ChatGPT Ads test: how many conversions before you judge the channel?

Say you're testing ChatGPT Ads on about 300 a month. After two weeks you've got maybe nine conversions, and the cost per conversion looks bad next to your Google Search numbers. My instinct is not to judge the channel on that yet, but nine conversions doesn't tell you much either way, since one good or bad week can swing the result a lot. What I'd look at first is whether people coming from the chat convert at a similar rate to your search visitors on the same offer. If the click-to-lead rate is close, the channel is probably fine and the volume just isn't there yet. If it's far off, that's a signal even at low numbers. How do you set the threshold before you pull the plug on a test?

Set the threshold in money before you launch, not in conversions. Pick the highest cost per lead you can live with, plus the budget you're willing to burn finding out. Comparing click-to-lead against search is shaky anyway, because search visitors already had intent and chat users may not. What's your break-even cost per lead?

Agreed on money first. Before you read the cost per lead, split the funnel into click to landing page view and landing page view to lead, because those two numbers fail for different reasons. If a big share of chat clicks never register as a page view, the problem is a slow page, a broken redirect, or a tracking gap, not the audience. If page views are healthy but leads are thin, test the offer or the form. You can usually see that split within a few days, long before nine conversions show up.

Write the stop rule as a spend figure too, something like "pause at 2x break-even CPL once $X is spent," so nobody has to argue about it on day 14. Break-even is lead value times close rate, and that's the number to compare against what search costs you per lead.

Agreed on the split. I'd start with an even simpler check though: compare the ad platform's click count with GA4 sessions for the same UTM tags and dates. Redirects and in-app browsers often strip the referrer or drop the tag before it fires, so some clicks end up as direct or don't show up at all. If the gap is more than 10 to 15 percent, fix that before you read anything into lead numbers, because your page-view rate is built on a bad denominator. It takes about ten minutes and usually tells you more than most dashboards.

Depends on volume. At around 300 a month, a 15 percent gap can be just a few dozen clicks, and that's often plain day-to-day noise. I'd check whether the gap holds steady across days before deciding the tag is broken.

Depends on whether consent is what's eating your sessions. If a chunk of visitors decline analytics cookies, GA4 just doesn't count them, and the gap can look like a broken tag even though it stays pretty steady from day to day. Check your consent rate before you go chasing the tag.

To tell the two causes apart, compare the ad platform's own landing page views against its click count instead of relying on GA4. Platform-side events are counted differently and usually hold up better when visitors decline cookies. If that gap is small but GA4 is far lower, the traffic is arriving and the missing sessions are a consent problem. If the platform's landing page views are also well below clicks, look at redirects or page speed instead. Either way, find out your consent rate before you use GA4 sessions as the denominator for any conversion rate.

Keep the numerator and denominator in the same system. Platform clicks go with platform conversions, GA4 sessions go with GA4 leads. Mix them and a channel can look broken when the tracking is just split across tools, and any chat vs search comparison built on mixed numbers will mislead you either way.