The conversion tracking setup most Google Ads accounts get wrong
by Harry Watts · · updated · 7 min read
The report says 47 conversions. You had maybe nine enquiries. Nobody's lying: the account is faithfully counting something, just not the thing you'd call a customer.
This is the fault I look for first in every audit, before bids, before copy, before structure. Not because it's the most interesting, but because everything else is measured with it. An account optimising toward the wrong signal doesn't stall — it confidently gets worse, and the reporting looks fine the whole way down.
Four faults cause most of it. Each takes minutes to check.
1. Soft actions counted as conversions
The most expensive one, and the most common. A page view of the contact page. A click on a phone number. A newsletter sign-up. Someone reaching 75% scroll depth. All useful signals — none of them a customer.
When these sit in the Primary column, two things happen. Your cost per conversion looks a fraction of its real value, so the account appears profitable when it isn't. And smart bidding, which optimises toward whatever is set as primary, learns to chase the easiest action available. That's always the soft one. Given a choice between finding someone who'll buy and finding someone who'll scroll, the algorithm does exactly what you asked and finds scrollers.
Where to look: Goals → Conversions → Summary. Read the Primary column. Only genuine business outcomes belong there: enquiry forms, purchases, calls over a meaningful duration. Move everything else to Secondary, where it's still recorded for observation but no longer steers bidding.
A phone-number click deserves special mention: it records the tap, not the call. On mobile a large share never connect, and none of them tell you the call lasted more than four seconds. If calls matter, use call tracking with a minimum duration.
2. The same action counted twice
One form submission, two conversion actions firing: a Google Tag Manager trigger and an imported GA4 event describing the same event. Cost per conversion instantly reads half its real figure.
It's rarely deliberate. It's what happens when tracking is set up once, then set up again by someone new who didn't know the first version existed, an agency handover being the classic cause.
Where to look: in the same Conversions list, read the action names and sources side by side. Two rows with similar names and similar volumes, one sourced from a website tag and one from an analytics import, is the signature. Then submit a test enquiry through your own site and watch: it should appear once.
3. The count setting nobody changes
Each conversion action has a Count setting: Every or One. It decides what happens when the same person converts repeatedly after a single click.
- Lead generation → One. Someone submitting your enquiry form three times because they weren't sure it sent is one lead, not three.
- E-commerce → Every. Three separate orders are three sales, and each carries revenue.
Get this wrong on a lead-gen account and you don't just inflate the numbers: you teach the bidding algorithm that repeat submitters are the goal, and it goes looking for more of them.
Where to look: click into each conversion action; Count sits in the settings panel.
4. Consent gaps quietly deleting conversions
The newest fault, and the one most often missed because it's nobody's obvious job. Under UK PECR and GDPR, advertising and analytics cookies need consent. If a visitor declines and your setup has no answer for that, their conversion is never recorded, even though the sale happened.
The tell is dramatic and easy to misread: conversions drop sharply on the day a cookie banner goes live, and the account looks like it broke. It didn't. It just stopped seeing part of reality.
Google Consent Mode v2 is the fix: cookieless signals while consent is denied, plus modelling to estimate what was missed. It has to be installed correctly, with the consent defaults declared before the tag manager loads; declared afterwards, tags have already fired and the whole exercise is decoration. (This site runs exactly that setup; view source if you want the working example.)
Where to look: plot conversions against the date your banner went live. A step change on that date, with no corresponding fall in clicks or enquiries, points here.
What good looks like
- Primary conversions are outcomes you'd recognise as business, nothing else
- Every action appears exactly once, from one source
- Count matches the model: One for leads, Every for e-commerce
- Consent Mode v2 configured, defaults declared before the tag manager loads
- Enhanced conversions on, so hashed first-party data recovers matches browsers would otherwise lose
- For long sales cycles, offline conversion import feeding CRM outcomes back, so bidding optimises toward closed deals rather than form fills
- Recorded conversions and real enquiries agree within a margin you can explain
Why this comes before everything else
Fixing tracking rarely shows up as a saving on its own. It shows up as every subsequent decision becoming correct: which keywords to cut, which campaigns to fund, whether the channel works at all. Run the wasted-spend checklist against bad conversion data and you'll get confident, well-reasoned, wrong answers. Do the cost-per-customer arithmetic on inflated conversions and you'll conclude paid search is cheaper than it is, then scale into a loss.
It's also the check that tells you most about whoever runs your account. Someone billing for management on an account where page views count as conversions either hasn't looked, or has looked and preferred the reporting. Neither is good. It's the reason tracking sits near the top of the list of things worth checking yourself.
Quick answers
Why don't my conversions match my actual sales?
Soft actions set as primary, the same action counted twice, count set to Every on a lead-gen account, or consent gaps dropping real conversions. Compare one week of recorded conversions against enquiries that actually reached your inbox: the direction of the gap tells you which.
Count Every or One?
One for lead generation: three form submissions from one person is one lead. Every for e-commerce: three orders are three sales.
Primary or secondary?
Primary counts in your Conversions column and steers smart bidding. Secondary is observation only. Real outcomes go in Primary; page views, scrolls and sign-ups go in Secondary, or bidding will chase whichever is easiest to generate.
Does cookie consent affect tracking?
Yes. Declined consent means conversions aren't recorded normally. Consent Mode v2 sends cookieless signals and models the gap. Without it, conversions fall off a cliff the day your banner goes live and the account looks broken when it isn't.
How do I check mine?
Goals → Conversions. Check what's Primary, the count setting on each, whether two actions describe the same event, and the Status column. Then send a test enquiry and confirm it lands once.
If the numbers don't add up
Untangling tracking somebody else configured is slow, specialist work, which is why it's a separate piece rather than something an audit fixes in passing. Start with a free intro call: thirty minutes is usually enough to tell whether you have a tracking problem or a performance one, and they need very different answers.