LinkedIn
You can hire someone to run your ChatGPT ads. You cannot hire them to know what a customer is worth to your business, which promises you can keep, or how much you can afford to lose finding out. If you want a launch rather than another platform to learn, use these seven decisions as your onboarding brief. Make the business calls; let the provider do the campaign work.
1. Set your maximum cost per qualified inquiry
Start with the number an agency cannot pull from an ad account: what you can pay for a real sales opportunity. A provider can see clicks and tracked inquiries. It cannot infer your margins or how often your team turns an estimate into a paying job.
Take the value of an average job, multiply it by your gross margin, then multiply that profit by your close rate. Say a job brings in $900 at a 40% margin and you close one in three qualified estimates. That is $360 in gross profit per job and about $120 in gross profit per estimate. So $120 is an approximate break-even media cost for a qualified estimate, before management fees and other costs. It is not a comfortable profit target. If your inputs are rough, label the result an estimate rather than pretending the third decimal place will save you.
This calculation also tells you what not to celebrate. At $4 a click, $1,200 buys about 300 clicks. If 4% become inquiries, that is 12 inquiries at $100 in media each. Looks fine against a $120 ceiling, until you learn that half came from people outside your service area. A dashboard can call them conversions. Your bank account does not have to agree.
Keep media paid to OpenAI separate from setup, creative, landing-page work and follow-up when you judge the test. The platform also has a $25-per-campaign daily minimum in the US, roughly $750 over 30 days, and a suggested $3–$5 starting max bid. A day can spend up to twice its daily budget, subject to a seven-day cap of seven times that budget. Your allowable inquiry cost has to sit beside a real spending limit, not float alone in a spreadsheet.
First step: give your provider your average job value, gross margin and close rate. Then write one sentence: “My estimated break-even media cost is $120 per qualified inquiry; show fees separately.” Change the number to yours. No agency should write it for you.
2. Confirm that you can run ads and will own the account
Eligibility is not setup trivia. It decides whether there is a campaign to set up at all. Before you pay for a launch date, check where your billing entity sits and whether your offer fits the ad rules. The supported-country list counted 63 countries as of October 14, 2026. Only Free and Go users see ads; under-18 targeting and health or politics adjacency are off limits. Financial, health and legal offers face restrictions, including case-by-case approval for US advertisers. If your category needs approval, tell the provider before anyone promises a start date.
I used to take access for granted with Google Ads. That habit is no use here. An agency cannot turn an ineligible offer into an eligible one by writing better copy, and it cannot make an account restriction disappear by putting a more optimistic date in its proposal. Check the offer against OpenAI’s ad policies before you commission assets for it.
Then settle ownership. Create the ad account yourself at ads.openai.com and invite the agency as a Member. The client-creates-then-invites arrangement reflects OpenAI’s agency guidance. It leaves you with an account you control and access you can revoke. If a provider proposes running your spend through its own entity to “handle all that,” walk away. Done-for-you management is not a reason to give away the keys.
Country, currency and timezone are permanent account choices; changing one requires a new account. A Brand Name change pauses serving for review, and a Legal Name change restarts business review. Home-country-only serving can also apply until verification and a home-country spend threshold are met. Those are decisions to check before account creation, not buttons to discover on launch morning.
First step: confirm eligibility, choose the correct account details and create the account in your name. For the fuller access check, use Stuck Setting Up ChatGPT Ads? Check Access Before You Hire Help. No access, no launch date, whatever the proposal says.
3. Pick the one offer that gets the first dollar
Do not hand a provider three services and ask it to discover your business model from the results. Choose one offer. Give it the exact name, price, service area and limits. A garage door company in Dallas might brief it this way: same-day spring replacement at $289, within 25 miles, no commercial jobs. That sentence gives the campaign a boundary a ten-page description of the company might never reach.
The mechanism matters. ChatGPT ads use freeform context hints at ad-group level, alongside options such as geo and first-party audiences; you do not bid on keywords. The hints inform a relevance-weighted second-price auction, but they are not exact-match targets and do not guarantee placement. A tight offer gives the provider a clearer buyer conversation to describe. “We do garage doors” leaves far more room for inquiries you cannot serve.
Price and limits also keep the landing page honest. If the offer applies only to standard residential doors, the page needs to say so. If you cannot reach a neighborhood, do not let the campaign imply that you can. The agency can turn your constraints into hints and copy. It cannot decide that the inconvenient ones no longer count.
First step: write one line with the offer, price, location and exclusions. Hand it to the provider with this instruction: “Run only this offer, at this price, in this area, until I approve another.” If the offer needs category approval, establish that before anyone builds the test.
4. Define the inquiry you will count as a success
I used to tell clients to count every form fill as a conversion. I was wrong. A form fill is an event; a paying customer is a business outcome. The gap between them can swallow a test budget while a report looks increasingly cheerful.
Define a qualified inquiry in the words your team uses when deciding whether to call someone back. For the garage door offer, that might mean a homeowner within the service area who needs a replacement in the next seven days, leaves a working phone number and books an estimate. “Lead” is too vague. If an inquiry comes from a commercial property you do not serve, the form may work perfectly while the campaign fails its actual job.
Your provider turns that definition into tracking and reporting. Before launch, have it install the OpenAI Pixel and Conversions API with the same event ID, preserve the click ID and add UTMs. Then insist on a report that separates all inquiries from the ones that meet your definition. Otherwise, you have no clean way to compare the campaign with the $120 ceiling from decision 1.
Give the test time to read. Impressions may refresh in about 15 minutes while spend lags seven to eight hours. A midday dashboard is a poor place to make a spending decision, and a handful of early inquiries cannot settle whether the offer works. Use the planned four-week test and the stop rule you set in advance rather than moving the goalposts whenever the latest number changes.
First step: write your qualification sentence and send it to the person who handles inquiries as well as the person building the campaign. Optimize and judge the test against that definition; report other form fills separately, not as victories.
5. Hand over the questions buyers already ask
Your front desk may have better targeting material than the platform panel. Pull five real questions from calls, texts or emails from the last 30 days. For the garage door shop, they could be: How much is a spring replacement? Can you come today? Is labor under warranty? Do you serve my neighborhood? Which brands do you carry?
Those questions do two jobs. They show the provider how a buyer describes the problem, and they reveal what the landing page needs to answer before someone calls. A page that states a $289 offer but says nothing about the service area still makes the buyer do detective work. An agency can write around that gap; it cannot know the answer unless you give it one.
It cannot buy those questions as exact keywords, either. Context hints influence relevance but do not reserve a placement on a particular prompt. If someone guarantees that your ad will appear every time a user asks one of your five questions, ask them to explain how. This is not Google Ads circa 2016 with a different chat window.
First step: copy five questions word for word into your brief, then ask the provider to show where each answer appears on the landing page. Use real buyer language, not the phrases your team wishes buyers used. The provider can shape the hints; your customers have already supplied the useful vocabulary.
6. Approve only claims you can defend on the page
The agency can draft copy. It cannot certify that your offer is true. Give it an approved list and a banned list before it writes the first ad.
For the garage door example, the approved list might say: same-day spring replacement, $289 flat for standard doors, licensed and insured in Texas, and a two-year labor warranty. The banned list might say: cheapest in Dallas, number one rated, guaranteed 30-minute arrival, and lifetime warranty. These are examples of the decisions an owner must make, not claims a copywriter should infer. If any approved line is not true for your business, it does not belong on your list.
OpenAI requires truthful ads without unfounded pricing or outcome claims, with a landing page that matches the ad. That last part is practical, not ceremonial. If the ad says “$289 flat,” the page needs to state the price and what it covers. If the offer excludes commercial jobs, the page should not quietly invite them. A persuasive ad that wins a click by making a promise the page cannot support has created a problem, not an opportunity.
This is also where an owner needs to resist their own most enthusiastic suggestion. The provider should push back on an unprovable superlative even when you are the one asking for it. You supply the facts and approve the boundaries; it turns those boundaries into copy that a buyer can understand.
First step: send both lists and the page that supports the approved claims. No new claim goes live without your written approval. That rule is faster than arguing over a rejected promise after money has been spent.
7. Cap the test and name the person who can stop it
A done-for-you test still needs an owner. Name the media cap, the fee cap, the stop rule and the approver before the first dollar goes out. Put them in writing rather than relying on a conversation everyone remembers differently.
For example: $1,200 in media over 30 days plus an agreed flat management fee; stop if qualified inquiries stay above $120 in media cost for two straight weeks; no budget increase without written approval. The $120 figure comes from the business math in decision 1, not from whatever CPA makes the report look good. Keep management, creative, landing-page and follow-up costs visible beside media spend. A campaign can meet its media target and still cost too much overall.
Be honest about test size. Earlier reported pilot minimums fell from $200,000 to $50,000 before being removed when self-serve opened on May 5, 2026. WebFX describes entry tests of $5,000–$10,000 a month for 1,000-plus clicks, with larger mid-level and scaled budgets. If you are testing smaller, say so. A fixed test such as Your First ChatGPT Ads Test: $1,000, 30 Days, No Blind Budget Increases is a clearer brief than an uncapped promise to “see what happens.”
Name who approves account changes, offer changes and additional spend. Set a review date and the evidence you will use: qualified inquiries against your cost ceiling, tracked through the Pixel and Conversions API. A provider should manage bids, context hints, geo, landing-page variants, tracking and reporting without sending you into a dashboard every morning. It should not mistake permission to manage for permission to redefine success.
That is the distinction worth paying for. At groas, Search Ads covers Google, Bing and ChatGPT for a flat $1,499 a month with no cut of spend, with a dedicated manager and a launch in under 24 hours from the onboarding call. If you are deciding whether to learn the platform instead, ChatGPT Ads Course vs Consultant vs Done-for-You: Who Actually Runs the Test? lays out the difference. A course you must finish before launch is homework. Management should remove the campaign work while leaving business judgment where it belongs.
First step: write the cap, stop rule, approver and review date in one paragraph. Nothing scales without your approval. If I could make you answer only one decision before hiring anyone, though, I would start with number 1. Your margin and close rate set the allowable inquiry cost; that cost tells the provider what the test must prove. Without it, “continue or stop” is just a meeting with a chart.