Google Ads conversions have dropped but clicks continue—how do I find out what changed?

Start by finding which result fell: real orders or leads, the events sent to measurement, or the conversions credited to your ads. Those can move differently. Clicks continuing tells you that some traffic is arriving; it does not establish that the amount, intent or cost of that traffic stayed the same.

Before changing bids, budgets or conversion settings, save the affected report, its dates and filters, and the current settings. Then check the actual purchase or lead journey. A checkout that rejects payments deserves an immediate repair and a decision about stopping affected spend. Preserving a comparison is useful; continuing to buy visits to a known broken destination is not a prerequisite for diagnosis.

If there is no obvious operational failure, work from the business event toward the report, then examine what happened inside the traffic total.

Give each number a job

Suppose someone clicks an ad on Monday, submits a quote request on Wednesday, and that request is uploaded on Thursday. These dates answer different questions:

Google Ads normally puts conversion credit against the ad-interaction date. For supported actions, its “by conv. time” columns instead organize credited conversions by when the conversion happened. Neither is an upload-date ledger. Processing also takes time, and a successful upload does not guarantee that every submitted event will appear as a conversion. Google’s reporting definitions and offline-import troubleshooting explain these distinctions.

So Wednesday’s business requests should not be compared directly with Wednesday’s ordinary Ads “Conversions.” Some of Wednesday’s requests may belong to Monday’s clicks, while Wednesday’s clicks may still produce requests later.

Choose the conversion action

Name the outcome you mean. A submitted form, an accepted lead and a completed sale are different events. In Goals → Summary → View all conversion actions, identify the exact action behind the decline, its source and the campaign goals using it. Compare that action’s “Conversions” and “All conversions” rather than an account total that also contains phone calls, purchases or page views. Goal selection and primary/secondary settings affect what appears in the “Conversions” column. If the action remains steady in “All conversions” while “Conversions” falls, inspect its inclusion in reporting and campaign goals first. Google explains primary and secondary actions.

Find where the evidence parts company

For the last healthy period and the affected period, obtain a dated export from the system that actually receives the outcome: saved form submissions, CRM stage changes, payment records or an order database. Use the same definition and time zone in both periods. A payment record is useful for a purchase investigation; it cannot tell you how many people submitted an earlier enquiry.

Keep an identifier and occurrence time where available. Ask the measurement owner for the corresponding event or import evidence: the action or destination used, the event time submitted, and any receipt or rejection. Start with several known outcomes from before and after the apparent break. A sample can locate a defect; it cannot establish the total missing volume. Broaden the reconciliation if the sample reveals a gap.

For an action that supports occurrence-date reporting, such as the quote-request import above, add “All conv. (by conv. time)” to the campaign table, segment by conversion action and retain the chosen action. Match the business-record dates and time zone. Use an ad-sourced subset of those records where it is reliably identified; otherwise the sitewide total is only a broader check. Google’s guidance on reconciling reporting dates.

Google excludes Store visit and Store sales conversions from these columns. If that is your selected action, an absent occurrence-date total is a reporting limit, not evidence of a measurement break. Compare the dated business records to establish whether real outcomes fell, and inspect the reporting and diagnostics supported for that action separately. Leave the unavailable occurrence-date reconciliation unresolved. This exception concerns those Google conversion types; ordinary offline lead imports are not automatically Store sales conversions. Google’s “Time of conversion” note.

Where the records support a comparison, the finding should change the next action:

Do not force the totals to match. Ads can include modeled or fractional credit; business records can include other channels; counting and attribution rules differ. The useful finding is a new, unexplained change in their relationship, followed by evidence of where it occurs. A stable sitewide lead total also cannot establish stable paid-search results: another source may have replaced the lost leads. Google’s conversion definitions.

Test the relevant measurement path

For a website action, open the actual landing page and complete the journey while observing the appropriate diagnostic tool. With Google Ads website tags, Tag Assistant can test the intended action; Google’s troubleshooting route starts from the landing page, not directly from a thank-you URL. Check both the real saved submission and the correct tag event. A thank-you message on screen proves neither by itself.

For a GA4-imported action, use DebugView to inspect the event and its parameters on your test device, then inspect the import/configuration in Ads. DebugView’s attribution is limited, and privacy or consent conditions can prevent events appearing there. For an offline action, inspect the actual upload results and diagnostics, including failures within an otherwise successful batch. Google distinguishes upload success from reported conversion credit.

A passing test shows that this journey worked under these device, browser and consent conditions. It does not prove that yesterday’s mobile visitors were measured, recover old missing events or guarantee an Ads-attributed conversion from your test. If mobile results are where the decline sits, test that journey too. Preserve the consent choices you are testing rather than treating consent-dependent absence as a broken tag.

Two dashboards are not independent witnesses when both rely on the same event. If a GA4 event stops firing and Ads imports it, both can fall while the form inbox remains busy. The inbox supplies the missing kind of evidence.

If the independent records do not exist

You can still check the current journey, the intended event, import errors and conversion settings. Ask the person receiving enquiries whether saved messages, booking records or call records can recover part of the affected interval. Keep that subset distinct from all leads, and do not invent channel attribution it lacks.

If history cannot be recovered, begin a dated record of actual outcomes now, independently of the analytics event, and compare it with prospective measurement receipts. This can detect a continuing problem; it cannot prove what happened last week. Meanwhile, repair any reproduced defect and keep spend within the business’s acceptable exposure. If no defect is found, label the historical split between measurement loss and business loss unresolved. That is a reason to collect the missing observation before a broad rebuild, not a reason to leave a broken journey running or to declare tracking healthy.

Give the clicks time to produce their outcomes

Now ask whether the acquired traffic produced fewer credited outcomes. Use ordinary interaction-date conversion columns for this comparison, so conversions stay with the clicks and cost that preceded them.

Recent clicks have had less opportunity to convert. To see the historical delay, Google’s documented route is a campaign, ad-group or Search-keyword report ending at least 30 days ago—or longer for a longer conversion window—then Segment → Conversions → Days to conversion. Bid-strategy reports can also indicate that more conversions are expected. Those expectations are estimates, not received outcomes. Google’s conversion-delay guide.

Choose equal-length periods with the same weekdays, the same action and unchanged definitions. For a completed retrospective comparison, allow the configured conversion window and relevant processing time to elapse after each period’s final interaction. If using a shorter age based on the account’s historical delay, state the remaining tail and treat the result as provisional. A past delay pattern need not hold through a new problem.

If the affected period is too young, save its report and set a specific revisit when the chosen observation age is reached. Use current business receipts and journey tests in the meantime. You cannot reconstruct “what last month looked like at age three days” from its fully updated total unless you kept that earlier snapshot. Without one, compare mature periods when available; do not manufacture a matched-age baseline.

Also count the clicks. “Still receiving clicks” can describe 1,000 clicks becoming 600. If the rate remained 5%, those totals would produce 50 versus 30 conversions. That is a traffic-volume question before it is evidence of worse conversion performance.

Look inside the total

Even equal click totals can hide a substantial change. Here is an illustrative Search campaign with two non-overlapping ad groups, Brand and General. The requests below are credited conversions for one form-submission action. Assume all clicks in this example are eligible for that conversion, both periods have matured, measurement is consistent, and these two groups contain all the traffic being compared.

Neither group’s conversion rate fell. The campaign bought far more of the group that converts at 2%, and far less of the group that converts at 10%. Its total fell from 84 to 36 requests because the mix changed.

The most useful calculation is to apply each earlier rate to its later clicks:

200 × 10% + 800 × 2% = 36 requests.

That is exactly the later total in the illustration. At this level of grouping, the mix accounts for the decline. Investigate why the allocation changed: did Brand lose traffic after a budget, targeting or negative-keyword change; did General expand; did demand shift? Inspect the corresponding settings history and search terms before choosing a remedy. A lower rate alone does not justify excluding General—it may still supply profitable new business that Brand cannot replace.

Now change just one observation. Suppose General produced 8 later requests instead of 16. Its rate is now 1%, and the campaign total is 28. The old rates applied to the new traffic still predict 36. Mix explains the movement from 84 to 36; another 8 requests are missing relative to that calculation.

The next useful investigation is inside General: its search terms, landing pages, devices, offers and measurement path. A new mobile-form error could explain a loss there; a successful desktop test would not clear it. If the terms changed substantially within the ad group, finer traffic mix may explain the difference. If comparable traffic deteriorated while the measurement path still represented outcomes consistently, investigate the journey, offer and demand conditions. The arithmetic locates the shortfall; it does not prove its cause or that a small sample represents a lasting change.

If a group had no earlier clicks, it has no earlier rate to apply. Show that new traffic separately. You can identify its contribution to the changed mix and inspect its observed outcomes, but cannot fill the historical gap with another group’s rate and call it a diagnosis.

Build this comparison in your account

To build this comparison in your account, use the ad-group table for the affected Search campaign. Use ordinary interaction-date columns for both periods, keeping conversions with the clicks that preceded them. Export the clicks for both periods with campaign and ad-group IDs.

Then use Segment → Conversions → Conversion action, export “All conversions,” and retain only the exact action you chose. This holds the action under observation even if its primary/secondary setting changed.

Join those counts to the traffic export using the same IDs and period. Conversion-action segmentation can leave clicks and cost blank; take those from the unsegmented export, once per ad group, rather than adding a denominator for every action. Google’s segment documentation.

For this click-based comparison, calculate each group’s requests divided by clicks. Across campaign types, Google’s conversion-rate metric uses eligible interactions, which can include more than clicks; do not combine unlike interaction types into this Search example. Keep the groups mutually exclusive and include the whole selected population, including a remainder group if needed.

Where the August 17 bidding change fits

For declines around that period, check Google’s August 17, 2026 target-bidding update. The relevant case is a campaign that was budget-constrained, used a cost per action (CPA) or return on ad spend (ROAS) target, and had been achieving a lower CPA or higher ROAS than that entered target. Google says affected campaigns now optimize more consistently toward the entered target; multi-channel allocation can also change. The update notice.

Check the strategy, the target in force, evidence of budget limitation during the affected interval, and mature achieved performance before and after it. An account notification is not enough to establish that combination. For example, a campaign spending $1,000 for 50 leads achieved $20 CPA. If its entered target was $40, movement toward $40 could mean about 25 leads at the same spend. That is an illustration of the relationship, not a prediction for your account.

If those conditions fit, review whether the entered target still represents what the business can pay. A tighter target may reduce spend or volume; it does not guarantee the old result. If the campaign was not budget-constrained, Google says this update does not change its target-based behavior. And Google specifically advises against applying data exclusions solely because of this update. The update FAQ.

The point of these checks is to leave with a finding specific enough to act on: a submission is being lost, a recent click period is incomplete, a traffic group lost share, or comparable traffic is producing fewer outcomes. Make the repair or bounded business decision that finding supports, record when it starts, and keep the evidence needed to check the result. A calmer chart later will not, by itself, tell you which explanation was right.