You do not need to learn ChatGPT ads. You need to make seven business decisions nobody should make for you, then delegate the campaign work. This list is for the owner who typed “who can just do this for me” into Google and meant it. Done-for-you should remove campaign work, not business judgment. Put these decisions into a one-page brief, and your provider can execute without guessing what your business needs.

1. Pick the customer action worth paying for

Start with one commercially meaningful outcome: a completed purchase or a qualified enquiry with a name, a need, and a way to contact the buyer. Not a chat started. Not a click. Not a mention inside an answer.

ChatGPT inventory is billed per 1,000 impressions or per valid click, never per conversion. The thing you buy from the platform is not necessarily the thing your business needs. If your brief treats engagement as success, your provider can deliver plenty of activity without proving demand.

Pick the hard action first, even if volume looks thin in week one. I want the reporting to distinguish a potential customer from someone having an interesting afternoon.

Your first step is one sentence:

We pay for X. We observe Y, but we do not optimize toward it.

Say you spend $20k a month on Google and want to test ChatGPT with $3,000. For a drain company, X might be “paid booking for a drain repair, with address and phone.” Y might be site visits and time on page. Those can help explain what happened. They do not replace the booking.

A sharp brief also names target ROAS or CPA and why it matters. If that target is still a guess, say so. Decision No. 4 covers the economics. For now, get agreement on what counts before anyone starts counting.

2. Name the offer and the buyer it actually fits

Give your provider one offer, one page that can fulfill it, and a clear description of the enquiries you will reject.

ChatGPT matches ads to conversation context rather than simply to a list of keywords you picked at midnight. A vague offer gives the manager less useful context to work with. “Plumbing services” leaves plenty of room for renters, DIYers, and job seekers. “Same-day drain unblock for homeowners in Austin, $149 callout credited to work” tells the buyer much more before the click.

Specificity is a filter, not just a copywriting preference. Your manager can describe the offer. Only you can say which jobs you actually want.

Your first step is to write three lines:

  • Offer: What happens, how pricing works, and where you deliver it.
  • Fit: The customer and problem the offer serves. For example, a homeowner with an urgent drain problem.
  • Reject: Out-of-area enquiries, DIY advice, or jobs whose economics do not work for you.

Do not make the provider discover those exclusions through paid enquiries. If landlords shopping three quotes are a bad fit for this particular offer, put that in the brief. A campaign manager should not have to infer your margins from which leads you complain about.

3. Set a media-spend limit separate from the management fee

There are two numbers here. Keep them separate: media spend goes to the platform; the management fee goes to whoever runs the work.

Self-serve buying on ads.openai.com opened May 5, 2026 without the earlier pilot spending minimums. That does not mean every campaign setting has no floor. The buying guidance lists a $25 per campaign per day minimum in the US and an OpenAI-recommended $3 to $5 starting maximum CPC.

The budget detail I would underline is less exciting: daily budgets are seven-day averages and can bill up to twice that amount on a single day. Set $100 a day and you can see a $180 day. An average daily budget is not a hard daily ceiling. Your card does not care what the proposal called it.

Your first step is to put these on separate lines:

  • Total test authorization: For example, $2,500 in media spend.
  • Average daily budget: For example, $85, with the potential daily variation acknowledged.
  • Approval rule: Only you can authorize more total spend.
  • Management fee: A separate amount, with the work it covers stated plainly.

Ask the provider how they will monitor the total authorization and when they will pause or seek approval. Do not accept an average-budget setting as the whole answer.

The groas monthly management fee is $1,499; platform spend is separate and paid to the platform. There is no percentage-of-spend fee quietly rising when the media budget rises. If a proposal blends media and management into one unexplained number, send it back.

4. Define the economics the manager must work within

A manager needs to know what a customer is worth before deciding what an enquiry can cost. Give them your margin per sale, your close rate from qualified enquiry to customer, and lifetime value if repeat purchases matter.

Without those numbers, a ROAS or CPA target is guesswork. That is why a useful brief includes margins and lifetime value, not just the acquisition target you would like to see in a report.

Take the draft economics of a home-services offer: a $90 lead cost looks different if one in three qualified leads closes at $1,400 a job. But revenue alone still does not settle the question. The margin matters. Cheap leads can lose money, and expensive ones can be worth buying.

Your first step is to give ranges and label the unknowns. For example:

Gross margin is about 60% on a $400 average order. Qualified enquiry to customer is around 20%, but that is a guess from last quarter. Treat $48 per qualified enquiry as a provisional first-order breakeven, not a proven target.

That $48 figure is a lead-cost estimate, not a cost-per-customer target. Keep the denominator visible. Otherwise two people can agree on “CPA” while talking about entirely different actions.

An honest unknown beats an invented target. Ask your provider to work within an agreed provisional ceiling and revisit it once the close-rate data is credible. Do not let a guess become policy because it looks tidy in a spreadsheet.

5. Approve the claims and destination before launch

Decide what the ads can promise, what they must never say, and which page has to fulfill the promise.

In a chat interface, your ad sits close to an explanation. A mismatch is hard to hide. If the ad promises same-day service and the page offers appointments in 48 hours, you have paid to create distrust. Better copy will not fix the scheduling problem.

This is business approval, not a request for you to rewrite every headline. Your provider can produce variants. You supply the boundaries those variants must respect.

Your first step is a three-part approval note:

  • Allowed claims: Two promises you can support with a page or policy.
  • Restricted language: Words such as “cheapest,” “guaranteed,” “number one,” “instant approval,” and “free.” Ban them unless you can substantiate and fulfill them.
  • Destination: One URL you control and can change within a day.

Check the offer and the page together. A truthful sentence can still mislead when the conditions are missing from the destination.

If you cannot fulfill it on that page, do not let anyone advertise it. Make that the approval standard, not whether the headline sounds punchy.

6. Keep account ownership and access explicit

Your business should own the advertiser account and retain the billing relationship. Delegating management is not the same as renting access to your own advertising history.

The setup includes your work email, legal business name, logo, billing, and verification. The account setup guidance describes verification and access steps that can take around 15 days. Do not confuse a provider’s campaign-build timeline with a promise that platform access will arrive on schedule.

One owner creates the account and invites others. If an agency manages the campaigns, the business should normally retain account ownership and billing. Put that in writing before the build starts, not during an awkward departure.

Your first step is to confirm two things before paying anyone:

  1. Your buyers are reachable. ChatGPT ads began as a US test on February 9, 2026, with availability expanding gradually. Ads appear on Free and Go, while Plus, Pro, Business, Enterprise, and Edu remain ad-free. Headline user counts are not your reachable audience. If your buyers mostly use enterprise seats, that distinction matters.
  2. Your access survives the relationship. Confirm who owns the account, who pays the platform, and what access the manager receives. Use the checks in Hiring Someone to Set Up ChatGPT Ads? Verify These 5 Things Before Sign-Off, including approved spend and tracking evidence.

If the provider wants to run your spend through their account, that is a no. You are hiring execution, not giving away the keys.

7. Decide who can pause, change, or expand the test

Write down who can stop spend, what the manager can change without asking, and when you will review the results. Otherwise every disappointing afternoon becomes a strategy meeting.

A new campaign needs enough stability to produce a useful read. The operator setup guidance recommends starting on CPC, building context around scenario and intent, and leaving the launch alone for 72 hours before making routine changes. That does not mean ignoring broken tracking or an unauthorized promise. Patience is for evaluating performance, not tolerating faults.

Your first step is to separate permissions from stop conditions:

  • The manager may: Adjust campaign execution and test approved copy within the agreed offer and spending authorization.
  • The manager may not: Increase the spending authorization, change the offer, or send traffic to a different page without your approval.
  • Pause and investigate: Tracking fails, an ad makes an unapproved claim, or spending exceeds what you authorized.
  • Review together: Set a date to judge the test against the customer action from No. 1.

Plan for a 14- to 21-day test, while keeping permission to stop a broken one. A bad result and a broken meter are different problems.

You do not need to install the tracking yourself. You do need evidence that it works before spend starts. The setup checklist includes pixel and server-side event checks, deduplication, and prelaunch verification. Ask the provider to show that the agreed conversion records correctly. If enquiries appear in your inbox but not in the reporting, fix measurement before declaring the media a failure.

Once the brief is agreed, hand over the work: setup inside your business-owned account, tracking verification, context development, copy variants, launch checks, spend monitoring, and landing-page coordination. Ask for plain-language reporting: what spent, what each paying action cost, what changed, and what happens next. A dashboard is not an explanation.

That is the operating split behind groas: you set goals and budget, execution runs around the clock, and a named strategist owns the strategic accountability. The flat management fee does not rise just because you authorize more media spend. The work has to justify keeping the relationship, not simply enlarging the budget.

If I could settle only one decision before hiring anyone, I would start with No. 1: the customer action worth paying for. A clean build and a fair fee cannot rescue a test judged on chats started when the business needs jobs booked. Put the paying action in writing first. Then hand over execution and judge the work on revenue, not activity.