Advertising in ChatGPT has moved from rumor to working tool: OpenAI opened its own ad manager, and ads now appear right under the model's answers. For media buyers and marketers this is a new traffic source with a nearly empty auction — few competitors so far and a huge audience. This guide covers how OpenAI Ads works, how it differs from the familiar Google Ads and Facebook, where the budget traps hide, and how to launch a first campaign without stepping on the rakes early testers have already found.
What OpenAI Ads is and where the ads appear
OpenAI Ads is the platform for buying traffic from ChatGPT. An ad appears under the model's answer when the conversation topic matches what you promote. The unit is modest: brand logo and name, an Ad label, a headline up to 50 characters and a description up to 100. No images or video — text and a link only.
The key difference from classic contextual advertising: only users on the Free and Go tiers see ads. Plus and Pro subscribers live ad-free and cannot be reached through the manager. That matters for audience estimates: paying subscribers are usually the wealthier segment, and they are out of reach.
The second difference is targeting. There are no keywords in the usual sense. Instead there are Context hints: you describe the situation and user intent, and the algorithm decides in which dialogue to show your ad. More on this below, because Context hints decide whether your ad lands in target conversations or scatters across random ones.
Who can access the manager and what budget to start with
At launch the manager is open in 52 countries: the US, Canada, the UK, EU countries, India, Japan, South Korea, Australia, the UAE and others. OpenAI publishes the current list in its documentation, and it keeps growing.
The minimum daily budget is low — roughly $9 a day equivalent, though the manager pre-fills a sum several times larger, which is worth checking before launch. The budget also hides the platform's main trap: the Daily budget you set is an average spend, not a hard cap. On a single day the system may spend up to two daily budgets, and up to seven over a week. Set $25 a day and be ready for $50 on a peak day. For a two-day test this is critical: plan around the upper bound.
How the manager is structured: three levels
The structure is familiar to anyone who has used Google Ads: Campaign → Ad Group → Ads. First a campaign with budget, geo and objective, inside it an ad group with bid strategy and context hints, inside the group the ads themselves.
The left menu: Overview with general stats (impressions, clicks, average CPC over 7, 14 or 30 days), Campaigns, Tools with conversions, audiences, feeds and change history, Billing and Settings.
An important detail on the Campaigns tab: the Active toggle being on does not mean the ad is running. The real state is in the Status column. A campaign can sit in Not serving with Active on — usually due to unfinished billing tasks, budget or audience issues, or ad moderation.
Step-by-step campaign setup in OpenAI Ads
Step 1. Account and billing
After sign-up you land on Overview, where the Tasks block in the top-right lists mandatory setup items. A task marked Required blocks launch: until you confirm payment details, the campaign stays in Not serving. Check this block first — most "why isn't it running" cases are solved here.
Step 2. Campaign and objective
Create → Create campaign. Set a name (at least three characters); the only type for now is Standard. Then the objective: Clicks or Conversions. Start with Clicks: the system optimizes for visits and needs no conversion data you do not have yet. Conversions makes sense only once the pixel is set up and verified — otherwise the algorithm trains on junk and burns budget "optimizing" toward empty events.
Step 3. Budget and schedule
The Budget & schedule block. Check the pre-filled Daily budget and lower it to your figure. Below the field the manager honestly shows maximum daily and weekly spend — read those numbers, not just the main field. In Schedule the campaign runs indefinitely by default; for a test enable Set an end date so you do not discover a week later that it is still spending.
Step 4. Geo and platforms
Targeting sets locations. The default is the whole account country; you can narrow to regions and cities. Locations to exclude takes priority over included ones: if a city is in both lists, it gets no impressions. Platforms: Android app, Android web, Desktop web, iOS app, iOS web — all five on by default. For a desktop offer keep only Desktop web.
Step 5. The Advanced block: UTM, audiences and text rewriting
This block is collapsed by default and most often skipped — a mistake. Three things inside:
- Landing page query parameters — UTM tags with macros: {campaign_id}, {ad_group_id}, {ad_id} are filled in by the system. Without them your tracker cannot tell campaigns apart.
- Custom audiences — connecting your own audiences, for example a customer base to exclude from delivery.
- Text customization — on by default, it lets the AI rewrite and translate your headlines and descriptions without approval. For offers where every word is vetted (finance, legal wording, brand-approved copy), switch it off — otherwise versions you never saw will run.
Step 6. Ad Group and Context hints — the core targeting
At the group level you pick a bid strategy: Maximize results (the system manages bids within budget) or Manual: Max CPC (you cap the click price). For first tests Maximize is more sensible, to collect real CPC data.
Then Context hints — the heart of the platform. The manager lets you leave the field empty, but then you hand targeting to the algorithm. The right approach is to describe the dialogue scenario: who the user is, what they are looking for, what task they are solving. A comma-separated word list works poorly; a detailed description of intent works well.
A weak hint: "tracker, analytics, media buying". A strong one: "The user is choosing a tracker for ad campaigns, comparing cloud and self-hosted options, and asking which suits a small team on a limited budget." The more specific the situation, the better the system understands where your offer belongs.
Below is the Default Ad Destination URL: use a lander for the specific dialogue topic, not your homepage.
Step 7. The ad
Headline up to 50 characters, Description up to 100, with a counter next to the fields. Suggested ads offers ready variants based on campaign settings and site content — take them as a base and edit. Live Preview shows how the ad will look under a ChatGPT answer.
Step 8. Review and launch
The final screen gathers everything: name, objective, geo, platforms, budget, bid strategy, Context hints and ads. The pencil icon on each card returns you to editing. Publish launches. Check here while mistakes still cost nothing.
Conversions and Identifier coverage
Tools → Conversions handles pixel setup and target actions. It has a metric other managers lack: Identifier coverage — the share of events sent with data that can identify the user (hashed email, phone, session cookie). That is how OpenAI knows whether the person saw an ad in ChatGPT before converting.
Practical rule: if Identifier coverage is near zero, the system cannot link conversions to impressions, and switching to the Conversions objective is premature — it would optimize blind. Get proper coverage first, then change the objective.
Audiences, feeds and change history
Audiences takes customer emails and phones (raw or SHA-256 hashed). The most useful application is a suppression list: exclude people who already bought so you do not pay for their clicks twice. Feeds is for online stores — a product catalog for dynamic ads. Change History stores every account change, handy for teams.
What it means for media buying
OpenAI Ads is a young auction with low competition and moderation geared toward white-hat offers: SaaS, education, e-commerce, licensed financial products, subscriptions. The free ChatGPT audience is mass-market with a strong skew toward informational queries: people ask for advice rather than where to buy. Two consequences: offers that fit naturally into a consultative dialogue will convert, and the lander must continue the conversation rather than restart it.
Budget tests around the double daily limit, set an end date, tag everything with UTM and run it through a tracker — otherwise you will never know which Context hints brought cheap clicks. We have a separate blog guide on tracker and postback setup.
Frequently asked questions
How much does advertising in ChatGPT cost? The minimum daily budget is about $9 equivalent; click price is set by auction and depends on geo and topic. On a single day the system may spend up to two daily budgets.
Who sees ads in ChatGPT? Users on the Free and Go tiers. Plus and Pro subscribers see no ads.
Are there keywords in OpenAI Ads? No. Targeting is set via Context hints — a text description of the user's situation and intent that the algorithm matches to dialogues.
Can I run ChatGPT ads from any country? The manager is available to advertisers in 52 countries, and the list keeps growing — check OpenAI's current documentation.
Which objective for the first campaign? Clicks. Conversions only after the pixel is set up with proper Identifier coverage.
Conclusion
OpenAI Ads is not "another Google Ads" but a different logic: dialogue descriptions instead of keywords, a text line under the model's answer instead of banners, an averaged budget with a double daily ceiling instead of a hard cap. Three things decide a test: close every Required task before launch, write Context hints as a conversation scenario rather than a word list, and read the fine print under the budget field. After that it is business as usual: tests, tracker numbers and scaling what pays back. Share your first ChatGPT-traffic results in the Traffic Sources section of the forum — the source is new, and every number is gold right now.