How to Advertise on ChatGPT: Approval, Budget and Reporting for Agencies

Yes, an agency can run ChatGPT ads for clients. OpenAI's self-serve Ads Manager at ads.openai.com tells anyone managing several advertisers to create a separate ad account for each, invite staff as Admin, Member or Viewer, and switch accounts on one login. The safe route is a client-owned account, written approval of creative and budget, and a capped test.

Two caveats shape everything after that. OpenAI's help centre still calls the product Ads Manager Beta, so its rules change often, and every claim below names the page it came from. And since 30 October 2026 creative enhancements are on by default for self-serve campaigns, which lets OpenAI rewrite, translate and re-point an approved ad without showing you each version. The agency-specific work is the part OpenAI's pages leave out: how the client signs off, how the first budget is capped, and how results get reported without overselling.

Follow me on Instagram@sagnikteaches

What OpenAI documents, and what to confirm before you promise anything

Sort every statement you are about to make to a client into one of three bins. It keeps proposals honest while the product moves.

Connect on LinkedInSagnik Bhattacharya

There is also a promotion on OpenAI's ads page: when I read it, new advertisers were offered a $500 ad credit for spending $500, with the offer ending 31 December. Treat that as something to check on the day you create the account, not as a plan.

Subscribe on YouTube@sagnikteaches

Step one: the client approval pack, before any account exists

Agencies lose most arguments after launch, over things nobody wrote down. A one-page approval pack, signed before the account is created, prevents them. Fill it in for each client; an illustrative version for a small bathroom-fittings retailer follows.

CLIENT APPROVAL: CHATGPT ADS TEST
Advertiser (legal entity): [client company name], billed to client
Account owner: client (Admin); agency staff invited as Member
Landing page(s): /bathroom-fittings (OAI-AdsBot may visit it to check policy)
Objective: clicks to the landing page; conversion = enquiry form sent
Test budget: $1,500 campaign-total budget over 30 days (about $50 a day)
Stop rule: pause if $300 is spent and the form event has not fired once
Creative sign-off: client approves every title and body before launch
Creative enhancements: opted out for this campaign (Manage settings)
Claims allowed: only those on the client's own website
Reporting: weekly, same five figures; call at day 14 and day 30
Out of scope: Plus/Pro/Business users (they see no ads); chat transcripts

Each line answers a question that otherwise arrives mid-campaign. The stop rule matters most: without it, a conversion event that silently fails will let money run for weeks. The enhancements line comes a close second, because with enhancements on, the wording the client approved is only the starting point for what people see. For why that tracking check comes first, set up conversion tracking before you let AI spend your ad budget covers it.

Step two: account ownership, billing and the brand review

OpenAI's documentation describes an ad account as the owner of campaigns, creative, feeds, audiences and conversion sources. Three names live on it: a legal name used for billing and business verification, an internal account name shown in Ads Manager, and a public brand name shown in ads. OpenAI's setup guide says the country or region, currency and time zone cannot be changed after the account is created, so decide them with the client's finance team, not on a whim.

Two facts shape the agency set-up. First, OpenAI's availability page says that to use self-serve Ads Manager, the legal entity that will advertise and be billed must be based in a listed market, so a client's registered entity decides eligibility, not your agency's location. Second, an account cannot serve ads until brand review is approved. OpenAI's help centre says a brand-name change triggers a fresh brand review and pauses serving until it is approved, and a legal-name change restarts business review and may pause serving; only the internal account name can be edited without a review. Do not rename the brand during a live test.

Access is the question agencies usually guess at, and OpenAI's account-setup guide answers it. Create a separate ad account for each advertiser. The client should create it, or at least hold the Admin role, so the account survives a change of agency. An Admin invites people from Settings, then Users, then Invite, choosing Admin (full access), Member (create and manage campaigns) or Viewer (view only). Your staff then sign in with their own OpenAI logins and switch between the client accounts they belong to. Two limits matter as you grow. Once a login belongs to 10 or more ad accounts it cannot create another itself, though an Admin can still invite it to more. And some brand-new self-serve accounts can advertise only in their home country until identity verification is approved and enough spend has run there, so a client that sells abroad should hear that before the plan promises it.

The documentation also notes that account-wide spending limits are available only for accounts on postpaid invoice billing, and that reading or setting them needs billing permission. For most small clients on card billing, your cap therefore lives at campaign level, which is why the approval pack states the budget in full. Card details go in under Billing before any campaign can deliver, and OpenAI notes that a card may show a temporary authorisation hold, separate from ad charges, so warn the client's bookkeeper.

Step three: check the client's category against the ad policy

OpenAI's ad policy says ads during the initial period are primarily limited to consumer verticals such as lifestyle and household goods, local services, travel and experiences, and digital products or education. Financial services, healthcare and legal services may be approved case by case for approved advertisers, sometimes with proof of licensure, and only in certain markets; the policy lists them with their own conditions. Disallowed categories include political advertising, gambling, alcohol and tobacco, adult content and others. The policy also prohibits ads for individual job listings or homes for rent or sale, which catches property and recruitment clients out, and says ads or landing pages its systems cannot review are not eligible to run.

Run a quick screen for each client:

1. Is the category listed as allowed, restricted or disallowed? Read the current policy page, not a summary.
2. Does the landing page contain anything disallowed? The policy says assets and landing pages are reviewed end to end.
3. Do any claims in the copy promise outcomes? The policy bars misleading claims about capabilities, pricing and outcomes.
4. Does the creative imitate ChatGPT's interface? The policy prohibits ads that could be mistaken for the product itself.

For legal clients in particular, ChatGPT ads for law firms: can you advertise, and is it allowed? goes through the licensing conditions.

Step four: build a capped test, with the arithmetic shown

OpenAI's budgeting documentation separates the objective, the campaign budget, the billing event and the bid strategy. The options for an agency running a first test are few.

The API documentation says new resources start paused, so setup can be finished before delivery begins, and that a conversion objective needs a conversion event set up first. Ads Manager's sign-up flow is less forgiving: OpenAI says ads created during sign-up can go live once they pass review, so pause anything built before the approval pack is signed. The minimum daily budget depends on the billing currency; OpenAI's table listed $25 a day for a USD account when I checked.

For a first test, use the clicks objective. It needs no conversion history, and it tells you whether anyone responds. Then pick the budget type with the client's cap in mind. OpenAI describes a daily budget as an average across a Sunday-to-Saturday week: one day can spend up to twice the figure and the week up to seven times it. It calls a campaign-total budget the stricter control for total spend; that type pairs with a fixed click bid, because the maximise-clicks strategy needs a daily budget. OpenAI suggests newcomers start with a daily budget so they can watch delivery, which is fine if you check spend every few days, but for a client who signed a hard $1,500 cap the campaign-total budget is the honest choice. One more trap: on an account's first campaign the start and end dates may only appear after publishing, and OpenAI advises against publishing just to reach them, so put day 30 in your own calendar too.

The sum below is a hypothetical, with the cost per click left for you to fill from the real account, because OpenAI's own FAQ says the platform has no performance benchmarks yet across advertisers or industries.

Budget: $1,500 campaign total (about $50 a day x 30 days)
If cost per click turns out to be $2.00: 1,500 / 2.00 = 750 clicks
If 4% of clicks send the enquiry form: 750 x 0.04 = 30 enquiries
Cost per enquiry: 1,500 / 30 = $50

Same budget at $4.00 a click: 375 clicks, 15 enquiries, $100 each.

The point of showing both is that the cost per click alone halves or doubles the result. Tell the client the test exists to discover the real figures, and agree in advance what cost per enquiry would justify continuing. For comparison with a retail cost view, ChatGPT ads cost for a small store sets out a test budget in that setting.

Step five: write the ad to the field limits, then get it signed off

OpenAI's ad documentation sets the creative limits: a title of 3 to 50 characters and a body up to 100, with a destination URL that its ad crawlers can reach. That is short, so each word must work. An illustrative pair of options for a bathroom-fittings retailer:

Title (46): Bathroom taps and fittings, fixed delivery fee
Body (72): Compare brass and chrome taps, see sizes and finishes, and order online.

Title (39): Replace a tap without calling a plumber
Body (68): Step-by-step fitting guides with every tap, plus clear measurements.

Count the characters rather than trusting a draft, because AI tools miscount; paste each line into a spreadsheet and check it with =LEN(A2) before anything goes to the client. You can use an AI assistant to produce a dozen variants, then screen them against the policy. A prompt that helps:

Write ten ad variants. Title: 50 characters maximum. Body: 100 maximum.
Use only these facts about the client: [list from client's website].
Do not promise results, savings or guarantees. Do not use comparative claims
about other companies. Do not imitate a chat reply. Return each with its
character counts.

Then check each against the client's website: every fact must appear there. Send the shortlist to the client with the approval line "approved for launch" next to each. Sign-off by email is enough as long as you keep it with the campaign record.

Next, deal with creative enhancements, because they can undo that sign-off. OpenAI's help page says they are on by default for new and existing self-serve campaigns from 30 October 2026, and that they can write new headlines and descriptions from the conversation and the landing page, translate the ad text, resize images and point the destination URL at a different page of the site. There is no preview of each version. Where the client must approve exact wording, opt out under Manage settings in Ads Manager; a campaign-level setting overrides the account default, so check both levels. If the client is relaxed about variants, record that decision in the approval email instead, and make sure every page an enhancement could pick is one the client is happy to send people to.

Step six: targeting that matches what the client actually sells

OpenAI's targeting documentation covers location (country, region or market, with up to 2,500 IDs per list), platform (OpenAI's campaign guide lists Android app, Android web, desktop web, iOS app and iOS web) and custom audiences, plus context hints that describe when an ad might be useful. For a small client, three decisions are enough:

• Location. Only where the client can deliver. A service firm that works in one region should not buy a whole country; use the geographic lookup to pick the region.
• Platform. Start with all, then split by platform only after you have clicks to compare.
• Context hints. Short descriptions of the situations when the product is useful, such as "replacing a dripping kitchen tap". They guide placement rather than act as strict keywords, so write them as situations, not search terms.

Custom audiences are rarely worth it on a first test: OpenAI's documentation says inclusion and bid-multiplier audiences need at least 25,000 matched users, which most small client lists will not reach. An exclusion list has no such minimum, so excluding existing customers is a practical use.

Step seven: tracking, before the first dollar is spent

OpenAI documents a measurement pixel for web pages, a server-side Conversions API and an image tag for sending events without JavaScript. Create the pixel in Ads Manager, add it to the client's site, and fire a test event before launch. Agree what counts as a conversion; for the example client that is the enquiry form being sent, not a page view.

Know the reporting defaults, too. The documentation says insights default to a 30-day click window and a 1-day view window, with conversions including both click-through and view-through. A client comparing the report with their own analytics will see different totals, because their analytics tool credits conversions by its own rules, often to the last click, and cannot see view-through conversions at all. Explain this at the start, not after the first complaint.

Reporting to the client without overselling

OpenAI says advertisers get aggregated, non-identifying reporting such as views and clicks, not chats or personal details. Build the weekly report around figures it provides and your own analytics, and say plainly what you cannot see. A template, with illustrative week-two numbers:

Under the table add three lines: what changed this week, what you will test next, and one sentence on limits ("this shows ad clicks and form sends, not conversations inside ChatGPT"). For wider reporting habits, how marketing agencies automate client reporting with AI shows how to draft the narrative. Check the arithmetic yourself: click-through rate here is 140 divided by 12,400, about 1.13%.

The sign-off email that keeps the record clean

Most clients will not read a form closely, but they will reply to a short email. Send the creative and the cap in the body, not as an attachment, and ask for one word back. An illustrative version:

Subject: ChatGPT ads test: please reply "approved"

Hi [first name],

Here is what goes live on Monday if you approve:
- Budget: $1,500 in total over 30 days (about $50 a day), set as a hard cap
- Two ads (titles and bodies below), both using only text from your site
- The wording stays exactly as below: OpenAI's automatic variations are off
- Landing page: the bathroom-fittings page; we count a sent enquiry form
- We pause on the day we have spent $300 with no enquiry recorded

Reply "approved" and we will launch. Reports go out every Friday.

[your name]

Keep the reply with the campaign folder. If a dispute arises about spend, creative or results, that single line settles which version the client agreed to.

Deciding at day thirty: continue, change or stop

Agree the decision table before the test starts, so the numbers choose rather than the mood on the call. The thresholds below are illustrative; set them from the client's own margins.

A single test is a small sample, so write the caveat into the review: a month of results from one client in one category tells you about that client, not about the channel in general.

Running several clients without mixing them up

An agency with five clients on the channel needs discipline more than tools. Use a naming pattern across everything: client code, objective, month. For example, "BF-clicks-2026-11" for the bathroom-fittings clicks campaign. Keep each client's approvals, creative, pixel details and reports in one folder named the same way. Invite each person to each client account with the lowest role that does the job: Member for whoever builds campaigns, Viewer for an account manager who only reports. OpenAI's API documentation says each Ads API key is scoped to one ad account and tells partners to keep keys in a server-side secret manager, and the same instinct applies to people: never share a login, and remove staff access the day someone leaves the team or the client ends the contract. When a client moves to another agency, hand over the account itself, since the client owns it, along with the pixel IDs and a written list of what is running.

Explaining the channel to a sceptical client

Clients often ask whether ChatGPT ads are "worth it yet". An honest answer has three parts, and AI can help you phrase them, but the facts come from you. First, it is a new, beta-stage product whose rules and availability change. Second, ads reach people on the Free and Go plans, not Plus, Pro or business-plan users, so the audience is a slice. Third, the cheapest way to learn is a capped test with a written stop rule, which risks a known amount. A line you can adapt: "We are not betting the budget on this. We are spending $1,500 to find out whether people who ask ChatGPT about bathroom fittings click and enquire, and we will stop earlier if the tracking shows nothing."

If the client wants a comparison, put the test next to the cost per enquiry from their best current channel, and say what difference would change the decision. For the client who also wants organic visibility inside answers, point them to how to monitor brand mentions in ChatGPT week by week, because that work is separate from anything you buy.

Where a first ChatGPT ads test goes wrong

• The tracking event never fires. Symptom: clicks, no conversions after $300. Cause: the pixel is on the wrong page or blocked by a consent tool. The stop rule catches it.
• The landing page fails review. OpenAI's crawler page says OAI-AdsBot visits pages submitted as ads to check them, and its ad-creation guide says landing pages must not block OAI-AdsBot or OAI-SearchBot. Check that the firewall allows both, and that the page loads without a login.
• The live ad says something the client never approved. Creative enhancements were left on. Opt out before launch, or get written agreement to variants.
• The category was not allowed. A restricted category approved for the wrong market, or one that needs proof of licence. Screen first.
• The client expects chat recommendations. Ads are separate from answers. Reset expectations in the approval pack.
• The audience is smaller than assumed. Only Free and Go users see ads, so a client whose buyers mostly pay for Plus or a business plan is reaching fewer of them.

Launch checklist and the first thirty days

1. Approval pack signed; the client holds Admin on its own account and has invited your staff.
2. Brand name and icon approved; do not rename during the test.
3. Pixel installed; test event seen in Ads Manager.
4. Campaign created paused; budget, dates and location checked line by line.
5. Creative approved in writing; character counts verified.
6. Creative enhancements opted out at account and campaign level, or variants approved in writing.
7. Activate; check delivery after 24 hours.
8. Days 7, 14, 21 and 30: weekly report, stop rule checked, notes kept.
9. Day 30: one-page review with the client: continue, change or stop.

The decision at day 30 should compare cost per enquiry with what the client already pays through other channels, not with hopes. If you also test Google's AI surfaces, ads in AI Overviews: a pre-spend checklist gives the same discipline for that channel.

Questions agencies ask about ChatGPT ads

Which ChatGPT users can see the ads?

OpenAI's help page says ads may appear on the Free and Go plans, while Plus, Pro, Business, Enterprise and Edu accounts do not get ads. Ads are not shown to accounts identified as under 18, in Temporary Chats or near sensitive topics such as health or politics, and Free users can switch to an ad-free option with lower limits. Tell clients the reachable audience is a slice of ChatGPT users.

Can a client's ad influence what ChatGPT answers?

OpenAI states that ads do not influence the answers ChatGPT gives, run on separate systems from the chat model, and are labelled. Advertisers cannot shape or rank responses. Say this plainly to clients who assume an ad buys a recommendation inside the answer itself.

What can I report to a client?

OpenAI says advertisers receive aggregated reporting such as views and clicks, not conversations. Its documentation lists impressions, clicks, conversions, spend, click-through rate and cost per click, with selectable attribution windows. Keep reports to those figures and your own site analytics, and never imply you can see user chats.

How long should a first test run?

Long enough to spend the whole test budget and collect enough clicks to judge, usually several weeks for a small budget. Set the stop rule in advance, such as a fixed spend with a minimum number of clicks, and agree it with the client before launch so nobody judges results on day three.

Further reads

• ChatGPT Ads for Realtors: Access, Cost and Fair-Housing Rules — Access and rules for another regulated-adjacent category.
• Best AI Tools for Advertising Agencies: Copy Variants, Mock-Ups, Reporting — Other AI tools an agency may add to its stack.
• Search Console and AI Overviews: What Agencies Can Report to Clients — What can be reported from Google's AI features.
• Can AI Manage Your Google Ads Without an Agency? — Where AI-run ads work and where they do not.
• SEO for AI Search: A Client Audit Checklist for Agencies — Audit a client's AI-search readiness alongside paid tests.
• How Much Should You Spend on AI-Run Ads Before Judging Results? — How much to spend before judging any ad test.
• How to Buy Ads on ChatGPT: A Small Business's First Budget — Eligibility gates, account set-up, a four-week first budget with the sums worked out, and the Ads Manager settings to check before you spend.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OpenAI Ads documentation (developers.openai.com/ads), OpenAI Ad policies, OpenAI crawler documentation, OpenAI Help Centre (Ads in ChatGPT; Ads Manager Availability; Ads Manager Beta account setup; Creating campaigns; Create ads; Creative enhancements; Ads FAQ) and ads.openai.com, all read on 4 November 2026. Ads Manager is a beta product; check the pages for changes.