Running ChatGPT Ads: what the first 7,000 impressions taught us

Most of what's written about ChatGPT advertising is paraphrased from OpenAI's documentation. This isn't. Calibre has been running ChatGPT Ads on our own account since the channel opened to self-serve advertisers: our own money, our own name, our own data.

This post is for marketers, founders and in-house teams weighing the emerging channel: what the platform actually is, the real numbers from our first 7,300+ impressions, and the five lessons we'd hand any business entering now. If you're deciding whether ChatGPT ads deserve a share of budget this quarter, this should save you the first month of trial and error, and help you focus the test properly.

ChatGPT advertising in one minute

OpenAI spent years saying no to digital advertising: in late 2024 the company publicly rejected advertising plans. That position reversed on February 9, 2026, when ChatGPT ads launched in beta for Free and introductory tiers, the clearest sign yet of how serious the ad revenue push is. The pilot expanded to Canada, Australia and New Zealand in March 2026, with early coverage also pointing to tests in South Korea. Today the channel is open to advertisers in five markets (the US, UK, Canada, Australia and New Zealand) and still limited everywhere else, including the EU.

The rules the platform enforces

• Ads are served only to adult users (18+) on the Free and Go tiers. Logged in users on paid tiers like Plus and Pro don't see them.

• Every placement is one of the clearly labeled ads formats, kept separate from ChatGPT's responses: a sponsored card below the chatbot's answer with a brand name, favicon, headline and description.

• Ad placements do not influence the AI's core responses. Advertisers cannot change or steer what the artificial intelligence says; the organic answer stays earned, not bought.

• OpenAI maintains conversation privacy: ads are matched to the topic of the user's conversation in the moment, not to chat history or identity.

• Access runs through OpenAI's Ads Manager or approved partners, which is where an agency account like ours lives.

Why marketers care about the audience

Scale first: ChatGPT reaches a reported 900 million weekly active users (the world's most-used of the new ai tools) with hundreds of millions of free users on tiers where ads appear. More interesting than scale is the mode: these are consumers and professionals in active research, and surveys put AI use in purchase decisions at around 44% of users.

That's the pitch of this channel in one line: reach people at a pivotal moment, while they're comparing options and making decisions, not idly scrolling a feed.

1. Context hints are a different discipline, not a new keyword field

In Google Ads you buy keywords. On Meta you buy audiences. ChatGPT advertising targets neither: you describe, in plain language, the situation your buyer is in, and the platform decides whether a live conversation matches it.

Writing hints that match real conversations

Keyword thinking produces context hints that are too broad to match anything well. What performs is situation phrasing: the moment someone is comparing options, the problem they're describing in their own words, the decision they're about to make. Intent, written out as a sentence. For example, "a founder comparing ways to advertise on new AI platforms" will outperform any list of category words.

One practical note: the platform is young, and parts of the ads manager still behave like a beta. Some settings are effectively write-only: you set them, and verification means checking change history rather than reading them back. Plan your campaigns and account structure before you build, because auditing them afterwards is harder than it should be.

2. What a good click-through rate looks like right now

Across our first 7,300+ impressions, our ads have held a click-through rate around 1.4 to 1.5 percent. We created two variations: a straight product framing and a story-led framing, and they landed within a tenth of a point of each other.

That near-tie is the finding. The ad format is small and standardised, so there's less room for creative to dominate the way it does in feed-based digital advertising. In our data, the leverage sits in the context hint (showing up in the right conversations) more than in copy variations. Iterate the hint first, the copy second, and keep an A/B test running because generating variations is cheap.

For calibration: 1.4 to 1.5 percent on cold, top-of-funnel placement beats typical display banners by a wide margin and holds its own against good social prospecting. The conversational context does real work for relevance.

3. Your ad is read as part of an answer, not an interruption

A chat_card appears directly below ChatGPT's reply, mid-conversation, while someone works through their actual problem. Banner instincts fail here. Copy that reads like an interruption gets ignored; copy that reads like a useful continuation of the answer gets clicked.

In practice, the headline does the work, but in the register of the conversation, with a clear value proposition and zero ad-speak. The messages that feel relevant to the conversation get the clicks. It's closer to "the answer, continued" than "the pitch, inserted."

The same lesson applies to your account setup: precise, craft-level descriptions of what your company does clear review faster and match better than broad category labels. How you describe the business literally shapes where your ads appear.

4. Measurement is the real build

This is the lesson that separates running the channel from testing it. ChatGPT's audience skews toward ad blockers and strict cookie settings, and buyers who click an ad often don't convert in that session; they come back later through Google branded search, or type the site in directly.

The stack we actually run

A browser pixel alone will undercount this channel badly. Our stack is the pixel plus the server-side Conversions API with hashed identifiers, UTMs on everything, and, just as important, watching branded search and direct traffic during flights, because that's where ChatGPT-driven customers actually surface in the data.

We built our own server-side attribution bridge for this, and it's the piece of performance marketing infrastructure we'd least want to run without. If a report on this channel only shows platform-tracked conversions, it's not showing the channel.

5. The economics favour whoever builds history first

The auction runs on familiar CPM and CPC pricing models. On our spend to date, blended CPM has landed around USD $22, early inventory pricing well below the premium CPMs this placement will command as more brands arrive, the auction fills, and OpenAI scales its advertising revenue.

Cheap impressions are nice. What matters more is that campaign history, creative learnings and account signals compound. Advertisers who are live now, in the early access phase of the channel, are building an advantage that later arrivals will pay more per impression to catch up with. As broader access rolls out, enterprise budgets included, that gap widens.

Know the audience while it lasts, too: the Go tier skews toward founders, freelancers and small-business decision-makers. People with real buying authority, which is why considered B2B and service purchases are the natural early focus for this strategy.

Quick answers for busy marketers

Do ChatGPT ads change what ChatGPT says?

No. Ads sit below the response, clearly labeled and separate. Advertisers can't steer the AI's answers; user trust in the organic response is the thing OpenAI is protecting hardest.

Who actually sees ChatGPT ads?

Adult users on the Free and Go tiers, in the US, UK, Canada, Australia and New Zealand. Paid tiers don't see ads, and the EU isn't open yet.

How are ChatGPT ads priced?

CPM and CPC models through OpenAI's Ads Manager. Our own blended CPM sits around USD $22, cheap for placement inside a high-intent conversation, and unlikely to stay that way.

Can you target keywords or audiences?

Neither. Targeting is by context hints (plain-language descriptions of buyer situations) matched against the live conversation, with country-level geography. No chat history, no user profiles, no lookalikes.

What we'd tell a brand entering now

Treat it as a pilot, not a launch. Correct setup, server-side measurement from day one, one or two sharply defined context hints, an honest read after a few thousand impressions, then scale what the data supports. In our opinion it's the most interesting test in digital advertising right now: an ai powered search surface where attention is cheap, intent is high, and most of your competitors haven't turned up yet.

That's exactly how we run it, for ourselves and for clients. If you'd rather skip the trial and error and get in touch today, the sprint is fixed-scope and takes about five days.

Calibre Studio is a ChatGPT ads agency running done-for-you campaigns in every open market, with market pages for Australia and the UK. Sprint details at ads.calibrestudio.co.