Google Ads users try my chatbot but rarely subscribe. How do I find out why?

Start with the first thing a new user gets done, then compare what happens next. A signup, a first message and a paywall view tell you that someone entered the chatbot. They do not tell you whether the visit solved a problem worth subscribing for.

Compare Google Ads signups with all signups from a source that produces paying users, at the same age after signup. Find the first meaningful difference in useful work, return use, paywall behavior or purchase. That difference points you toward an acquisition promise, a product experience or a payment path to investigate. It does not, by itself, prove a cause.

If you already have one paid Google Ads subscriber, keep that person in the count. A universal “Ads users cannot check out” story has become harder to defend. One successful purchase still cannot clear every plan, device, payment method or path to the paywall.

A first message is a greeting, not a job finished

What would make this chatbot useful enough to revisit? Decide that before looking for a flattering event. For example, suppose the product helps someone write a customer reply. “Sent a message” is too early; “received a draft, refined it for the actual customer and copied the usable reply” is closer to a finished task. If the product instead sells quick image creation, one good image may be the whole task. A short session is not automatically a failed session, and a long chat is not automatically valuable.

Look at the tasks people who pay actually came to complete, but do not define “useful” as “whatever payers did.” That would make the answer true by construction. Check the candidate task against users who did and did not pay: what did they try, what result did they get, and did they have a reason to return? Product events can count a proposed milestone; a few direct user conversations can tell you whether it meant what you thought. Amplitude's activation guidance similarly starts with an event that represents value and follows retention after it.

Now build two signup cohorts, with one account per person and the acquisition source captured at signup. Use the same signup period where possible, and give each person the same elapsed window: for example, first useful task within 24 hours, another task within seven days, and subscription within 14 days. Do not compare yesterday's Ads signups with six-month-old referrals. Do not compare all Ads signups with only the other source's purchasers; that would select the winner after the race. Source labels need one consistent meaning, such as first acquisition source rather than whichever session brought a person back. Google Analytics distinguishes first-user acquisition from session acquisition; mixing the two changes who is in each group.

Here is an invented readout. It is a method example, not a chatbot benchmark or a report from any advertiser. Every number is a count out of 100 signups; rows are separate milestones, not necessarily a strict sequence.

Scroll sideways to read every column. With a keyboard, focus the table and use the arrow keys.

The purchase gap matters. The 18-versus-51 task gap is 33 per 100 signups and gives you a place to look first, but these separate rows do not show what each person saw first. Check account-level event order: did users who missed the task leave before any paywall or checkout prompt, or did one interrupt the task? In the first case, inspect the promised task and first-run experience before redesigning checkout; in the second, test that interruption too. The one Google subscriber also prevents the attractive but absolute diagnosis that no one from Ads can pay. Compare device, location and entry route before assigning the gap to intent: Ads users may be meeting a different first-run experience.

There is a catch worth keeping in the frame: maybe those Google users wanted a single free image and got it. If one image was their intended useful result, this table's “useful task” definition is wrong. Fix the definition or compare task types separately before declaring them unmotivated. If the product sells a recurring subscription but the ad recruits one-off jobs, you can change that promise or consider an offer suited to one-off use. A paywall can be seen on the way out, the way a museum gift shop can be seen by someone who came only for one exhibit.

Let the first gap choose the next investigation

Suppose the chosen task survives the one-off-image check and, among users who miss it, account-level paths show no paywall or checkout prompt before they leave. Ads users often make one quick image, while users from the paying source tend to complete and return to a task the subscription supports. First inspect the promise that brought the Ads cohort in. Do the creative and landing page invite a repeat problem the paid product solves, or do they mainly promise a free one-off trick? If instead a prompt appeared before task completion, inspect that interruption and its first-run route before changing the ad promise.

In Google Ads, identify the actual campaign type, the ads or assets people saw, the landing-page promise, available audience and placement evidence, and the campaign's bid strategy and conversion goal. CPC bidding is available on both Search and Display; “we run CPC” does not tell you this is a Search campaign. If the campaign is Search, the search terms report can show a subset of actual queries that triggered ads. Google omits some low-volume terms, so the visible list is evidence to inspect, not a complete transcript of intent. For other campaign types, use the acquisition evidence their reporting actually supplies. In particular, a Performance Max placement report gives impressions on some placements and is not a full performance report.

If a creative promises “make an image in seconds,” quick image use may be exactly what it recruited. If the chosen useful task is valid and account-level paths show users who missed it left without seeing a paywall or checkout prompt, a subscription business can test making the paid, recurring use case clear in one campaign's ad and landing page, then compare a new same-age cohort on the same first-task and subscription measures. If Search queries reveal a clearly mismatched group, you can also narrow that group; avoid blocking a term just because its few visible users did not buy. The trade-off is fewer easy signups, which is acceptable only if useful tasks and eventual subscriptions hold up. This is a test of a promise, not proof that Google found “bad people.”

Also inspect what the campaign is set to optimize. Google distinguishes primary conversion actions used for bidding from secondary observations, subject to the campaign goal and custom-goal settings. If the campaign bids toward a free signup, the account has asked it to favor signups. If it uses a click-focused strategy, do not claim Google is optimizing for those signups at all. Record the actual setting before proposing a bidding change; one subscription is not enough to make a purchase-focused bidding switch self-evidently sound.

Now change one important fact in the example. Suppose both sources complete the useful task and return at similar rates, but Google users who see the paid offer rarely start or finish a purchase. The acquisition-promise theory loses priority. Compare the exact paywall version, plan and price shown; the step from paywall view to checkout start; payment attempts and failures; and purchase records against your analytics events. Test the same device and route as the affected users. Other-source purchases and one Ads purchase show that some routes work, not that every route works. If people complete useful work but prefer to stop at the free tier, that points to the value and price of the subscription rather than a broken card form. This later-gap case earns a product or checkout change before another targeting change.

If you cannot join the click to the task

An Ads dashboard with a signup total and an app dashboard with a first-chat total is not a source-to-task comparison. The two totals might not describe the same people. Preserve an acquisition label with the account at signup, then join it to task, return, paywall, checkout and successful-subscription events by a stable account ID. Keep the source definition and event definitions fixed for the comparison. Mixpanel's full-journey account explains why joining pre-signup source to identified product behavior matters; Amplitude's freemium guide separates activation from free-to-paid conversion.

If that join was never stored, old channel-level product behavior may be unknowable. Tag a small prospective cohort, verify that a signup retains its source and that a completed payment appears once under the same account, then allow both sources the same observation window. You can still inspect the campaign promise and test checkout immediately. You cannot reconstruct a precise historical Ads-to-purchase rate by dividing unrelated dashboards. And if some signups look automated, verify abuse with stronger evidence before removing them from either denominator; an odd email address is not a measurement rule.

The decision is narrower than “Ads works” or “Ads fails.” If the task definition holds and Ads users miss it without an earlier paywall or checkout prompt, compare the promise, audience and first-run routes; test a promise or audience change if those checks point there. If a prompt interrupts users before they finish the task, inspect its timing, offer and route first. If the cohorts do useful work alike and diverge at payment, examine the offer and purchase path. If you cannot see the first divergence yet, build the smallest trustworthy join and wait long enough for both groups to have the same chance to subscribe. One paid Ads user remains in the all-signup Ads cohort—as a real purchase, not a magic eraser.