Clicks without sales can reflect a tracking fault, a mismatch between the ad and the offer, site friction, visitors who need more time, or poor-quality traffic. Start by checking the outcome and following the paid visits before deciding which explanation fits. A low conversion rate alone cannot tell you whether bots are clicking your ads.
This guide uses Button's published observations from three small websites. It explains how to investigate; it does not report a new experiment or a measured increase in sales.
1. Check what “no sales” means
Compare the store's order records with the event your advertising report calls a conversion. A cart visit, a thank-you-page load, a submitted form and a paid order answer different questions.
If orders exist but the advertising report shows none, investigate measurement and attribution first. Check the event definition, consent behaviour, duplicate tags, checkout handoffs and whether the event reaches the intended account. Our collection and advertising-tag troubleshooting guide covers Button's part of that investigation.
If orders are also absent, establish how long a customer normally takes to buy. Compare cohorts with the same follow-up time. Yesterday's visitors and last month's visitors have had different opportunities to place an order.
Record the actual counts alongside the rate. Two sales from a small cohort and two from thousands of clicks need different investigations; neither supports a universal bot percentage.
2. Follow paid visits to the point where they stop
Start with identifiable paid visits, then follow their same-site journeys through the relevant product, cart, checkout and outcome pages. Deduplicate repeated observations of the same click. Keep visits with incomplete collection separate from those with a visible journey.
An ad click identifier helps link observations. It does not, by itself, prove that the platform charged for the click. Missing identifiers also do not prove that a visit was organic.
| Observation | Next useful check |
|---|---|
| Paid visits are recorded, but few reach the advertised product | Confirm the destination URL, page loading and whether the offer matches the ad |
| Visitors reach a product but rarely reach the cart | Review availability, price, delivery information and the intent of that campaign |
| Visitors reach payment but no order is confirmed | Check checkout errors, payment failures and external checkout visibility |
| The gap is concentrated in one campaign or device group | Compare that group with its own baseline and inspect individual journeys |
| Collection stops before the journey is visible | Investigate the observation gap before calling the visitor a quick leaver |
These are diagnostic branches, not findings about your site. Check each against your own records.
3. Check whether the campaign reaches the intended market
In our wrong-country study, foreign IPs accounted for 15 of 241 ad-click IPs on a UK lettings site. The campaign breakdown exposed a more specific pattern: all nine ad-click IPs associated with one Google campaign were outside the UK.
The observations covered 9 October 2026 at 11:31 through 13 October at 20:48 UTC. Nine IPs are a small sample, and we did not have that campaign's targeting settings. Overseas property interest might have been intentional. The result justified checking the campaign's intended audience and settings; it did not establish fraud or prove which setting caused the pattern.
Google distinguishes targeting people in or regularly in a location from also reaching people who show interest in it. Review the current location options against the business's actual service area before proposing a change.
4. Inspect traffic quality without discarding buyer counterexamples
Network reputation can help you choose what to inspect. It cannot settle the outcome on its own.
In our residential-proxy study, three of 596 flagged IPs reached a payment-success page, compared with four of 681 unflagged IPs. That comparison included all eligible IPs in those two groups, not just paid clicks. It covered about 46 hours on one store, and the outcome was a page observation rather than an independently reconciled order.
That is a concrete reason to inspect apparently successful journeys before imposing a broad network rule. It is not evidence that all flagged visitors were legitimate or that proxies are generally harmless.
For suspected automation, examine browser evidence and activity alongside the paid-click linkage. Our companion guide explains what counts as evidence of bots clicking ads.
5. Write down the next test
A useful investigation ends with a specific uncertainty and a way to resolve it:
- Tracking hypothesis: orders exist but events are missing. Trace a known order through the measurement path.
- Targeting hypothesis: one campaign reaches an unintended market. Inspect its settings and compare appropriate cohorts.
- Site hypothesis: visitors stop at the same checkout step. Reproduce that step and reconcile it with payment records.
- Automation hypothesis: paid visits carry corroborating technical evidence. Inspect examples and known legitimate counterexamples before testing a response.
For any change, record its date, affected population, success measure and reversal condition. More blocked visits or fewer recorded conversions do not establish improved acquisition. Judge the result against meaningful outcomes and the risk of losing legitimate customers.
Start with your own traffic
Install Button to observe traffic quality and visitor journeys, then compare the campaigns that need attention. Use order or CRM records separately to verify commercial outcomes; this guide does not assume they are automatically connected.
Questions advertisers ask
Do clicks without conversions mean click fraud?
No. They establish a performance gap. Tracking, targeting, the offer, the site and the time allowed for a sale are competing explanations. Evidence about automation must come from additional observations.
Should I block suspicious IPs immediately?
First establish why they are suspicious, which paid visits they relate to and whether the same rule includes potentially valuable customers. Choose a response that addresses the observed problem.
Sources and method
The examples come from the linked October 2026 proxy and geography studies. Their public dataset and reproduction scripts document the per-IP unit, site-specific windows, crawler exclusions and collection gaps. They are short observational samples, not controlled tests of advertising changes. No new traffic was queried for this guide.
AI disclosure: Written by ChatGPT (AI) from Button's published research and the linked platform documentation. Research signals and original data: Nathan Pankowsky. Editor: Nathan Pankowsky.