Google Ads Conversion Tracking: How to Verify Yours Is Measuring Something Real

2026-09-28

The pages I could retrieve for this term — Google's own product and Help Center pages, the Ads API documentation, a Shopify explainer, a Tag Manager tutorial — concentrate on installation: place the tag, define the action, fire the event, confirm it appears. Several of those extracts were partial, so read that as a description of what was in front of me rather than a verdict on everything published. Either way, the installation work matters and the official documentation covers it well. But installation is not the same as measurement. A tag can be installed correctly and still produce a number you cannot safely act on.

This article deals with the part after setup: how to establish that the conversions in your account correspond to real business outcomes, what each figure actually counts, and how much evidence a given decision needs before you spend against it.

Where conversion tracking sits in diagnostic order

When you are working out why a campaign is not producing revenue and there is no obvious broken thing, the useful order of investigation runs: does a qualified, reachable market exist at economics the business supports; is the data real enough to make decisions on; does the right person understand the offer; is the right person arriving in learnable volume; does the path from arrival to purchase hold together; is the value of non-buyers being captured; and do the unit economics support spending more.

Conversion tracking is the second of those. That position is the whole point. Each stage is a precursor to the ones after it, so a failure at measurement readiness makes any conclusion you draw about traffic, pages, or economics unreliable. If the event is misdefined or not firing, any decision that rested on that number rested on something other than the action you believed it was counting.

So work this stage before you rewrite an ad, rebuild a landing page, or raise a budget.

Write down what the event counts, and what it divides by

Before you use any conversion rate in a decision, state in writing exactly what behaviour counts as the conversion and what population it is divided by. "Conversion rate" means different things in different tools and to different people in the same company. If a figure sounds impressive or alarming and nobody can say what it counts, you do not yet have a metric — you have a number.

Do the same for the word "conversion" itself inside your account. A form submission, a button click, a thank-you page view, and a completed payment are four different events with four different meanings. Naming all four "Lead" in the interface does not make them equivalent.

Skip this if every metric in play already has a written definition with an explicit denominator.

Perform the action yourself and watch for the event

If an expected conversion is not appearing — or dropped to zero, or a new variant is live and reporting nothing — complete the action on a real device and confirm the event arrives before concluding anything from its absence. Fill in the form. Click the button. Pay with a real card if that is what the event tracks. Then look.

Be precise about what that test buys you, in both directions. If the event lands, you have established that it fires on the path, device and occasion you tested — not that tracking is complete across every other path, free of duplicates, or attributing correctly. If it does not land, you have established that this action, on this path, on this device, on this occasion, was not recorded. That is a thread to pull, not a verdict on your history: a consent prompt, one browser, a single broken variant or an intermittent fault can all produce a failure on one route while others record fine. Widen the test — another device, another browser, another entry path — until you know how far the gap extends. Either way it answers one question cleanly, which is more than a dashboard argument does.

Skip this if the event has been verified since the last change to the page, the funnel, or the tracking. Any change resets that.

Instrument every point where the visitor changes tools

Draw the actual technology path a visitor takes from ad click to outcome, and assign a verified event to each point where the domain, platform, or tool changes. Ad platform to landing page builder, page to booking tool, booking tool to payment processor, payment processor back to your CRM — each of those boundaries is a place a signal can fail to cross.

Watch particularly for the thank-you page used as a proxy for an action it does not prove. A confirmation page view tells you a page loaded. It does not tell you a payment cleared or a booking held.

Skip this if the funnel runs on one platform with no boundary crossings and completion is verified.

Reconcile the whole count chain, not just the last step

Line up impressions, clicks, page views, sessions, on-page actions, leads, and bookings or purchases for the same period, and investigate the largest unexplained gap. Do this before concluding anything about conversion, because it tells you which stage to look at rather than which one you assumed.

One gap deserves specific mention: if clicks are not becoming page views — the ad platform reports far more clicks than your behaviour tool reports sessions — measure page load on a real mobile device and check redirects and script placement before attributing anything to the content. Load time, redirects and script placement can stop a conversion event firing, so treat them as conversion variables rather than as cosmetic concerns.

Skip the reconciliation if the chain has already been reconciled for this period and the gaps are explained. Skip the speed and redirect check if the click-to-page-view ratio is already healthy and load times are known to be fine on mobile.

Decide which system is authoritative for each outcome

For every outcome you care about, identify which system actually receives it — the CRM, the calendar, the checkout, the payment processor — and reconcile Google Ads reporting against that system. Platform-reported conversions and your own records can disagree. When they do, the system that receives the money or the booking is the one that settles the question.

This matters in two specific moments: when you are about to act on platform-reported conversions, and when someone says there are no leads. In both cases, check the authoritative record before you act on the dashboard. If your records and the platform disagree by an amount nobody has explained, reconciling your CRM pipeline numbers against what the platform reports is the work that has to happen first.

Skip this if platform numbers and first-party records already agree within an understood margin.

Check the reporting context before you diagnose from it

Before interpreting any performance data, verify the date range, timezone, active filters, campaign status, destination URL, which variant is live, the event configuration, and the sample size. Do this especially when something looks suddenly and dramatically wrong, or a campaign appears inactive, or a page appears to have collapsed overnight. A timezone mismatch and a genuine collapse look identical in a screenshot.

Keep a written register of the operational abnormalities too — setup errors, outages, budget accidents, tracking defects, broken variants — with dates. Then report performance both including and excluding them. If an average looks strange and something operationally abnormal happened in the period, that register is what stops you diagnosing a campaign problem that was actually a two-day tracking outage.

Skip the context check if you have already run it on this data set and nothing has changed since. Skip the anomaly register if the period was operationally normal and you can say so.

Choose the event you optimise toward on depth against signal

Your conversion action list and your optimisation target are separate decisions. Pick the optimisation event by trading outcome depth against signal density: the deepest real business event that still receives enough volume to learn from.

Three situations tell you the trade is worth reassessing. Optimisation pointed at a cheap shallow event while lead quality is poor. Optimisation pointed at an event so sparse that there is very little to learn from. Optimisation pointed at an event that is not firing at all. Each is a reason to review the target, not a verdict on what caused the performance problem — audience, offer, message and funnel execution all remain live possibilities. And treat a change to the optimisation event as a change to measure: record what it was, when you switched, and what happened to the downstream outcome afterwards.

Where the real outcome is too rare to learn from at your current volume, a shallower proxy event is a reasonable concession — but keep it labelled as a proxy and track the real outcome separately and continuously. A proxy that quietly becomes the thing being optimised stops being a measurement shortcut and becomes the objective.

Skip the proxy entirely if you have enough volume at the real outcome. Then use the real outcome; a proxy is a concession to sparsity, not a preference.

Make the number decision-ready before you act on it

A verified event still does not guarantee a usable figure. Four things can stand between "the tag works" and "I can decide on this."

• Counts, not percentages. Before declaring anything, look at the actual number of conversion events and ask what one more event would do to the rate. Decide how much evidence the decision requires before you look at the result, based on how expensive and how reversible it is. Confirming that an event fires needs one successful test. A large, hard-to-reverse increase in spend needs considerably more.
• Mixed populations. Report outcomes separately by source, period, destination version, and offer version. Keep the combined total for operations, never as a test result. Blend cold and warm traffic into one conversion rate and that rate cannot tell you which one moved. The same applies when the offer or the campaign objective changed mid-period.
• Immature cohorts. Where buying takes time, link each sale back to the enquiry that started it and report cohorts that have not finished converting separately from matured ones. If the cohort has not matured, you are reading a half-finished result and calling it a verdict. This is also why a daily view can show more sales than enquiries and look like a tracking error when it is not.
• Units. If you are importing conversion values and calculating return, align currency, time window, and the meaning of "revenue" before dividing one number by another. Two figures gathered under different definitions produce a ratio that describes neither.

If you are working from a conversion rate specifically, the same discipline applies to the denominator you chose — the checks that make a landing page conversion rate decision-ready are the checks that make a Google Ads conversion rate decision-ready.

When a recorded conversion is a stage rather than the result

Some funnels end at the payment. If yours does — someone buys, the product ships or downloads, and there is nothing after it that determines whether the customer got the result — then the recorded conversion is the outcome and this section is not yours.

Where meaningful stages do sit after the tracked event, track them separately: payment, attendance, participation, completion, and outcome. A small payment does not secure the stages after it. If paid registrations are being counted as attendees, or a purchase is being treated as the final outcome while non-completion goes unmeasured, your conversion tracking can be accurate while your picture of the business is still wrong.

Skip this if those later stages either do not exist in your funnel or are already tracked separately.

Run the pre-launch check every time

Before anything goes live, check budget type, amount, dates, projected spend, optimisation event, audience, destination URL, and tracking — then re-check the live settings immediately after publishing. This is a gate rather than a diagnosis. It applies whenever something is about to launch, and you can skip it only if the check has already been run on this exact configuration since the last change.

If these checks do not isolate the cause

Working through this list does not prove your tracking is sound by elimination. Insufficient data, execution problems, timing, and interactions between several of these factors can survive every check above. What the checks do is tell you where to look, not what to conclude — and they rule out a set of specific, known ways a conversion number ends up describing something other than what you assumed.

If you reach the end and the numbers still do not reconcile with what the business is experiencing, the honest position is that the evidence is not yet sufficient to diagnose it, and the next step is further first-party investigation rather than another round of optimisation against the same figure.

If you have worked through the verification and still cannot tell which stage is losing the money, GrowthHackr's Campaign Leak Audit is a free diagnostic pass over the same sequence, aimed at locating the stage where the evidence actually breaks down.