By Joel Horwitz, founder of Synter. All numbers in this post come from our own OpenAI Ads account, pulled via the API on June 11, 2026. Nothing is estimated.
In This Guide
TL;DR
We have been running ChatGPT Ads since June 2, 2026. Nine days of delivery, five markets, 23 ad groups, every dollar tracked through the API:
- 129,663 impressions, 448 clicks, $4,428.84 spent
- Blended CTR 0.35%, blended CPC $9.89, blended CPM $34.16
- CPMs are nearly flat across markets ($26 to $37). CTR is not. The UK clicked at 0.72%, more than double the US at 0.33%, which made UK clicks cost $5.10 against $10.62 in the US.
- Targeting is context hints, not keywords. You describe the buyer's situation in plain sentences and OpenAI matches your ad to conversations.
- Three of our ads were flagged under "Platform integrity & misrepresentation." All three claimed autonomy. The rewrites that passed lead with what the user gets, not what the AI does.
Most of what has been published about ChatGPT Ads so far is written by people who have never run them. This post is the data.
Why this post exists
ChatGPT Ads opened to advertisers this spring. The guides that rank for it today were largely written before their authors had account access: speculation about how targeting probably works, recycled OpenAI launch-post language, best practices inferred from other platforms.
We run paid media for a living, so we put real budget on it the week we got access. This is what the account actually shows: spend, impressions, clicks, rejections, and the targeting methodology we used. Where our data is thin, we say so.
How OpenAI Ads actually works right now
Five things that surprised us coming from Google and Meta:
- The unit is a chat card. A title, a one-or-two-sentence body, an image, and a target URL, rendered inside a ChatGPT answer. There are no headlines-by-the-dozen responsive formats. One card, one message.
- Targeting is context hints, not keywords. Each ad group takes a handful of natural-language descriptions of the conversation you want to appear in. Ours read like buyer situations: "marketing ops teams that need pipeline reporting without manual data stitching." OpenAI matches your card to conversations semantically. There is no search-terms report yet, so the hint is both your targeting and your only lever.
- Billing is by impression. You set a max CPM bid per ad group (ours range from $40 to $65) and a lifetime spend cap per campaign. There is no daily budget in the form you know it, which makes delivery spiky.
- Geo targeting is country-level. We run separate campaigns per country: US, UK, Canada, Australia, New Zealand.
- Conversion events exist and attach at the campaign level. We track trial signups; volume is too small at our spend level to publish split tests yet, and anyone telling you they have statistically significant ChatGPT Ads conversion data on a few thousand dollars is rounding up.
The numbers
Nine days of delivery, June 2 to June 10, 2026. The US campaign went live first; we expanded to the other four markets on June 7.
| Market | Impressions | Clicks | CTR | CPC | CPM | Spend |
|---|---|---|---|---|---|---|
| US | 103,133 | 337 | 0.33% | $10.62 | $34.69 | $3,577.61 |
| UK | 8,989 | 65 | 0.72% | $5.10 | $36.90 | $331.73 |
| Canada | 9,181 | 33 | 0.36% | $8.33 | $29.95 | $274.96 |
| Australia | 5,714 | 10 | 0.17% | $17.59 | $30.78 | $175.88 |
| New Zealand | 2,646 | 3 | 0.11% | $22.89 | $25.95 | $68.66 |
| Total | 129,663 | 448 | 0.35% | $9.89 | $34.16 | $4,428.84 |
Daily delivery was anything but smooth. Three impressions on June 2. A ramp to 10,559 on June 4. A spike to 45,419 impressions and $1,636 in a single day on June 8, followed by a drop back to 8,022 the next day. With impression billing against lifetime caps and no daily pacing control, the platform decides your rhythm. Budget accordingly.
Finding 1: you pay the same CPM everywhere. CTR decides everything.
The spread between our cheapest and most expensive market CPM is about $11. The spread in CTR is 6.5x.
The UK ran the same topic stack as our other expansion markets and clicked at 0.72%, with our single best ad group (B2B demand generation, UK) hitting 0.76% CTR at a $4.80 CPC. New Zealand, same methodology, clicked at 0.11% with a $22.89 CPC. When impressions cost roughly the same everywhere, CTR is the entire efficiency story, and CTR is a function of how well your context hints match real conversation volume in that market.
Practical takeaway: do not copy your US budget split into ChatGPT Ads. Launch wide and cheap, read CTR per market within the first week, then concentrate.
Finding 2: inventory follows conversation volume, not your bid
We ran the same four-topic structure in every market: B2B demand generation, CRM-integrated advertising, pipeline attribution, and programmatic advertising. The platform did not distribute impressions evenly, no matter the bid.
In the US, one topic (AI creative and ad automation) absorbed 76,545 impressions, 74% of our US volume, at a 0.34% CTR. Meanwhile the programmatic ad group got 188 impressions in the US, 49 in Canada, 3 in Australia, and exactly 1 in the UK, across nine days, at bids comparable to everything else.
The lesson: ChatGPT Ads inventory is a mirror of what people actually ask ChatGPT about. People are clearly asking it about AI tools for ad creative. Almost nobody is asking it about programmatic DSPs. No bid fixes a conversation that is not happening. Pick topics by conversation volume, and if an ad group is starved after a week, the conversations are not there. Kill it.
Finding 3: what gets rejected, verbatim
Three of our ads were flagged with the same review reason: "Disallowed - Platform - Platform integrity & misrepresentation." These are the exact titles:
- "The AI Agent Operator for Ads"
- "Autonomous Media Buying for Growth Teams"
- "Stop Babysitting Your Ad Campaigns"
The pattern is unmissable: every flagged ad led with the claim that the AI acts on its own. We market autonomous AI execution, so this stung, but the policy logic is coherent. Inside a ChatGPT answer, a card claiming an AI will run things for you reads like the assistant itself making a promise. OpenAI appears to hold that line hard.
These rewrites were approved:
- "Your growth team, running at full speed" / "Synter ships campaigns while you focus on strategy. You set direction, it builds and launches."
- "From brief to live in 30 minutes" / "Describe your campaign. Synter builds it across Google, Meta, LinkedIn, and TikTok. You approve."
- "Run paid media across 15 platforms" / "One conversational interface. Direct API to Google, Meta, LinkedIn, TikTok, Reddit, and more."
What changed: the approved versions lead with the user's outcome and keep a human in the sentence ("you set direction," "you approve"). Concrete mechanics passed review. Autonomy claims did not. If your product is AI that does things, write the card about what the buyer gets and where they stay in control.
The methodology: seed context hints from AI-visibility data, not keyword tools
This is the part nobody else is doing, and we think it is the most repeatable idea in this post.
Keyword tools tell you what people type into Google. They tell you nothing about what people ask ChatGPT. But AI-visibility tracking does: we already monitor how often Synter and its competitors appear in LLM answers across a library of tracked prompts, grouped by topic, using Gauge. Each topic carries a visibility score: how often we show up in AI answers about it.
We turned that directly into campaign structure:
- Rank topics by tracked AI-conversation visibility, not search volume. Our top tracked topics became our ad groups.
- Split defend from conquest. Topics where we already show up organically in AI answers get "defend" ad groups to own the conversation. Topics where competitors get recommended and we do not get "conquest" ad groups.
- Write hints as buyer situations, not keyword strings. Five per ad group, each a sentence a real buyer might be living: "performance marketing agency managing 20 client accounts wants AI to automate the entire campaign workflow."
- Reuse the topic stack across markets, then let CTR vote. Same hints in the UK and New Zealand produced a 6.5x CTR difference. The data tells you where the conversations are.
The general version, if you do not track AI visibility: spend an hour asking ChatGPT the questions your buyers ask, note where competitors get named, and write hints for those conversations. The worst input you can feed this platform is your Google Ads keyword list.
What we are changing next
Our own next moves, so you can sanity-check this against what we actually do:
- Shift budget share toward the UK and hold Canada. Same CPM, half the CPC.
- Kill the starved programmatic ad groups everywhere and the attribution groups in Australia and New Zealand (one click on 2,997 combined impressions).
- Keep US creative-automation funded. 74% of US volume at account-average CTR is reach we cannot buy this cheaply elsewhere.
- Let conversion events accumulate before claiming anything about CPA. Impressions and clicks are real today; revenue math needs more data.
Synter connects to every major ad platform via direct API.
Sign up freeFAQ
What are ChatGPT ads?
Sponsored chat cards that appear inside ChatGPT answers. Each card has a title, a short body, an image, and a link, and is matched to conversations via advertiser-written context hints rather than keywords.
What do ChatGPT ads cost?
Across our account: a $34.16 average CPM, a $9.89 average CPC, and per-market CPCs from $5.10 (UK) to $22.89 (NZ). Billing is per impression against a max CPM bid; our bids range from $40 to $65 CPM.
How does targeting work without keywords?
You write context hints: natural-language descriptions of the buyer's situation. OpenAI matches your card to semantically similar conversations. There is no search-terms report, so hint quality is the whole game.
What gets ads rejected on ChatGPT?
In our account, every rejection carried the reason "Platform integrity & misrepresentation," and every flagged ad claimed the AI acts autonomously. Cards that lead with user outcomes and keep the human in the loop passed.
Is ChatGPT Ads worth it for B2B?
At our scale: yes for reach into high-intent AI conversations, unproven for direct response. CTRs of 0.3 to 0.7% at ~$34 CPM are workable, and the inventory is uncrowded. Treat conversion claims, ours included, as premature.
Methodology
Every number is from Synter's own OpenAI Ads account, June 2 to 10, 2026, pulled via the OpenAI Ads API on June 11 and cross-checked against per-campaign daily insights. Spend figures are USD and include every test dollar, including the failures. We will publish a follow-up with conversion data when the sample is large enough to mean something. Synter sells cross-platform campaign execution, including OpenAI Ads support, so read our enthusiasm with that in mind; the rejections section should help you trust the rest.