The Nine Questions Enterprise Procurement Asks Before Approving ChatGPT Ad Spend
By Ansh Khandelwal 10 min read
In brief
Documented answers to nine procurement questions about ChatGPT Ads, including data, contracts, billing and unresolved gaps.
Last verified: 12 September 2026 | Version: 1.0 | Next scheduled review: 12 December 2026
This is written to be forwarded. Each question is the one procurement actually asks, the answer is the documented position with its source and date, and where OpenAI publishes nothing, that is the answer given rather than a paraphrase of something adjacent.
Nothing here is legal advice or a compliance opinion. Your legal and privacy teams should read the underlying documents.
1. What happens to our customer data?
The documented answer. The Ad Tools Data Processing Addendum, effective 19 August 2026, sets OpenAI and the customer as independent controllers for most ad tools processing. OpenAI acts as processor only for what the DPA calls Restricted Processing, meaning audience matching and opt-out data. Data categories include email addresses, phone numbers, cookies, online identifiers, device and browser information, and event and transaction data.
The Ad Tools Terms, published 24 August 2026, restrict input to first-party data collected directly from your users, with no third-party enrichment, and prohibit a named list of sensitive categories: health, sexual orientation, race or ethnicity, religion, financial distress, genetic data, disability status and children's data.
What to flag. Independent controller status is the line privacy teams need to see before they model this as a standard processor arrangement, because it is not one. EEA and Swiss data routes through OpenAI Ireland Limited; UK data through OpenAI OpCo, LLC under the UK Standard Contractual Clauses. Sub-processors are published on OpenAI's ad tools sub-processor list.
2. What controls do we have over adjacency and brand safety?
The honest answer. Effectively none, and the reporting to verify it does not exist.
There is no exclusion layer of any kind: no negative keywords, no topic blocklist, no category exclusions. There is no adjacency reporting, no conversation data and no search terms report, so you cannot see what conversation your ad appeared beside either before or after the fact.
What to flag. Say this plainly. A brand safety reviewer used to a placement report and a blocklist will assume both exist. What you can offer instead is structural: ads are labelled Sponsored and visually separated from the response, ads run on separate systems from the chat model and advertisers cannot shape or alter responses, and OpenAI's ad policies disallow whole categories including political and contested social issue advertising. Those are real mitigations. They are not controls in your hands.
3. What contract are we actually signing, and can we negotiate it?
The documented answer. OpenAI publishes the Advertising Terms, updated 25 August 2026, the Ad Tools Terms, published 24 August 2026, the Ad Tools DPA, effective 19 August 2026, and the Ad Policies, updated 10 September 2026 at version 1.6. These are online terms accepted by use.
Key provisions procurement will find: OpenAI does not guarantee performance, reach, frequency or results. Liability is capped at the total amount paid in the six months before the event giving rise to it, excluding gross negligence, wilful misconduct and indemnification. The customer indemnifies OpenAI for claims arising from the ads and content it provides. Either party may terminate on 30 days' notice for any reason, and OpenAI may terminate immediately for non-payment, policy violation or legal compliance. Terms changes take 30 days' notice except for security or legal issues.
What to flag. Nothing published indicates a negotiated enterprise agreement path. If your procurement policy requires a signed MSA, an uncapped data protection indemnity, or a liability cap above six months of spend, that is a blocker to raise before the business case, not after.
4. How are we invoiced, and what are the payment terms?
The documented answer. Billing is postpay. Account setup requires a legal business name, a business tax ID such as an EIN, a payment method, a logo, a favicon and business verification. The Advertising Terms state that invoicing is available on request, that overdue amounts accrue a 1.5 percent monthly finance charge, that fee disputes must be raised within 30 days, and that adjustments are provided as non-transferable credits or makegoods usable before termination rather than as cash refunds.
What to flag. Postpay generally helps the business case. The 30-day dispute window and the credits-only remedy are the two clauses finance will want to know about in advance, because both are shorter and narrower than a typical negotiated media agreement.
5. Can we audit what we were charged for?
The honest answer. Not independently. The Advertising Terms state that metric-based fees are determined by OpenAI's measurements and data. There is no published audit right for the customer, and no third-party verification, ad server log access or independent impression measurement exists for this inventory.
What to flag. For an enterprise used to third-party verification on programmatic buys, this is a genuine downgrade and should be stated as one. The available substitutes are the 30-day dispute window and your own CRM reconciliation. Reseller record-keeping obligations exist in the terms but are not a customer audit right.
6. Can we defend the numbers to finance?
The documented answer. Measurement runs through the OpenAI pixel plus a Conversions API. Reporting covers impressions, clicks, spend, CTR, average CPC, average CPM and conversions at campaign, ad group and ad level. Configurable click-through attribution windows shipped in 2026.
What to flag. Three absences shape what you can claim. There are no cross-advertiser benchmarks, so a forecast has no external reference point. There is no per-hint reporting, so you cannot attribute performance below the ad group. And if you also run Google AI Mode or Microsoft ads, two attribution systems will both claim the same conversion, so anything leaving the marketing team should be reconciled against the CRM first. Say this before someone builds a board slide on platform totals.
7. Are we even eligible to advertise?
The documented answer. OpenAI's ad policies, updated 10 September 2026, restrict financial services, health services and legal services to the US and approve them case by case, with legal services limited to licensed attorneys. Political and contested social issue advertising is disallowed. Restrictions also apply to housing and job listings, and to unsubstantiated wellness claims. Prohibited categories include adult content, alcohol and tobacco, counterfeit goods, gambling, recreational drugs and scams.
What to flag. Eligibility is a gate, not a risk to manage. If your category is in the restricted list, run the eligibility question before the budget question. Case-by-case approval means a decision you cannot predict.
8. Which legal entity holds the account, and does that work for us?
The documented answer. Advertiser account eligibility follows the country of the legal business entity. OpenAI now lists 52 self-service countries, including India and much of Europe. The UAE is not on the current Ads Manager availability list. Confirm buyer-location targeting inside the live campaign picker.
What to flag. For a group with entities in several countries, this becomes a real structuring question: which subsidiary signs, which entity is invoiced, and whether the intra-group flow of customer data to the entity holding the ad account is itself documented. The DPA's independent controller structure makes that second question sharper than it would otherwise be.
9. How do we get out?
The documented answer. Either party may terminate on 30 days' notice for any reason. There is no minimum lifetime commitment and the daily budget minimum is 25 USD, so spend can be stopped at any time. Credits and makegoods must be used before termination, which means unused credit is lost on exit.
What to flag. Exit is cheap, which is the strongest single argument in this document. The precedent worth citing is Perplexity, which launched an ad product and then exited advertising entirely in February 2026 to go subscription-only. Build nothing expensive and single-purpose on top of this channel: no bespoke infrastructure, no headcount whose whole role is one surface, no annual commitment. Keep the exit as cheap as the contract makes it.
What we cannot tell you
- Whether a negotiated enterprise advertising agreement exists. Nothing published indicates one, and no advertiser has reported obtaining one.
- Whether third-party measurement verification is planned. No roadmap, partner or accreditation is published for this inventory.
- What proportion of users see ads. Ads serve to Free and Go plan users only, users can choose an ads-free Free plan or disable personalisation, and take-up of both is unpublished.
- What a realistic budget or return looks like. No cross-advertiser benchmarks exist, and InPromptAds runs no client campaigns, so we have no first-party figures to offer.
- How review decisions are made in restricted categories. Case-by-case approval is documented, the criteria are not.
Quick answers
Is OpenAI a processor for our ad data? Only for Restricted Processing, meaning audience matching and opt-out data. For most ad tools processing, the Ad Tools DPA effective 19 August 2026 makes OpenAI and the customer independent controllers.
Can we block our ads from appearing beside certain topics? No. There is no exclusion layer, no topic blocklist and no adjacency reporting. Mitigation comes from OpenAI's own category prohibitions and the Sponsored label, not from controls in your account.
Can we negotiate the terms or get an MSA? Nothing published indicates a negotiated path. The terms are online and accepted by use, with liability capped at six months of spend and a customer indemnity. Raise this before the business case if your policy requires otherwise.
Can we verify what we were billed for? Not independently. Fees are determined by OpenAI's measurements, no customer audit right is published, and no third-party verification exists. You have a 30-day dispute window and your own CRM reconciliation.
Which entity should hold the account? One registered in a country where advertiser accounts are permitted. OpenAI currently lists 52 self-service countries, including India; the UAE is not currently listed.
How hard is it to stop? Not hard. Either party can terminate on 30 days' notice, there is no lifetime commitment, and the daily minimum is 25 USD. Unused credits and makegoods are lost at termination.
Sources
| Claim | Source | Tier |
|---|---|---|
| Independent controller structure; processor only for Restricted Processing; data categories; EEA/Swiss and UK routing and UK SCCs; sub-processor list | OpenAI Ad Tools Data Processing Addendum, effective 19 August 2026 | Confirmed, primary |
| First-party data only, no third-party enrichment, Prohibited Data list, advertiser holds notices, consents and legal bases | OpenAI Ad Tools Terms, published 24 August 2026 | Confirmed, primary |
| No guarantee of performance, reach, frequency or results; fees determined by OpenAI's measurements; invoicing on request; 1.5 percent monthly finance charge; 30-day dispute window; credits and makegoods only; liability capped at six months of fees; customer indemnity; 30-day termination either party; 30 days' notice of terms changes | OpenAI Advertising Terms, updated 25 August 2026 | Confirmed, primary |
| Category restrictions and prohibitions; case-by-case approval for financial, health and legal services; political and contested social issue ads disallowed | OpenAI Ad Policies, updated 10 September 2026, v1.6 | Confirmed, primary |
| Ads labelled Sponsored, visually separated, run on separate systems from the chat model | OpenAI Help Center, Ads in ChatGPT, 2026 | Confirmed, primary |
| Measurement via pixel and Conversions API; reporting at campaign, ad group and ad level; configurable click-through windows shipped 2026 | OpenAI Ads Manager documentation, 2026 | Confirmed, primary |
| Postpay billing; setup requires legal business name, business tax ID, payment method, logo, favicon and verification | OpenAI Ads Manager documentation, 2026 | Confirmed, primary |
| Minimum daily budget 25 USD, no minimum lifetime commitment | OpenAI Ads Manager documentation, 2026 | Confirmed, primary |
| No exclusion layer, no adjacency reporting, no cross-advertiser benchmarks, no per-hint reporting, no customer audit right, no third-party verification | OpenAI documentation, by absence | Absent |
| Serving markets US, UK, Canada, Australia, New Zealand, Japan, South Korea; advertiser eligibility follows country of business registration; India and UAE on neither list | Trade press and agency write-ups, 2026 | Reported |
| Perplexity exited advertising in February 2026 and moved to subscription-only | Trade press, February 2026 | Reported |
| Reconcile against the CRM before reporting outside the marketing team | InPromptAds | Inference, ours |
Related reading
- Consent and the OpenAI Pixel: GDPR, UK GDPR and India's DPDP
- Where the ChatGPT Ads Budget Should Come From
- ChatGPT Ads Brand Safety
- How Postpay Billing Works on ChatGPT Ads
- ChatGPT Ads Eligibility: Who Can Actually Buy
Changelog
12 September 2026, v1.0. First publication. Answers the nine standard enterprise procurement questions against OpenAI's published August and September 2026 terms, and names the three with no documented answer: adjacency control, independent billing audit, and a negotiated enterprise agreement.
Field kit
Tools
Site
OpenAI (primary)
Ansh works across GEO strategy, B2B research, and execution. At InPromptAds, he translates new AI advertising products into clear operating advice, tests, and measurement questions for marketing teams.
