Most agencies will not tell you this: the most expensive word in a done-for-you setup is done. It can mean an account exists, a campaign sits in draft, a pixel fires, or a report arrived. None of those proves your ads are delivering or that the right action is being counted.

I wrote this for the owner who started ChatGPT Ads, got stuck halfway, and now needs someone else to finish it. Read these terms in the order you are likely to meet them. Each definition comes with the practitioner’s gloss: what gets confused with what, when the distinction costs you money, and what you can check for yourself.

Eligibility and availability

Eligibility and availability means whether OpenAI will let your business advertise at all. The headline about availability across 89 countries does not answer the question for your account. Eligibility follows where your legal entity is based, and new self-serve accounts can initially advertise only in their home country until verification and spend thresholds are met.

A provider saying “we offer ChatGPT Ads setup” does not prove you can run ads yet. Ask to see your advertiser account’s eligible country and account status. If that screen is missing, the blocker is platform access. Campaign work cannot make an ineligible account deliver.

Account ownership and permissions

Account ownership and permissions means who can keep the account, pay for it, and remove the other party’s access. Shared access sounds reassuring until the relationship ends. Then the useful questions are less friendly: Can you sign in as an Admin? Is billing in your name? Can you retain the account and its history without the provider?

I used to tell clients shared access was fine. I was wrong. Access is not the same as control. Before work starts, ask the provider to show you your Admin seat and billing details, then establish how you will remove their access if you part ways. Do not accept “we’ll sort that out later” as an ownership plan. Later is when you need it.

Management fee and media spend

A management fee pays the provider; media spend pays the ad platform. They are not the same bill. A proposal that blends them hides what changes when the budget rises. Google-agency pricing commonly uses a percentage of spend or a flat fee, with setup, tracking, and landing pages sometimes charged separately. That is a useful warning when you read any setup proposal, not a price list for ChatGPT Ads.

Get four things in writing: who pays OpenAI, whose card is charged, what the management fee covers, and what triggers an extra charge. If you get one blended number instead, you cannot tell whether an increase buys more traffic or gives the provider a raise. Split the two costs before you agree to either.

Campaign setup

Campaign setup means the advertiser account and campaign are configured for an ad to deliver. It is a deliverable, not a result. For ChatGPT Ads, country, currency, and timezone lock after advertiser account creation. Billing, brand details, verification, and user roles also need attention. Drafting an ad during signup does not mean it passed review.

The loose version of “setup complete” is a campaign draft under someone else’s login. The version worth paying for has evidence you can inspect: your account details are correct, your Admin access works, billing and verification are in place, and the campaign is delivering or scheduled rather than sitting in draft or review. If you cannot see its status yourself, setup is started, not finished.

Ad copy and landing-page alignment

Ad copy and landing-page alignment means the ad’s promise matches the page a visitor reaches after clicking. ChatGPT Ads can be easy to misread here. Sponsored units appear below the answer for Free and Go users, not inside the answer. You provide plain-language context hints, and OpenAI matches ads to conversation intent. Clever copy alone does not complete that job. Paid placement does not buy a place in the answer, either.

The moment that matters is the click. The visitor has just read an answer and now lands on your page. Does the page address what the ad promised, or make them hunt for it? Ask for each context hint paired with its landing URL, then open every URL on your phone. If the promise disappears after the click, you are paying to lose the visitor.

Launch-ready versus live

Launch-ready means the checklist is complete; live means the platform is serving your ad and spending money. What a deck calls launch-ready, I call a draft with a bow on it. Structure built, copy written, tracking installed, billing present: good. None of that is delivery.

For live, look for passed review, delivering status, and impressions and spend visible under your own login. Ask for a timestamped delivery screen and the review status of each ad before you accept a launch claim. A builder preview is not proof. You bought the rehearsal, not the show.

Conversion event

A conversion event is the business action you tell the ad platform to count. A ChatGPT Ads campaign needs one active standard event as its optimization goal; a custom event cannot serve as that goal. That makes the choice more consequential than a line item labelled “tracking installed.”

Teams often count every form fill, then wonder why they get more forms rather than more future customers. Ask your provider to name the chosen event in plain English and show how it is configured in the account. Keep other actions distinct for reporting. If nobody can name the one action the campaign is optimizing for, you have set a counter, not a goal.

Tracking validation

Tracking validation is a test showing that the right action is counted once: not twice, and not never. ChatGPT Ads can receive events through an oaiq JavaScript pixel, a server-side Conversions API, and a no-JS image tag. Pixel and server events need matching event names and IDs for deduplication. Merely pasting code onto a page does not prove any of that works.

The failure points are concrete: a redirect strips the oppref click parameter; pixel and server event IDs do not match; amounts arrive in the wrong units; or a bad event causes a batch to be rejected. Only events from the last seven days are accepted. Ask for a live test purchase or form submission that appears once in the platform within a day, with the click parameter intact. No test record, no validation.

Attribution window and conversion lag

An attribution window is how long after an ad interaction the platform can take credit; conversion lag is how long you wait for that credit to appear. ChatGPT Ads defaults to 30 days for a click and one day for a view. Conversions are dated to the ad interaction, not the purchase, and daily processing adds at least a day’s lag.

Confuse the two and you can declare a sale missing when the report is simply not ready. I have seen that panic in Google Ads for years. A B2B order that closes on day 35 can also fall outside a 30-day click window; waiting longer will not put it inside one. Ask to see the configured windows, click and view-through results separately, and days-to-conversion. Judge closed periods, not today’s unfinished numbers. A shorter window than your sales cycle is a measurement decision, not proof the pixel broke.

Guardrail

A guardrail is a spending or performance limit the buyer sets before ads run, with an agreed response when it is reached. Think maximum test spend, a CPA ceiling on qualified outcomes, and who must approve a higher cap. A campaign budget is not the whole conversation: a manager may propose changing it to chase volume. Your approval rule says when they must stop and ask.

Some proposals call any budget a guardrail because the word sounds sturdy. I want the less glamorous detail: who monitors the limit, what happens when it is hit, and where that decision is logged. Put the answers in writing on day one. A limit nobody has to act on is a suggestion.

Test budget and scaling approval

A test budget is the amount you agree to risk to learn whether ChatGPT Ads works for you; scaling approval is separate permission to spend more. Permission to test is not permission to scale.

Say you approve $2,000. That buys a test, not a promised result. Before it starts, the manager should say what decision the test will support, which CPA on qualified outcomes you will examine, and when you will decide. If the results justify more spend, you can approve a new cap. If they do not, “the campaign is still ramping” is not an authorization to keep charging your card. No new cap in writing, no new spend.

Optimization and reporting

Optimization changes something in the account to improve the outcome; reporting describes what happened. One is work. The other may be a PDF.

In ChatGPT Ads, an optimization might rewrite a context hint that attracts the wrong conversations, pause a placement that spends without qualified actions, fix a deduplication problem, or tighten a guardrail after the first useful results arrive. A chart showing CPA fell does not tell you what the manager did to cause it. It may not tell you they did anything.

Ask for an action log: what changed, when, why, and what result will be checked next. If the weekly update lists impressions and clicks but no decisions, you paid for narration. No action log, no evidence of management.

Done-for-you handoff

A done-for-you handoff is a written list of unfinished jobs, their owners, and the evidence that closes each one. It is how you stop “done” from covering five different meanings on the same status call. Put the checks in the contract, not in someone’s memory:

JobEvidence to ask for
AccessYour working Admin login, your billing details, and a clear way to remove the provider’s access.
LaunchPassed review and delivering status, with spend visible under your login.
TrackingOne live test counted once, with oppref intact and matching pixel and server event IDs where both are used.
AttributionThe configured windows and separate click and view-through reporting.
ManagementA dated log of changes, reasons, and the results they will be judged against.

If you started the setup yourself, first work out whether access, verification, or tracking is blocking you. Then give a provider only the rows still open, with an owner and a date for each. Anything without evidence stays open, whatever the status call said.

This is also how I would read the offer from groas. Its Search Ads service covers Google, Bing, and ChatGPT Ads for a flat $1,499 a month, with a named manager and around-the-clock work on bids, keywords, and budgets. The offer includes a landing page for every ad group and says campaigns can go live in your accounts within 24 hours of the onboarding call. None of that removes the checks above. Your business still needs to clear eligibility; your access and billing still need to be yours; and a promised launch is not a promise about which conversations will match or how many visitors will buy. Make each claim of completion visible in your account.

Qualified lead

A qualified lead is someone your sales team considers worth working, not merely someone who completed a form. This is the term I would retire from proposals, because it so often arrives without a definition. “More qualified leads” can mean “more form fills” with better lighting.

Keep form fills in the report, but agree on a written qualification rule, who applies it, and how the CRM records which ad-sourced names clear it. Track closed-won outcomes there, too. Then judge the work against those outcomes rather than celebrating a lower cost per form. If a provider promises qualified leads without asking how you score one, call the promise unfinished. Buy outcomes you can identify in the CRM, or do not buy at all.