ChatGPT Ads for agencies: organising multi-client management

The constraint that shapes everything: one account per advertiser

OpenAI's documented model is simple and rigid: one advertiser account per advertising business. To manage several advertisers you create that many separate ad accounts.

There is no manager account yet letting an agency connect to multiple advertiser accounts, the way a Google Ads MCC or a Meta Business Manager does. OpenAI indicates the capability may be offered at some point: that is not a schedule commitment, and it is not something to build an offering on today.

The direct consequence for an agency with ten clients: ten accounts, ten billing profiles, ten payment methods, ten permission sets to keep current. That is not an execution detail, it is an operating cost to price before you sign.

Who creates the account: the client, not the agency

OpenAI's documentation is explicit here: agencies should not create ad accounts on behalf of their clients. The expected pattern is that the client creates the ad account, then invites the agency's users.

This is not administrative preference. The ad account goes through identity and business verification, tied to the registration documents of the advertising entity. An account created in the agency's name for a client's activity puts the verified entity out of step with the advertising entity, with a real risk of getting stuck in review.

There is a hidden benefit to respecting the pattern, and it is worth selling to the client rather than enduring: the client keeps ownership of the account, its history and its data. At the end of the engagement there is nothing to hand over. The painful "give me my ad account back" negotiation simply does not happen.

The full creation walkthrough is on our ChatGPT Ads account page.

What does exist: one address, several accounts

The good news, also documented: a single email address can access multiple ad accounts, as long as a role has been assigned in each one. The user then switches between them inside Ads Manager.

So your consultants do not need one address per client, or one browser profile per engagement. That is already a lot of friction removed.

But be precise about what the switcher is not:

• it is not pooled billing;
• it is not a consolidated view of all your clients' performance;
• it is not a duplication mechanism for copying assets between accounts;
• it is not centralised access delegation: each role is granted account by account.

Roles and permissions: the distinction to grasp first

Two levels of rights coexist, and confusing them blocks an entire setup.

Organisation level. Global Admin and Ads Admin are separate roles. A Global Admin manages the tenant, identity settings and users, but being a Global Admin does not automatically grant permission to create an ad account. The Ads Admin role has to be assigned explicitly to the people who need it.

Ad account level. Once the account exists, each user or group receives an ad account role that controls what they can do inside that specific account, such as viewing or managing campaigns.

Another useful point: ChatGPT workspace membership and Ads access are managed separately. Ads-only customers manage their users in the Global Admin Console, while access to a given ad account still has to be assigned inside Ads Manager.

For organisations that provision access from a directory, OpenAI documents SSO and SCIM setup for Ads Manager.

Billing stays with the client, and that is the better outcome

Each ad account carries its own billing. The documented mechanism runs on a payment threshold: the payment method on file is charged once unsettled spend reaches the account's threshold. Approved advertisers generally start with a low threshold, which may be raised over time based on payment history and other factors.

Two operational points to fold into your processes:

• A failed payment can interrupt delivery until it is resolved. An expired card kills a campaign as surely as an exhausted budget, and you will not necessarily be the first to hear about it.
• When the card is added, the client's bank may show a temporary $100 authorisation during verification. It is not a charge, but it is exactly the kind of alert that triggers an anxious Friday-evening phone call. Warn them.

The agency model of fronting media spend on its own card and rebilling later hits a wall here: it presupposes an account in the agency's name, which is precisely what the documented pattern advises against. Billing detail is on our ChatGPT Ads pricing page.

Country, currency and time zone: locked, per client

At each account's creation, three values are permanently locked: the country, the currency and the time zone. For an international agency that produces a heterogeneous estate of accounts that no later manipulation will realign.

Three concrete effects to anticipate:

• The applicable minimum daily spend varies by the account's currency, with OpenAI publishing a table of minimums. Two similarly sized clients will not have the same floor.
• The time zone carves up reporting days. Consolidating exports from accounts on different time zones without reprocessing produces wrong totals.
• A mistake on any of the three is only fixed by creating a new account, with the loss of history that implies.

Our recommendation: make validating those three fields a formal step of your onboarding, with written confirmation from the client. Five minutes there avoids a complete rebuild.

Consolidated reporting is your job

No multi-account view is documented. Ads Manager Beta reporting includes impressions, clicks, spend, CTR, average CPC, average CPM and conversions, in table view, chart view and CSV export, in daily or cumulative values across the chosen period.

So the CSV export becomes your reporting backbone. A simple method that fits in a spreadsheet:

1. Export daily values per account, over an identical period.
2. Add a client column and a currency column at import.
3. Convert to a reference currency before any aggregation.
4. Check each account's time zone before aligning dates.
5. Consolidate the click-through Conversions column only.

That last point is not a detail: view-through conversions, shown for some accounts on a fixed one-day window, are separate supplemental reporting that OpenAI says should not be added to conversions or used for core metrics like CPA. Adding them into a client dashboard means promising profitability that does not exist.

What this changes in your commercial proposal

Turn these constraints into clauses rather than discovering them mid-engagement:

• Client prerequisites: the client creates the ad account, passes business verification and registers a payment method. That is their task, to be scheduled before any announced launch date.
• Access: the client invites named consultants, one role per person. Plan the removal procedure at end of engagement.
• Media billing: spend is paid by the client, directly to OpenAI. Your fees are invoiced separately.
• Test threshold: the documented minimum daily spend is $25 per campaign, varying by currency. A five-euros-a-day test does not exist.
• Reporting: manual consolidation from CSV exports, with a production lead time you state up front.
• Scope of guarantee: none. See the next section.

What never to promise a client

The shortest and most important section on this page. On a beta channel a few months old, an agency's legal and reputational exposure sits in what it writes into proposals.

• Promise no placement, no impression volume, no outcome. OpenAI offers no such guarantee, and nobody can offer one on its behalf.
• Do not promise an industry benchmark CPC or CPA. OpenAI states explicitly that there is no reliable, generalized per-industry performance benchmark yet. The only official figure is a recommended starting CPC bid of $3 to $5, which is an entry point into the auction, not a cost forecast.
• Do not imply that OpenAI approves, recommends or certifies your agency. No partner programme is documented.
• Do not promise demographic targeting, lookalike audiences or retargeting: those functions are not documented.
• Do not promise a firm time to go live: ads go through review, and review turnaround carries no public commitment.

The onboarding process we recommend

A sequence that fits on one page and avoids most rework:

1. Qualify the advertiser's category against the ad policies, updated on August 10, 2026 and built as an allowlist: only explicitly permitted categories may run. See ChatGPT Ads policies.
2. Have the client create the account, with written sign-off on country, currency and time zone.
3. Track verification of identity and business through to completion before announcing a launch date.
4. Obtain named access and document who holds which role.
5. Wire up measurement before the first campaign: pixel, Conversions API or both, depending on the client's technical context. See pixel and conversion tracking.
6. Check the landing page, in particular that OAI-AdsBot and OAI-SearchBot are not blocked. Our ChatGPT Ads landing page checker tests it directly, and it is the most frequently fatal point.
7. Build the structure: campaign, thematic ad groups, ads. See ChatGPT Ads campaign setup.
8. Launch, then leave it alone for at least 24 hours before any delivery diagnosis.

Selling the service without overpromising

An agency has two honest angles on this channel, and they are enough.

The first is timing. The inventory is young, the documentation changes monthly, and most advertisers have tested nothing yet. An agency that already knows the mechanics on the day the channel fills up holds a real lead. That is a learning argument, not a performance guarantee, and it sells perfectly well as such.

The second is continuity with organic. The technical requirements that gate the advertising, starting with letting OpenAI's crawlers reach your pages, also gate visibility in the answers themselves. One workstream serves two objectives, which makes the spend easier to justify. The organic counterpart is on our ChatGPT SEO page.

What does not sell, on the other hand: a cost-per-lead promise, a volume commitment, or the idea that this channel replaces an existing budget. Nothing in the documentation supports any of it.

What is not publicly documented

Points we flag as unconfirmed for lack of an official source:

• The timeline for a multi-advertiser management account carries no commitment.
• The maximum number of ad accounts one email address can access is not quantified.
• The full list of ad account roles and the rights attached are not reproduced here.
• There is no documented agency partner programme and no certification status.
• Account verification and ad review turnaround carry no public commitment.
• No guarantee of delivery, volume or outcome is offered by OpenAI, and nobody can offer one on its behalf.