-
I got into the ChatGPT Ads Manager beta — sharing my first campaign setup, will report back with data
Posted by on May 21, 2026 at 8:07 amA quick note up front: Everything in this post is from my own hands-on experience going through the OpenAI Ads Manager beta — every step, every screenshot, every observation is real. My original write-up was scattered (I was taking notes as I went), so I used AI to help me organize the structure and make it readable. The substance is mine; the formatting got help. Full screenshots included below for anyone who wants to verify. If AI-assisted editing is a dealbreaker, no hard feelings — skip away.
TL;DR:
- Got into the OpenAI Ads Manager beta (15-day wait after applying)
- Full walkthrough below: dashboard → campaign setup → context hints → ad creative → submit
- Real performance data coming in a few days
Applied for the OpenAI Ads Manager beta about two weeks ago and got approved a few days back — roughly 15 days from submitting the application to receiving the access email. Just finished setting up my first campaign.
Since the self-serve beta opened on May 5th there haven't been a ton of hands-on walkthroughs out there, so I figured I'd document the whole process — from the invite email through hitting submit — and come back in a few days with actual performance numbers.
Posting this now while it's fresh. Data update will go in the edits.
The invite email
Came through as a standard transactional email — clean, minimal, just a login button. Nothing fancy.
Dashboard — first impression
Once you're in, the UI is minimal — almost surprisingly so. Left sidebar has just four sections: Campaigns, Tools, Billing, Settings. No reports tab, no audience manager, no asset library. Compared to the sprawl of Google Ads or Meta Ads Manager, this feels stripped down.
The main view defaults to a Campaigns tab with sub-tabs for Campaigns / Ad groups / Ads — so the classic three-level hierarchy is there, just collapsed into one screen instead of separate pages. A green banner at the top confirms "Account verified — you're approved to launch ads in ChatGPT."
A few things worth flagging from the sidebar before diving into campaign creation:
- Tools only has Change History and Conversions right now
- Billing has the full set (Overview / Activity / Documents / Settings) — payment infrastructure is clearly more built out than the campaign tooling
- Settings has an API Keys section. So programmatic campaign management is on the table from day one, which is more than I can say for some platforms that have been around for a decade
Creating a campaign — Objective
First decision: campaign objective. Three options in the dropdown — Reach, Clicks, and Conversions — but Conversions is currently labeled "Coming soon." So right now you're choosing between optimizing for impressions (CPM) or clicks (CPC). The CPC option only launched with this beta — during the earlier managed pilot, CPM was the only buying model.
Worth noting: even though Conversions as an objective isn't live, you can still set up conversion tracking to measure results. The algorithm just won't optimize toward those events yet.
Country targeting
Location targeting at the campaign level is hard-capped to four countries: United States, Canada, Australia, New Zealand. That's it. The "Enter another location" field exists but the dropdown won't accept anything else.
Matches what OpenAI said publicly: ads only serve to Free and Go tier ChatGPT users in those four markets right now. If you're hoping to run ChatGPT ads in the UK, EU, or anywhere in APAC outside ANZ, you're waiting.
Budget
Standard setup — pick Campaign budget (total spend cap) or Daily budget (a daily ceiling). Same model as Google Ads and Meta. No minimum spend floor that I could find — which is a real change from the early pilot days when entry was reportedly $50K+.
Conversion tracking & the web-only problem
Before you can set up conversion events, you have to create a data source. The form asks for a name, then a Type: Web / iOS / Android.
Only Web is selectable. iOS and Android are both marked "Coming soon."
This is the single biggest limitation for anyone planning to run ChatGPT ads for an app. You can drive traffic to a web landing page and track it, but there's no native app-install or in-app event tracking yet. If your business is app-first, this beta isn't ready for you — though the fact that the form already has iOS/Android slots suggests it's on the roadmap.
The Web data source itself gives you a standard pixel: a
<script>tag with a unique pixel ID that you drop into your site's<head>. Once installed, you can define conversion events — currently the only base event type ispage_viewed, with a configurable conversion window (default 30 days). There's also a separate Conversions module under the Tools sidebar with a "Manage conversion keys" option — so the pixel/server-side infrastructure exists, it's just gated to Web for now.Ad group — where it stops looking like Google Ads
The Ad group page is where ChatGPT Ads starts to diverge from everything else. Four fields:
- Ad group name — standard
- Maximum CPC bid — range is $0.01 to $100.00. The default placeholder is $3.00, which is a tell: OpenAI seems to think that's roughly the right ballpark for early auctions. For context, average Google Search CPC sits around $1-$2 and Meta around $0.50-$1.50 in most verticals. ChatGPT Ads is launching with a higher floor signal than either
- Website URL — default destination for ads in this group. Standard ad-group-level setting
- Context hints — and this is the part that's actually new
OpenAI's own description of context hints:
"Describe the conversations, topics, or keywords where your products or services may be relevant; these hints guide matching but aren't exact-match targeting rules."
Read that twice. A few things stand out:
- It's called hints, not keywords or targets. Soft guidance, not hard filters
- "Guide matching" — the LLM uses these to understand what conversations your ad belongs in, not to do string-matching against user input
- "Aren't exact-match targeting rules" — they're explicitly telling you not to think in Google Ads terms
So this isn't keyword bidding. It isn't audience targeting either. It's something closer to: "here's a description of the context where my ad makes sense — figure out the rest, model."
The input UI itself is bare. Just a multi-line text box. No tagging, no chips, no autocomplete, no suggested hints, no keyword volume tool, no competitor research. You're writing into a blank text box and trusting the model. That's either liberating or terrifying depending on where you sit.
My read on how to write them (happy to be wrong):
- Too narrow probably under-delivers — the model can semantically expand from
transcribing meetingsto a lot of related conversations, so being too literal limits reach - Too broad probably tanks relevance —
productivityis so vague the model has nothing to anchor on - Sweet spot is probably scenario + intent, e.g.
taking notes during long meetingsrather than justnotes
No one knows the best practice yet. This beta launched two weeks ago. There's no data on what works.
Ad creative
Five fields:
- Ad name — internal label
- Website URL — inherits the ad group default but can override
- Headline — 50 character limit
- Description — 100 character limit
- Ad images — PNG/JPG, square recommended, minimum 256×256px
Character limits are tight. For comparison, Google Search Ads gives you 3× 30-char headlines plus 2× 90-char descriptions — roughly 270 characters total. ChatGPT Ads gives you 150 characters total. You have to land the value prop in one sentence.
The preview shown in the panel looks like a small card with a "Sponsored" label, brand name, headline (truncated at ~35 chars in the preview), description (also truncated), and a small thumbnail image. The ad unit is clearly designed to slot below a ChatGPT response, not to dominate the screen.
A few observations on the creative spec:
- Image displays small in the preview — fine detail in the image is wasted
- Headline gets truncated well before 50 chars in the actual preview — front-load the most important information
- No CTA button selection that I could find (no "Learn More" / "Download" / "Sign Up" picker)
- No A/B testing primitives at the ad level yet — if you want to test creatives, you build multiple ads under one ad group
Review and submit
The final step is a Review page that summarizes everything: campaign name, objective, locations, budget, conversion event, schedule, ad group settings, context hints, and the ad itself. Hit submit and the campaign goes into review.
No timeline given on how long ad review takes. I'll update once mine clears.
Surprises, friction, and what's missing
A few takeaways after going through the whole flow:
🟢 Surprises (in a good way)
- API Keys section exists in Settings — programmatic management from day one
- No minimum spend floor — the rumored $50K pilot entry is gone
- The three-level Campaign / Ad group / Ad hierarchy is exactly Google Ads — zero conceptual onboarding for anyone with Google experience
- Conversion tracking infrastructure (pixel + events + conversion keys) is mature for a 2-week-old self-serve beta
🟡 Friction
- Default CPC suggestion of $3.00 is high relative to other channels — feels like an anchoring move
- Context hints input is a blank text box with no guidance, no suggestions, no volume estimator — you're flying blind
- Character limits on headlines/descriptions are aggressive
- Preview truncates headlines around 35 chars even though the field accepts 50
🔴 Missing (for now)
- Conversions as an objective — Coming soon
- iOS / Android data sources — Coming soon
- Countries outside US / CA / AU / NZ
- CTA button selection
- A/B testing tooling
- Audience exclusion / frequency cap (didn't find these — happy to be corrected if I missed them)
What I'm doing next
The campaign is configured and ready. I'll come back with actual performance data once it's been live for a few days — impressions, clicks, real CPC, CTR, which context hints actually pulled volume, and how the ads end up looking inside ChatGPT.
For anyone else in the beta — would love to compare notes, especially on context hints. Did your first batch get approved on the first try? Any patterns on what gets flagged?
Will edit this post with the data update soon.
-
1 Reply
-
saved this, thanks for documenting it while it’s fresh. a few questions if you have time when the data lands:
– did you have to map the OpenAI Ads Manager pixel separately or does it accept GA4/Meta/whatever as a fallback for conversion attribution?
– on the “context hints” feature, is that a per-campaign config or per-creative? trying to figure out if it’s setting up the AI’s targeting model or telling it about your offer
– creative-side: did you upload static + video or are they keeping it text-only in v1? and is there any preview of how creative renders in chatgpt’s UI vs in-app vs across surfaces?asking because i’m planning to apply for the beta this week and want to know how much groundwork to do before submitting (eg if i need a separate landing page strategy because chatgpt traffic probably acts differently than search/social traffic). gut sense from the screenshots is that this is closer to discovery campaigns than search but i could be wrong.
also: 15 days is faster than i expected. did your application mention specific use case or vertical, or was it generic? i’ve seen mixed reports on what they prioritize for early access
-
Thanks for the write-up. Any intel on what context hints are landing well? Is it going broader or keeping the targeting quite tight?
-
Thank you!! I tried applying before reading your post, let’s hope it will be available in the EU soon.
-
tbh, that minimal ui sounds like a nightmare for b2b. the thing is, without audience targeting or any real reporting, how are you even gonna qualify leads? my last ad campaign was all about granular controls for specific icps.
this feels like a broad net, which is fine for some. but for b2b sales, you really need precision. curious to see what data you pull, but i’m skeptical about the roi for anything beyond brand awareness.
-
the scenario + intent framing matches what i’m seeing too, treating hints like a query you’d actually type beats dumping keyword-style lists, pain-based ones have pulled more volume than literal feature terms for me so far
-
Following
-
Is there only country specific targeting for those few countries or can you go down to the state level (e.g. New York)?
-
no audience targeting, no retargeting, no lookalikes. ur not running performance ads , ur buying attention and hoping the context model finds the right people.
attribution is the actual gap nobody’s asking about. if the pixel setup isn’t documented, all the “data coming in a few days” is sessions and clicks, not conversions. early betas on new ad platforms usually come with subsidized CPMs while they build out inventory. month 3 is the real test.
-
Thank you!
-
Is it still open, can you still sign up? If so, do you mind sharing the link?
Log in to reply.