Practical methods, not a client performance claim. Any worked example is fictional. Actions require the relevant evidence, authority and spending boundary.
Clicks are the start of a journey, not a visit guarantee
A traffic source reports clicks, but the landing-page report shows far fewer visits. Before buying more traffic or accusing a source of fraud, I would find the first point where the evidence stops matching. A billed click, a redirect request, a successful page response and an analytics session are different observations. None automatically represents a unique person.
This method applies to third-party traffic and retail landing-page reviews. It does not assume that every source counts an interaction in the same way. In particular, a pop arrival is not interchangeable with a deliberate click on a push creative.
Google documents several reasons Google Ads clicks and Analytics sessions can differ, including counting differences, redirects and page-loading or measurement limitations. Those are useful diagnostic categories, not a promised discrepancy range for other networks. Official clicks-versus-sessions guidance.
Draw the path before comparing totals
Write the actual route: ad interaction → tracking redirect → landing-page request → rendered page → intended action → validated outcome. For each step, record what system can observe it and what its count means. If you cannot observe a stage, mark it unknown instead of filling the gap with an assumption.
- Source: billed versus reported clicks, invalid-traffic adjustments, date basis, timezone and cost freshness.
- Redirect: response status, destination, parameter encoding and whether the required click identifier survives.
- Landing page: successful document responses, device compatibility, page errors and the visible path to the action.
- Measurement: whether the event is eligible to be collected, consent requirements, session rules and collection failures.
- Outcome: unique validated lead, paid order or approved acquisition, under an agreed definition.
Use one timezone and acquisition window where possible. Match a small authorized sample by the relevant click identifier, then investigate the unmatched records. Do not collect extra personal information or bypass consent to make totals look closer.
Separate delivery failure, measurement loss and poor fit
| Observation | Check next | What it does not prove |
|---|---|---|
| Redirect fails or loops | Reproduce the route on the affected device, inspect the status chain and confirm the final destination. | That every missing session had the same cause. |
| Page responds but the event is absent | Rendering, consent, tag eligibility and script errors. A server response alone does not show that the visitor saw a usable page. | That the person never arrived or was a bot. |
| Visits arrive but meaningful actions fall | Message match, device flow, placement mix, product availability and form or checkout friction. | Which one caused the decline. |
| A single slice looks unusual | Compare the same format, GEO, device and page version; examine permitted source diagnostics and repeatable technical behavior. | Fraud based on one IP address, short session or unusual click pattern. |
A universal “acceptable loss percentage” would hide the measurement contract. Compare like with like, and use your own stable baseline only after checking that collection and traffic mix stayed comparable.
Worked example: a gap is not a verdict
Fictional example. All counts and observations below are invented for this method, not campaign results.
One daily report contains 1,000 billed clicks, 720 successful landing-document responses and 610 measured sessions. These counts do not establish 280 fraudulent clicks or 110 failed visits: requests and sessions have different counting rules, and reporting delays may differ.
In a permitted test, the buyer reproduces a redirect on one mobile path that drops the source click identifier. The landing page still opens. That is evidence of an attribution defect in that path; it is not evidence that the defect explains the entire gap.
The next action is to repair that path in a controlled test, verify the identifier end to end and examine affected versus unaffected slices. Hold expansion while the scale of the problem is unclear. If the page itself is broken, stop exposure to the affected route within the agreed operating policy. Reconcile existing traffic before starting another acquisition test.
Close with a testable finding
A useful incident note states the affected route, first observed date, device or format, evidence, proposed fix, spending boundary, owner and readback result. “Traffic quality is bad” is too broad to guide the next person. “This redirect drops the identifier on this reproduced path” is actionable.
After a fix, keep the measurement change in the report. More recorded sessions or conversions can reflect better observation of the same traffic, rather than improved customer acquisition. Do not manufacture an FTD or other optimization event in a live campaign to validate the repair. Use supported diagnostics or an isolated test destination.