Conversions Disappeared: A 30-Minute Conversion Tracking Diagnosis

You open the account and yesterday shows zero conversions. Or the count halved with no campaign edits. Or Google Ads reports 340 leads while the CRM holds 190. These are three different problems with three different fixes, and the expensive mistake is the same in all of them: touching bids and budgets before establishing whether the drop is real.

What follows is a conversion tracking diagnosis you can run in about thirty minutes, from conversion action status through to CRM reconciliation. Order matters — each step is only meaningful if the previous one came back clean.

Separate three different problems first

“Conversions disappeared” hides three fundamentally different scenarios.

Symptom	Most likely class of cause	Start at
Zero conversions where there were dozens yesterday	Technical break: tag, site release, changed conversion action	Steps 1–3
Conversions down 20–50%	Partial loss: consent, one page type, one of several sources	Steps 2, 5, 6
Ads and CRM disagree but move together	Methodology: attribution, windows, deduplication	Steps 4, 7
Conversions intact, lead quality collapsed	Not tracking — training signal and traffic mix	A different investigation entirely

Rule for the first hour: until tracking is confirmed healthy, do not touch bids, budgets or bidding strategies. Reacting to a phantom drop costs more than the drop.

Step 1. Conversion action status — 3 minutes

Open Goals → Conversions → Summary and read the status column for every action:

• “Recording conversions” — data is flowing; the problem is elsewhere.
• “No recent conversions” — the action exists but nothing has fired for a while. First suspect.
• “Unverified” / “Tag inactive” — no events arriving at all. Almost always the website, not the account.
• “Inactive” — someone disabled it manually. Check change history; actions often get switched off during a duplicate cleanup.

Then check three settings that quietly rewrite the numbers:

1. Is the action still a primary goal? Moved to secondary, it stops appearing in the Conversions column and stops feeding bidding, while data keeps collecting. The split is explained in primary vs secondary conversions and lead quality.
2. Counting setting. Switching between “every” and “one” changes totals several times over on repeat purchases and multi-step forms.
3. Conversion window. Cutting a window from 30 days to 7 in a long-consideration market looks exactly like a crash in reporting.

Step 2. Is the event leaving the website? — 7 minutes

This step separates “the event never fires” from “it fires but never arrives”.

1. Walk the conversion path yourself. In debug mode (Tag Assistant or tag manager preview), complete the action on the live site and watch whether the tag fires and with what parameters.
2. Test every entry point, not one page. The classic: the footer form works, the modal form broke in last week’s release, and mobile serves a different template entirely.
3. Check the thank-you page. If the conversion counts on a thank-you page load and the post-submit redirect was replaced with an inline success message, the event vanished while the leads kept arriving.
4. Cross-check analytics. If the event appears in GA4 realtime but never reaches Google Ads, the break is in account linking or goal import, not in the tag.
5. Check consent state. It is easy to miss in debugging that tags do not fire until the cookie banner is accepted.

If the tag fires but data arrives only partially, the next suspects are browser restrictions and lost identifiers. That is precisely what enhanced conversions and server-side tagging with sGTM exist to recover.

Step 3. Duplicates and double counting — 5 minutes

The inverse problem is just as common: suspiciously many conversions. Typical sources:

• the tag deployed both hard-coded and through the tag manager;
• the same goal imported from analytics and configured natively in Google Ads;
• the conversion firing on thank-you page load while users refresh or hit back;
• a form submit firing both client-side and server-side with no shared transaction ID for deduplication.

The test is simple: compare account conversions with actual enquiries for the same day. A stable factor of 2.0 is a duplicate, not a lucky week.

Step 4. Conversion windows and lag — calendar, not panic

The most common false alarm. Conversions are credited to the click date, not the event date. Yesterday and the day before will always look worse than they eventually become, because part of the data has not landed yet.

In short-cycle markets — delivery, on-demand services — lag is measured in hours and the last day is nearly final. In B2B, real estate and education, 30–50% of conversions arrive three to fourteen days after the click. Judging yesterday’s ROAS there means deciding on half your data.

Measure your own lag once and set a rule: “data is considered settled after seven days”, for example. The mechanics are covered in conversion windows and conversion lag.

Step 5. Consent and privacy — 5 minutes

A drop that starts exactly on the day the consent banner changed is the most recognisable pattern in the whole diagnosis. Check:

• whether the banner’s design or logic changed — a more prominent “Reject” button drops consent rates immediately;
• whether consent mode is wired correctly and state signals are actually transmitted;
• whether conversion modelling is active and has enough volume to engage;
• whether a new script now blocks tags before consent in regions where it previously did not.

The goal is not to work around consent but to size the loss and compensate legitimately: enhanced conversions, server-side collection, offline imports. Details in Consent Mode v2 and advertising privacy.

Step 6. What actually changed — 5 minutes

Two timelines to overlay on the drop:

1. Google Ads change history. Who touched conversion actions, bidding strategies and statuses, and when. “Nobody changed anything” is frequently untrue. The workflow is in using change history to diagnose performance drops.
2. Site releases. Ask engineering for deployments in the affected window. A CMS template change, a moved form, a new CDN, altered post-submit redirects — each breaks tracking with zero edits in the ad account.

Check URL tracking parameters separately: a stripped or overwritten click identifier severs the link between click and conversion. How those identifiers work is in GCLID, GBRAID, WBRAID and tracking templates.

Step 7. Ads versus CRM: they will never match

A gap between Google Ads and the CRM is normal. What matters is its size and its stability.

Cause	Direction	Typical size
Click-date attribution vs enquiry-date reporting	Ads leads CRM in time	Depends on lag
Cross-device and modelled conversions	Ads higher	Single to double digit %
Click identifier lost in transit	Ads lower	5–20%
Spam and test submissions	Ads higher	Depends on filtering
Calls and messenger enquiries	Ads lower	Significant in services

The practical approach is not to chase parity but to establish a normal variance factor — say, Ads runs 15% above CRM — and monitor deviation from it. A jump in the factor is a signal; the gap itself is not. The proper way to close the loop is importing offline conversions from your CRM back into Google Ads, so bidding learns from confirmed deals rather than raw enquiries.

Segmenting: where exactly did conversions go?

Before fixing anything, localise the loss. Identical-looking drops have very different causes. Four cuts take five minutes and narrow the suspects sharply.

Cut	What you see	What it means
By conversion action	One action gone or all of them	One — a specific goal or page. All — tag, consent, global release
By device	Mobile down, desktop fine	Mobile template, different form, page speed
By campaign and network	One campaign only	Traffic, not tracking: bids, budget, auction
By geography and language	Specific countries only	Regional consent banners, localised site versions

The logic: conversions gone everywhere at once is technical. Gone in one place, gradually is traffic or market. Splitting those two in the first minutes stops you spending a day on tags when impression share is the real story. Building this as a saved report rather than a manual export is covered in custom columns and the report editor.

Three failure profiles you can recognise by shape

Profile 1: the cliff

Steady conversions, then zero on a specific day, and it stays there. Spend unchanged. Always technical: a release, a disabled action, a removed tag. Fifteen-minute diagnosis starting from action status and the deploy log. Speed matters here: automated bidding learning on an empty signal degrades a campaign faster than the missing data itself.

Profile 2: the step

Conversions drop 20–40% and hold at the new level. Spend and CTR unchanged. That is partial loss: a new consent banner, a broken mobile form, one of two event sources dropping out. Find the cut where the loss concentrates.

Profile 3: the slide

A gradual weekly decline with CTR and impression share moving too. Not tracking — market and auction: a new competitor, seasonal softening, fatigued ads. Fixing tags will not help; read impression share and auction insights instead.

Cliff means technical. Step means partial data loss. Slide means market. One minute spent identifying the profile saves half a day of diagnosis in the wrong direction.

Calls, messengers and offline

If a meaningful share of enquiries never touches a form, standard diagnosis covers half the picture. Nothing is broken in the tag; the handoff chain is. The symptom reads as “advertising stopped producing customers” while customers keep arriving — they are simply invisible.

• Calls. Verify that the dynamic number renders on every page template and that a static number has not been cached by the CDN after a release. Check pool size at peak hours too: when numbers run out, some visitors see the generic line and the click association is lost. Mechanics in call tracking and call campaigns.
• Messengers. Jumping into a messenger ends the session; without a parameter carried in the first message or a unique deep link there is nothing to reconstruct the source from.
• Offline deals. If only enquiries return to the account and not confirmed sales, bidding optimises for volume rather than value. Symptom: plenty of leads, flat revenue.

Cause frequency: where to start statistically

1. A website change nobody told the marketing team about — the most common cause of a sudden zero.
2. An edit to the conversion action itself (window, counting, moved to secondary).
3. Conversion lag mistaken for a drop.
4. Consent banner changes and a shifted consent rate.
5. Duplicates after the tag was deployed twice by different people.
6. Click identifiers lost to redirects and parameter stripping.
7. A genuine fall in demand or competitive position — sometimes it really is just that.

Conversion tracking diagnosis: the 30-minute protocol

1. 0–3 min. Conversion action statuses, exact start date of the drop, spend on the same days.
2. 3–8 min. Cuts by action, device, campaign and geography to localise the loss.
3. 8–15 min. Walk the conversion path in debug mode on the live site, mobile included.
4. 15–20 min. Change history plus the site deployment log for the period.
5. 20–25 min. Consent: banner changes and consent rate.
6. 25–30 min. CRM reconciliation on a settled period against your normal variance factor.

Preventing the next one

• One number, checked daily. A saved report with conversions per action for yesterday and the day before. A zero on an action that normally fires should be obvious on day one.
• A watchdog rule. An automated rule that alerts when a campaign records zero conversions with non-zero spend catches most technical failures. Setup patterns are in Google Ads automated rules.
• A release protocol. An agreement with engineering: any change to forms, thank-you pages, redirects or the consent banner triggers a notification and a tag check.
• A quarterly tracking audit. Every conversion action reviewed: what it counts, where duplicates hide, which goals are obsolete. A ready checklist is in the Google Ads account audit checklist.
• A baseline document. Normal CRM variance, typical lag, consent rate. Without a baseline there is no way to tell an anomaly from a Tuesday.

What a healthy setup looks like

It helps to know what you are aiming at, because “tracking works” is not a status anyone can verify. A healthy measurement setup has five properties, and each is checkable in minutes.

• One source of truth per conversion. Every important action is counted once, by one method, with a documented owner. No goal exists both natively and as an import unless deduplication is deliberate.
• A known lag. You can state, from data, how many days it takes for a cohort to settle, and reporting respects that window rather than reacting to yesterday.
• A known CRM variance. The gap between platform and CRM has a number attached to it, and deviation from that number triggers investigation.
• Value flowing back. Confirmed revenue or qualified-lead status returns to the account, so bidding optimises toward money rather than form fills.
• An alarm that fires on its own. Nobody should discover a tracking outage by opening a report on a Monday.

Teams that keep those five properties documented spend their diagnosis time on the actual anomaly instead of re-deriving what normal looks like every time something moves.

When tracking is not the problem

Sometimes everything checks out and conversions really are down. Then look at reach: impressions, impression share, auction dynamics. The order of checks is in why your ads are not showing. If volume held but audience quality dropped, start with the keyword set instead — see building a keyword set in Keyword Planner. For accounts where the configuration history is impossible to reconstruct, an outside review is faster; that is part of what our Google Ads agency work covers.

FAQ

Conversions dropped to zero in one day. What do I check first?

Conversion action status and the site deployment log for that date. A hard zero is almost always technical rather than market-driven.

Why does Google Ads report more conversions than GA4?

Different attribution models, different windows, different modelling and cross-device handling. A 10–30% gap between the platforms is ordinary; comparing them one-to-one is not meaningful.

What does “No recent conversions” actually mean?

The action is configured but nothing has fired for a noticeable period. For rare goals that is normal. It is alarming only if the action used to fire daily.

How long should I wait before treating a drop as real?

As long as your typical conversion lag. Hours in short cycles, seven to fourteen days in long ones. Measure it once and make it a rule.

How do I know conversions are being double counted?

Compare account conversions against actual enquiries for a single day. A persistent 2:1 ratio or exact multiples indicate double counting.

Can a cookie banner change really move conversion volume?

Directly. Button prominence and banner logic change consent rates, and consent rates change observable conversions. Always check the banner’s change date against the graph.

Do I need enhanced conversions if the tag works fine?

Yes. They recover events lost to browser restrictions and improve the completeness of the data your bidding learns from.

Can I just import conversions from analytics instead of configuring them natively?

You can, but watch for duplication if the same event also exists natively, and remember the attribution models differ. Pick one source of truth for optimisation.

Why does the CRM show fewer leads than the account?

Spam, tests, incomplete submissions, duplicates from the same person, plus cross-device and modelling on Google’s side. The stability of the variance factor matters more than its absolute value.

Is a zero-conversion automated rule worth setting up?

Yes — it is the cheapest way to catch a technical failure on day one. The condition “spend above zero and zero conversions in 24 hours” covers most cases.

What if engineering keeps changing the site without warning?

Formalise it: a list of tracking-critical pages and elements, plus a mandatory tag check after deployment. One document beats weekly arguments.

How often should a full tracking audit run?

Quarterly for a stable account, and after every major site release, on top of a daily automated check of the basic number.