ChatGPT Ads test budget: one tight intent cluster or several at once?

Something I keep coming back to with early ChatGPT Ads tests: on a small budget, do you spread across several intent clusters to learn which prompts convert, or run one tight cluster until you have a clean number? My instinct is the tight cluster. At a few dollars a day, splitting across five themes means none of them gets enough conversions to read, and the channel verdict you end up with is really just noise. One cluster gives you a yes or no faster, even if it tells you less about everything else. The counter-argument is that a narrow test can make a healthy channel look dead because you picked the wrong query. That risk is real, but it's cheaper to find out in week two than in month three. Has anyone seen a rough volume threshold where the numbers start to stabilize?

I agree with the tight cluster, but I'd measure it in conversions per week rather than calendar days. At a few dollars a day you might only get two or three conversions a week, which is nowhere near enough to call a channel dead. A common rough rule is around 30 conversions before you trust a cost per result. If that takes two months at your budget, the test is too small, not the channel.

I'd also count sales-qualified leads next to the platform number. A cheap conversion that sales never accepts can make a cluster look better than it really is.

Counting per week and checking SQLs both make sense. A tight cluster can still hide two intents under one label, though. A CRM tool might run on "best CRM for small teams" and "CRM pricing comparison," and those people sit at different points in the decision and often convert at very different rates. Lumped together, you get a middling blend that hides one strong sub-intent and one dead one.

You don't have to split the budget to catch this. Give each sub-intent its own UTM tag or landing path, so conversions stay in one campaign but you can read them apart in analytics. The cluster still gives you one clean verdict, and by week three you know which sub-intent to lean into. The catch is that each slice will be small at a few dollars a day, so treat the split as a direction to explore, not a number to bet on.

I don't think week three is a safe decision point. Going by Rowan's estimate of two or three conversions a week at this budget, you'd have maybe six to nine conversions total by then. Split across two sub-intents, that's three or four each, which is far too few to call one strong and the other dead.

Tagging also won't settle it. A UTM or landing path tells you which page someone ended up on, not what they were actually asking. If both sub-intents see the same ad, the tags mostly just show which page got the click.

I'd keep the split as a note to revisit once the cluster reaches around 30 conversions, not as a call to make at week three.

Fair on the ad side, but tagging isn't useless. If the two sub-intents land on different pages, say pricing for one and features for the other, the landing path works as a rough proxy for intent. That's better than one blended number, just not good enough to move budget at this volume.

Agreed that the landing page only works as a proxy if the two pages really differ, and ad copy can break that. If the headline mentions pricing and features together, the click goes wherever the ad promised, so the page split tells you about the ad, not the buyer.

At this volume, asking people directly may be cleaner. Put a single dropdown on the form or booking step, something like "What are you comparing right now?", with three or four options mapped to your sub-intents. It costs nothing extra in spend and tags every conversion. It's self-reported, so some people will pick the wrong box, but across 20 or 30 answers the pattern should still come through. You get intent per conversion without splitting the budget.

Keep the options short, though, because a long form will cut the conversions you're trying to count.

Make the dropdown optional and treat the blanks as their own bucket. If the people who skip it close at a different rate than the ones who answer, that gap often tells you more than the labels themselves, so compare both groups against your SQL rate.

Theo's blank bucket is a good idea, but I'd add one rule before building the dropdown: only split intents that need different ad copy. If a single headline works for both "best CRM for small teams" and "CRM pricing comparison," the ad is already treating those people the same, so the split only lives in your spreadsheet and the form question just tells you how the ad landed.

If the two intents need different headlines to pull the right click, they're really separate, and each deserves its own ad even at a few dollars a day. If they don't, keep one ad and one number. The dropdown then works as a check on your copy, which is more useful than chasing small slices of budget. Your ad decides who clicks, so the real question is whether the copy is attracting the buyer you described, and the form can tell you that.

If the copy really needs to differ, let the platform split the numbers instead of relying on tags. Two ads in separate ad groups report their own clicks and conversions, so you can read the sub-intents apart without any UTM work. Tags only earn their keep when the sub-intents share an ad and you still need to tell them apart, which is the situation Atlas was worried about.

For shared copy, the dropdown is most useful when you compare its answers against what the headline promised. Say most people pick "pricing" but the headline leads with features. That tells you the copy is attracting the wrong buyer, so the fix belongs in the ad, not in a conclusion about the market. I'd sort that out before reading anything into the cluster's conversion rate.

Separate ad groups fix the reporting, but they also split the budget, and that's the part people skip. My rule is simple: split only when two things are true. The headlines have to be different enough that the other group's audience wouldn't click them, and each slice has to realistically reach your readable number, say 30 conversions, in a timeframe you can live with. If the second part fails, keep one ad group and let the dropdown and the copy check do the work.

The reason is that a split only helps if each side can produce a verdict. Otherwise you've paid for clarity and ended up with two noisy numbers. Rowan's weekly math makes this easy to check: take your expected conversions per week, divide by the number of slices, and see whether the answer lands somewhere you'd accept waiting for.