Where the ChatGPT Ads Budget Should Come From
By Keshav Parsai 8 min read
In brief
A ChatGPT Ads test funded out of search budget will be killed by search maths, because it is not a substitution buy and cannot be judged as one.
Last verified: 12 September 2026 | Version: 1.0 | Next scheduled review: 12 December 2026
Take it from brand, display or an experimental line. Not from search. The reason is not political, it is measurement: money moved out of search arrives carrying search's evaluation rules, and ChatGPT Ads cannot survive them at this stage.
Why the source of the money decides the verdict
Budget carries its accounting with it. This is the part that gets skipped in internal conversations about "where do we find the 3,000 dollars".
Money taken out of paid search is judged on incremental cost per acquisition against the campaign it was taken from, usually within one reporting cycle, because that is what the search line is measured on. Money taken out of brand or display is judged on reach, cost efficiency and learning, over a quarter, because that is what those lines are measured on.
The same 3,000 dollars in the same ChatGPT campaign passes one test and fails the other. Nothing about the campaign changed.
Why this is not a substitution buy
A substitution buy moves demand that already exists from one capture point to another. Moving budget from Google search to Bing search is substitution: the same person, the same query intent, a different auction, and comparing cost per acquisition across the two is fair.
ChatGPT Ads is not that, for three documented reasons.
There is no query to substitute. Targeting is by context hints matched against conversation meaning. There are no keywords, no match types and no search terms report, so there is no unit of demand you can point at and say it moved.
The audience is not the same audience. Ads serve to Free and Go plan users. Paid plan users, Plus, Pro, Business, Enterprise and Edu, do not see ads at all, and users on the free tier can switch to an ads-free Free plan or turn off personalisation, per OpenAI's Ads in ChatGPT documentation. Whatever this reaches, it is not a clean overlap with your search audience.
The moment is different. The ad appears below a response that has already answered the question, labelled Sponsored and visually separated from it. On search, the ad sits ahead of the answer. Those are different positions in a decision, and treating them as interchangeable inventory is the error.
If it is not substitution, then a like-for-like cost per acquisition comparison against search is not a fair test, and a budget line that forces that comparison will produce a kill decision regardless of what the channel actually did.
What it can be evaluated on instead
Set the criteria before the money moves. Three that hold up.
Did it serve at all, in your category? Financial services, health services and legal services are allowed only in the US and only case by case as of the 10 September 2026 ad policies. Political and contested social issue ads are disallowed. Serving markets reported are the US, UK, Canada, Australia, New Zealand, Japan and South Korea. A test that establishes your category and geography actually clears has produced a real answer even if it produces no leads.
Did qualified pipeline appear, at any cost? Not cost per lead against search, but whether a lead arrived that sales agreed was qualified, and what it looked like. Three good conversations from an experimental line is a result. Three good conversations measured against a search cost per acquisition target is a failure.
What did the creative test teach you? With no search terms report and no per-hint reporting, the ad and the landing page are most of what you can move. A test that produces a working headline and a landing page pattern has produced an asset.
What to tell a CFO
The pitch that works is not "this is the future of search". It is a bounded option purchase. Say this, in this order.
The commitment is genuinely small and genuinely stoppable. Minimum daily budget is 25 USD, with no minimum lifetime commitment. The 200,000 USD pilot minimum is historic and no longer applies. A 90-day test at 50 USD a day is 4,500 USD and can be paused any day.
Billing is postpay, which is a cash-flow argument in your favour. You are invoiced after delivery. Account setup needs a legal business name, a business tax ID, a payment method, a logo and favicon and business verification, so finance is involved from the start rather than discovering a card charge later.
Say plainly what you are buying, which is information. The honest framing is that you are buying an answer to whether this channel reaches your buyers, six to twelve months before the answer becomes common knowledge and the cost of entry stops being 25 dollars a day. If the answer is no, the test cost less than one conference booth.
Say plainly what you cannot promise. No cross-advertiser benchmarks exist. OpenAI's advertising terms, updated 25 August 2026, state that OpenAI does not guarantee the performance, reach, frequency or results of campaigns, and that fees are determined by OpenAI's own measurements. A CFO will find that out eventually. Finding it out from you is worth more than the discomfort of saying it.
Name the reversal condition. "If by day 60 we have not had a single sales-accepted lead, we stop and report that as the finding." A test with a written stopping rule is a different object to a budget request, and it is the version that gets approved.
The counter-argument we take seriously
Everything above can be read as a way of avoiding accountability: move the money to a line where nobody measures anything and the test can never fail.
That is a real risk and the stopping rule is the answer to it. An experimental line without a written kill criterion, a named owner and a reporting date is not an experiment, it is a place budget goes to avoid scrutiny. The argument here is for a different measurement standard, not for the absence of one.
There is also a case for search funding that we will concede in one situation: if your search account is capped by budget rather than by demand, and you are genuinely choosing between two capture channels for the same buyer, then the comparison is fair and the objection above mostly falls away. That describes fewer accounts than claim it.
What we cannot tell you
- Whether ChatGPT Ads takes demand from search. No overlap or incrementality study has been published by anyone, and no cross-platform identity signal exists to run one.
- What share of free-tier users see ads. Users can opt out and choose an ads-free Free plan. Take-up is unpublished.
- What the channel costs at scale. The three to five dollar CPC and 60 dollar default max CPM figures circulating in 2026 are reported from agency write-ups, not published by OpenAI as floors.
- What a reasonable test budget is. We have no first-party data. InPromptAds runs no client campaigns, so the 50 dollars a day figure above is an illustration of the minimums, not a benchmark.
Quick answers
Should ChatGPT Ads budget come out of my search budget? Generally no. Money moved from search is judged on incremental cost per acquisition against search within one cycle, and ChatGPT Ads is not a substitution buy, so that comparison is not a fair test.
Where should it come from instead? Brand, display, or an explicit experimental or innovation line. Those lines are measured on reach, efficiency and learning over a quarter, which matches what this channel can currently evidence.
What is the smallest credible test? The platform minimum is 25 USD a day with no lifetime commitment. A bounded test of 60 to 90 days with a written stopping rule is the shape that survives internal review.
What does a CFO most want to hear? That the commitment is small, stoppable and postpay, that the deliverable is information rather than a pipeline promise, and that there is a named date and condition on which you will stop.
Is there a case for funding it from search? Yes, in one case: if your search account is capped by budget rather than demand and you are choosing between two capture points for the same buyer. That is rarer than it is claimed to be.
What should I promise about results? Nothing quantitative. OpenAI's advertising terms state no guarantee of performance, reach, frequency or results, and no cross-advertiser benchmarks are published to forecast from.
Sources
| Claim | Source | Tier |
|---|---|---|
| No guarantee of performance, reach, frequency or results; fees determined by OpenAI's measurements | OpenAI Advertising Terms, updated 25 August 2026 | Confirmed, primary |
| Ads serve to Free and Go plan users; paid plans are ad-free; users can choose an ads-free Free plan or disable personalisation | OpenAI Help Center, Ads in ChatGPT, 2026 | Confirmed, primary |
| Ads appear below the response, labelled Sponsored and visually separated | OpenAI Help Center, Ads in ChatGPT, 2026 | Confirmed, primary |
| Financial, health and legal services restricted to the US and case-by-case approval; political and contested social issue ads disallowed | OpenAI Ad Policies, updated 10 September 2026, v1.6 | Confirmed, primary |
| Minimum daily budget 25 USD, no minimum lifetime commitment | OpenAI Ads Manager documentation, 2026 | Confirmed, primary |
| Postpay billing; account setup requires legal business name, tax ID, payment method, logo, favicon, verification | OpenAI Ads Manager documentation, 2026 | Confirmed, primary |
| 200,000 USD pilot minimum is historic | Trade press, 2026 | Reported |
| Serving markets US, UK, Canada, Australia, New Zealand, Japan, South Korea | Trade press and agency write-ups, 2026 | Reported |
| Three to five dollar CPC and 60 dollar default max CPM guidance | Agency write-ups citing OpenAI guidance, 2026 | Reported |
| ChatGPT Ads is not a substitution buy and should not be funded from search | InPromptAds | Inference, ours |
Related reading
- What a Google Ads Manager Gets Wrong in Their First ChatGPT Campaign
- What Do ChatGPT Ads Cost?
- Who Should Not Buy ChatGPT Ads Yet
- Budget calculator
- Daily Budget or Total Budget: Which One Controls Your Spend
Changelog
12 September 2026, v1.0. First publication. Argues the funding line determines the evaluation standard, and sets out the not-a-substitution-buy case with a CFO script and a stopping rule.
Field kit
Tools
Site
OpenAI (primary)
Keshav studies how AI systems retrieve, verify, and cite brand information. At InPromptAds, he leads source research and turns platform documentation into practical guidance for advertisers.
