organic visibility and competitor ads

Paid placements work better when you already know where your brand stands in ChatGPT’s answers. Three checks are worth running before the first campaign.

Get your brand into ChatGPT’s organic answers

Before paid placements matter, organic visibility matters more. ChatGPT mentions brands far more often than it links to them, so a brand can be the dominant cited source in a category and still see almost no referral traffic. Organic visibility runs through OAI-SearchBot crawl access (ChatGPT’s live retrieval runs on OpenAI’s own search index), referring-domain authority, and third-party mentions on Wikipedia, Crunchbase, G2, Capterra, and trade press.

Our AI SEO guide covers the eight retrieval factors that drive AI citations. For ChatGPT specifically, the high-leverage ones are crawler access for OAI-SearchBot and ChatGPT-User, entity strength on third-party sources, and answer-first paragraphs that are easy to extract.

Monitor whether competitors are buying ChatGPT ads

If a competitor is paying for placements on queries in your category, you have a category-defense problem regardless of your own budget. Run your top 50 category queries through ChatGPT (or a monitoring tool) and check ads[] for competitor brands. The internal budget conversation gets easier when you can show the share of your high-intent queries where a competitor’s ad is the answer and your brand isn’t even mentioned.

The technical side of running that monitoring at scale lives in our monitor ChatGPT ads walkthrough; the ChatGPT Ads API is the productized version with brand, creative, and country breakdowns parsed for you.

Plan your first Ads Manager campaign

Use the pilot data to start small and learn fast:

• Define your top 100 priority queries (the category’s high-intent prompts where ChatGPT visibility matters)
• Model expected campaign cost from your historical search-ads CTR and a starting max CPC of $3 or more
• Document where your top 5 competitors appear in ChatGPT answers today (organic mentions, citations, paid placements)
• Split context hints into separate ad groups so you can see which conversation types convert

Don’t wait for ads before fixing your crawler access

ChatGPT’s organic citation pipeline runs through OpenAI’s own crawler and search index, not Google’s. If OAI-SearchBot and ChatGPT-User are blocked in robots.txt or at your CDN, your ChatGPT visibility is capped regardless of how good your Google SEO is. The simplest pre-work for ChatGPT advertising is making sure those user agents can actually crawl your priority pages.

What self-serve changed

The pilot pointed to self-serve on a Bing Chat timeline: Microsoft started exploring ads in Bing Chat in March 2023 and moved them into standard Microsoft Advertising tooling within about a year. OpenAI moved faster, opening Ads Manager six months after its February launch.

The minimum fell further than most forecasts expected: from $200,000 to a $25 daily budget, not to a tiered $1,000 to $5,000 monthly floor. Ads Manager also ships the pieces the pilot lacked: campaign, ad group and ad structure, a choice of CPC, CPM or conversion bidding, reporting on impressions, clicks, spend and conversions, and the OpenAI pixel for conversion tracking (OpenAI’s Help Center). What it does not have is keyword targeting. Context hints replace keywords, so the work of matching an ad to a conversation still sits with the advertiser.

For measurement, the monitor ChatGPT ads guide shows how to see which competitors appear on your queries once they can all buy in.

Compliance: IAB disclosure rules are voluntary today, FTC’s aren’t

Beyond OpenAI’s own ad policies, two external frameworks are worth knowing as disclosure rules catch up.

The IAB released its AI Transparency and Disclosure Framework in January 2026, recommending consumer-facing disclosures for AI use in advertising backed by machine-readable C2PA metadata. It’s voluntary today, but the direction of travel is toward formal disclosure standards. IAB Tech Lab’s agentic advertising initiatives are extending those standards to AI agents that act on ads autonomously, which is a separate problem ChatGPT advertisers will face within the next year.

On the regulator side, FTC AI content disclosure rules apply to AI-generated ad creative regardless of platform. If you use AI tools to generate the ad copy or imagery you submit to OpenAI, you carry the disclosure burden under FTC rules even though OpenAI handles placement. Worth flagging to legal before any AI-generated creative ships.