A ChatGPT Ads test plan fixes three things before launch: what you’re testing, how much you’ll spend, and what result counts. A month is long enough to learn from that, built around how OpenAI’s platform reports, paces and reviews. The numbers are yours to fill in, because OpenAI says it has no benchmarks to give.
The plan itself is our view. Figures about the platform are OpenAI’s.
The short answer
- Settle the account, tracking and landing pages before the first ad runs.
- Write the target, the budget and the stop rule down before launch.
- Keep the structure small, then judge each week on what its data can show.
- On day 30, scale, reshape or stop. Don’t drift.
Before day one
Most failed tests fail before launch, so work through this list first.
- Your category is allowed in the countries you’ll target.
- The account belongs to your business and verification is done.
- Country, currency and time zone are right.
- Landing pages are specific to each offer and open to OpenAI’s crawlers.
- Conversion tracking sends test events, and the right event is attached to the campaign.
- Every ad carries UTM parameters.
- The success line and stop rules are written down.
Four of these have traps. The rest are checks.
The account and its settings
Your business should create the account and invite any agency, as OpenAI’s setup page advises. Verification can take up to 7 business days, so start it first. The policy tracker shows whether your category is allowed.
Country, currency and time zone cannot change after the account exists. The currency also sets your minimum daily budget. Choose them once and check them twice.
The first campaign’s dates
OpenAI says a first campaign may not show start and end date options before it is published. Those settings appear when you edit it afterwards. Its advice is not to publish a campaign just to reach them. Your ads could start running before your schedule is set.
Landing pages and crawlers
Each ad’s landing page has to be valid and must not block OpenAI’s user agents, OAI-AdsBot and OAI-SearchBot. OpenAI lists blocked pages as a common reason an ad is rejected. Test the page from outside your network as well.
Tracking before traffic
The conversion event has to be attached to the campaign before traffic starts. OpenAI says historical events from a mismatched setup won’t backfill, so a mistake costs you the data. Send a test event, then watch for it. Our conversion tracking guide covers the setup, and ad clicks showing as Direct explains why UTM tags matter.
Set the test’s numbers first
A test with no target can only report what happened. It can’t say whether that’s good. OpenAI says ChatGPT Ads has no benchmarks across advertisers, industries or campaign types yet. So the target has to come from your own economics.
| Decision | How to set it |
|---|---|
| Conversion value | Average order value or, for leads, your close rate times your average deal value. |
| Target cost | What you can pay for one conversion and still make money. |
| Break-even return | One divided by your gross margin, if you judge on return on ad spend. |
| Conversions needed | A number you pick before launch. Fewer conversions means a noisier answer. |
| Budget | Enough to reach that number within the month, at your expected click cost and conversion rate. |
| Stop rule | A limit set in advance. For example: stop or reshape if cost per conversion is double the target once half the budget is spent. |
A worked example
The numbers here are illustrative. Work them in order.
- A $100 daily budget over 30 days is $3,000.
- At $4 a click, $3,000 buys about 750 clicks.
- At a 2% conversion rate, 750 clicks give about 15 conversions.
- $3,000 across 15 conversions is $200 each.
If a conversion is worth $300 to you, that clears the line. If it’s worth $150, it doesn’t, and no amount of tuning inside the month will change the arithmetic. Open this example in the calculator and swap in your own figures.
Keep the structure small
Keep it small, because a thin budget spread across many campaigns teaches you little. Our context hints guide covers the hints. OpenAI’s guidance points the same way: separate campaigns for meaningfully different objectives, regions or product categories, and several ad groups inside each.
| Level | What to set | Why |
|---|---|---|
| Campaign | One per goal and market | Each answers one question, with budget enough to answer it. |
| Ad groups | Two or three, each built around a different buying situation | OpenAI suggests several to test themes and value propositions. |
| Hints | Context hints for that one situation | A hint should describe one situation, not a list of keywords. |
| Ads | Several, each with a different angle | OpenAI recommends multiple titles and descriptions per offer. |
| Budget | A daily budget | OpenAI suggests newcomers start with one, so they can watch delivery first. |
| Objective | Conversions if your event fires reliably and often, otherwise Clicks | Our bidding guide explains the choice. |
| Locations | Include your country explicitly | Excluding other countries doesn’t limit a campaign to yours. |
The ChatGPT Ads test plan, week by week
Budget weeks run Sunday to Saturday in your ad account’s time zone, so compare weeks on those lines.
Week 1: launch and check delivery
Confirm that every ad has passed review and that the campaign, ad groups and ads are all active. A serving ad shows the status Serving.
Impressions and clicks refresh about every 15 minutes. Spend lags 7 to 8 hours, so a zero early on doesn’t mean nothing is accruing. OpenAI says to allow at least 24 hours after launch before treating zero delivery as a problem, unless there’s a clear blocker.
Check that clicks reach your analytics under your UTM tags. Test conversions should appear in Ads Manager within 24 to 48 hours. Leave bids alone for the first days. OpenAI’s low-bid warning isn’t raised within 24 hours of an ad-group change, so early silence proves nothing either way.
Week 2: read the early signals
Compare click-through rates across ads and ad groups. Clicks arrive faster than conversions, so creative is the first thing you can judge.
Replace the clearly weakest ads with new angles, and don’t rewrite everything at once. Editing an ad sends it back for review, so each change costs a delay. Use the Platform and Country breakdowns in Insights to spot a device or market that lags.
Check pacing too. A campaign that spends its weekly allowance early can stop delivering until the next budget week, even while it’s marked active.
Week 3: judge conversions
Look at cost per conversion by ad group, leaving out the last two days, which are still filling in. Reconcile Ads Manager with your analytics and CRM for the same dates and time zone. They won’t match exactly, and OpenAI says a difference doesn’t necessarily mean an error. A large gap usually points to tracking.
Then make one change per ad group: a bid, a budget or new context hints. OpenAI advises against changing several settings at once if you want to know what helped. Note the date.
Week 4: decide
| Result | Next step |
|---|---|
| At or under target cost | Scale in steps. Raise budgets on the ad groups that work, keep testing new ads, and check the cost holds. |
| Close to target | Reshape. Try new landing pages, sharper context hints or a different objective. A new objective means a new campaign. |
| Far from target, or the stop rule triggered | Stop, and write down what you learned. Re-testing one variable later is cheaper than drifting on. |
What you can change mid-test
Some settings are fixed once a campaign exists. A few changes have side effects too, so check this table before you edit.
| Setting | Can you change it? |
|---|---|
| Bids and bid adjustments | Yes, per ad group. |
| Budget amount | Yes. After a mid-week change, the seven-day limit is prorated from the change to the next Sunday at midnight. |
| Context hints and ads | Yes. Edited ads go back through review. |
| Audiences, locations and platforms | Yes. Custom audience members can be updated without rebuilding the campaign. |
| Objective, billing option or conversion event | No. Create or clone a new campaign. |
| Budget type | Only from campaign total to daily, not back. |
Reading the results
Three facts about how OpenAI reports can change the answer.
First, know what’s in the total. OpenAI’s Help Center says that with the view-through window set to one day, the Conversions total includes view-through conversions. Decide beforehand whether your target counts them.
Second, watch the window. Changing attribution windows changes reporting only, not bidding or billing.
Third, allow for the beta. OpenAI says delivery and spend can fluctuate while it improves its systems. For conversion campaigns it adds that its goal is better conversion results, not simply spending the whole budget.
Keep a log of every change and its date. Without one, week 4 turns into guesswork.
How a first test goes wrong
Four failures account for most unreadable tests, in our view.
Spreading the budget too thin
Five campaigns on a small budget each get too little traffic to show a result. One or two campaigns with real budgets answer the question. Add more only after the first answer.
Changing too much at once
If you change bids, hints and ads in the same week, you can’t tell which helped. Change one thing per ad group, note the date, and wait a full week.
Judging on clicks
A low cost per click proves little. Conversions come later, and they are what the target measures. Wait for the reporting delay, then judge on conversions.
Letting the stop rule slide
A stop rule works only if it’s written before the money is spent. After a bad week, any limit looks arbitrary. Hold the one you wrote.
Questions
How much budget does a 30-day test need?
Enough to reach the number of conversions you’ll judge by. That depends on your cost per click and conversion rate, so run your numbers through the cost calculator. The platform minimum is set by your billing currency, $25 a day per campaign in US dollars. Our cost guide covers how the auction sets prices.
Can a test be shorter than 30 days?
It can, but the answer will be noisier. Conversions take 24 to 48 hours to report and delivery varies from day to day. Short tests therefore give noisy answers. Two full budget weeks is the least we’d judge, and a full month is better.
Should a test use Maximize results?
It can. OpenAI selects it by default for eligible new ad groups, and it sets and adjusts bids for you. But it prioritizes result volume and doesn’t guarantee a particular cost per click or conversion. If you must stay under a bid limit, use a manual max bid.
What if I get clicks but no conversions?
Check tracking before blaming the ads. The event you send must match the conversion event on the campaign, or Ads Manager shows zero. Then check the attribution window, which decides how long after a click a conversion still counts. Our tracking guide has a checklist.
What if the campaign doesn’t spend its budget?
That’s allowed. OpenAI says pacing doesn’t guarantee a campaign spends its full budget, and low bids can still limit delivery. Check review status, the end date and billing first. Then check the bid warning, remembering that it’s guidance and not a rule.
What does a good result look like?
OpenAI says there are no performance benchmarks across advertisers yet, and results depend on objective, creative, landing page and relevance. So a good result is one that beats your own target. You write it down first so it can be checked later.
Can someone run the test for us?
Yes. Our launch and setup service includes a 30-day test plan, with the budget, success metrics and stop rules agreed in writing. It also covers the account, the tracking and the first campaigns, so the test starts on a sound footing. The account stays yours throughout.
