Conversion tracking that tells the truth

Automated bidding is extremely good at getting more of whatever it is pointed at. That single sentence is the reason measurement sits above structure, landing pages and bidding in every ordering I use.

Point Smart Bidding at a conversion action that fires on a newsletter signup, and it will go and find you people who sign up for newsletters. It will do this efficiently, relentlessly, and it will report excellent results while doing it. The account will look healthy. The bank account will not.

This is not a hypothetical failure mode. It is the default one, because marking every useful action as a conversion feels responsible. Newsletter signups, PDF downloads, contact-page views — each is genuinely useful information. None should be a primary conversion, because bidding will optimise toward whichever is cheapest to produce, and that is almost never the one that makes money.

So before touching a bid, a keyword or a landing page, the question to answer is: does the number in this account correspond to something the business actually received?

Most of the time, the answer is no, and nobody has checked.

Open Google Ads → Goals → Conversions → Summary. You are looking at the list of things this account believes are worth money.

For each action, work through four questions.

1. Does this represent money, or the possibility of money?

A completed purchase represents money. A qualified enquiry represents the possibility of money. A page view represents neither. Only the first two belong as primary conversions.

2. Is it primary or secondary?

Google splits conversion actions into primary (used for bidding, counted in the "Conversions" column) and secondary (observation only, reported separately, ignored by bidding). This distinction is the single most useful control in the interface and it is chronically underused.

You do not have to delete a soft conversion to stop it damaging your account. Move it to secondary. You keep the data, bidding stops chasing it.

A well-configured account usually has one or two primary actions and several secondary ones.

3. Is the counting setting right?

"Count" has two options and getting it backwards is common:

• One — count a single conversion per click. Correct for lead generation. One person filling in your form three times is one lead.
• Every — count every conversion per click. Correct for ecommerce, where a customer genuinely buying twice is worth twice as much.

A lead form set to "Every" inflates your conversion count with duplicate submissions and teaches bidding to buy people who fill in forms repeatedly. This is a two-click fix that materially changes what your account optimises for.

4. What is the conversion window, and is it honest?

The default click-through window is 30 days, and can be set up to 90. A longer window attributes more conversions to ads, which flatters the account. It is not wrong to use one — if your genuine sales cycle is 60 days, a 30-day window will systematically undercount.

The test is whether the window matches your actual buying cycle rather than whichever number makes the report look better. Set it deliberately, write down why, and do not change it mid-analysis.

Everything above is settings hygiene. This is the test that actually tells you whether your data is real, and almost nobody runs it.

Take last month. Write down the number in the Conversions column. Now go to whoever handles enquiries — the sales inbox, the CRM, the person who answers the phone — and count how many genuine enquiries arrived from the website in the same period.

Compare.

If Google Ads says 84 and the business received about 80, you are in good shape; the gap is modelling and cross-device attribution.

If Google Ads says 84 and the business received 30, then every optimisation decision made this year was based on fiction, and no further work on the account is worth doing until you find out why.

Where the gap usually comes from

In rough order of how often each turns out to be the cause:

• Soft actions marked primary. The 84 includes 50 newsletter signups.
• Duplicate counting. Form confirmation page reachable by refresh, or counting set to "Every".
• The tag firing on a page reachable without converting — a thank-you URL someone can navigate to directly, or a tag placed on the form page rather than the confirmation.
• Phone-click conversions counted as calls. A click-to-call tap counted regardless of duration counts misdials and accidental taps as customers. Google's call reporting lets you set a minimum call length; without one, this number is close to meaningless.
• Cross-device and modelled conversions. Legitimate, but they widen the gap and should be understood rather than assumed away.

Only the last of these is acceptable. The rest are bugs.

Do this quarterly, not once

Tracking breaks silently. A website redesign moves the thank-you page, a developer removes a tag, a form plugin updates. Nothing in Google Ads will tell you — the number simply changes, and by the time anyone notices, bidding has been learning from bad data for weeks.

Put the reconciliation in the calendar.

The cleanest setup in 2026 is Google Tag Manager firing a Google Ads conversion tag directly, with GA4 alongside it for analysis.

The important distinction, because it causes a great deal of confusion: conversions imported from GA4 and conversions tracked natively in Google Ads are not the same number and will not agree. GA4 uses its own attribution model and its own session logic. Google Ads uses click-based attribution.

Pick one as the source of truth for bidding. My default is a native Google Ads conversion tag, because it is the most direct measurement of the thing bidding is optimising for and has the fewest moving parts between the event and the bid. Use GA4 for understanding behaviour, journeys and assisted paths — the things it is genuinely better at.

What you should not do is import the same conversion from GA4 and track it natively, both marked primary. That double-counts, and it is a surprisingly common state to find an account in.

A workable implementation

1. GTM container on every page, with a documented data layer.
2. A trigger on the actual success event — the form's success callback, not a thank-you page URL that can be reloaded or reached directly.
3. Google Ads conversion tag on that trigger, with a conversion value if you have one.
4. GA4 event on the same trigger, for analysis.
5. Consent Mode v2 configured before either tag fires.

Attribution

Data-driven attribution is the default and is generally the right choice — it distributes credit across the path rather than giving everything to the last click. Last-click remains available and there are narrow cases for it, but if you are on last-click by inheritance rather than by decision, you are systematically undervaluing everything that happens before the final touch.

The caveat worth knowing: data-driven attribution needs data. On very low conversion volume it degrades toward something close to last-click anyway.

These two are frequently discussed as separate optional improvements. They are neither separate nor, for most advertisers, optional.

Enhanced conversions

Enhanced conversions send hashed first-party data — email address, phone number, name, address — alongside the conversion event. Google matches the hash against its own signed-in users and recovers conversions that would otherwise be lost to cookie restrictions, cross-device journeys and delayed conversions.

The hashing happens before anything leaves the browser (SHA-256), so you are not transmitting the address itself. Reported uplift is commonly cited in the 5–15% range, though the real figure depends heavily on how much of your audience is signed in to a Google account and how much first-party data you can pass.

The important part is not the uplift in the report. It is that recovered conversions feed bidding, so the algorithm is making decisions on a fuller picture. That compounds in a way a reporting uplift does not.

Turn it on. There is very little argument against it, and the setup for a lead-generation site is usually a matter of passing the email field from the form into the conversion tag.

Consent Mode v2

If you serve users in the EEA or the UK, Consent Mode v2 is a legal requirement, not a recommendation. It introduced two parameters beyond the original consent signals — ad_user_data (may user data be sent to Google for advertising) and ad_personalization (may ads be personalised).

When a user declines, Google's tags do not simply stop. They fire in a restricted mode sending cookieless pings, and Google uses conversion modelling to estimate what would have been measured with consent. Without Consent Mode v2 configured, you get neither the consented data nor the modelling — you get a hole.

Two things that catch people out:

• Enhanced conversions will not work correctly without Consent Mode v2 wired properly. They are dependent, not parallel.
• As of the April 2026 GA4 update, correct Consent Mode v2 configuration became a hard requirement for GA4 data to flow accurately into Google Ads conversion tracking in the EEA and UK, and behavioural modelling thresholds tightened globally.

If you serve India, the US or other non-EEA markets exclusively, this is less urgent — but "we do not really get European traffic" is worth verifying in your analytics rather than assuming.

Everything so far measures the lead. For any business where leads are qualified by a human before becoming customers, that is the wrong optimisation target, and the gap between the two is where most B2B and high-value local accounts quietly fail.

Consider two campaigns. Campaign A produces 100 leads at ₹500 each. Campaign B produces 40 leads at ₹1,200 each. On cost per lead, A wins comfortably and any sensible bid strategy will shift budget toward it.

Now add close rates. A closes at 2%, producing two customers at ₹25,000 each. B closes at 20%, producing eight customers at ₹6,000 each. B is four times better and the account has been actively defunding it.

Offline conversion import fixes this by sending the outcome back.

How it works

1. Capture the GCLID on the landing page — Google appends it to the URL of every ad click — and store it against the lead in your CRM.
2. When that lead progresses (qualified, quoted, closed), upload the GCLID back to Google Ads with a conversion action and, ideally, a value.
3. Bidding now optimises toward qualified pipeline rather than form fills.

Uploads can be scheduled from Google Sheets, sent via the API, or handled by most mainstream CRMs natively.

If you cannot capture GCLIDs — some form platforms make it awkward — enhanced conversions for leads offers a route using hashed email or phone instead, matching the uploaded outcome back to the original click without needing the click ID.

The realistic version

Full offline import is a project, and for many businesses it is genuinely the highest-value work available. But the crude version delivers most of the benefit and takes an afternoon: create two extra conversion actions, Qualified Lead and Closed Won, upload to them monthly from a spreadsheet, and mark Qualified Lead as primary once it has enough volume to bid on.

Even without automation, simply knowing which campaigns produce customers rather than enquiries changes how you allocate budget — and that decision is worth more than the automation.

The question this all comes back to

When someone becomes a customer, do you know which click brought them?

If yes, paid search is a solvable engineering problem. If no, then closing that loop is the priority, ahead of any campaign work — because spending into a channel you cannot measure produces a number at the end of the month that nobody can interpret and an argument about whether it worked.

A checklist to audit against. If you can tick all of these, measurement is not your problem and you should move on to structure.

Conversion actions

• One or two primary actions, each representing real business value
• Soft actions present but marked secondary
• Counting set to "One" for leads, "Every" for sales
• Conversion window matched to the actual buying cycle
• No conversion imported from GA4 and tracked natively as primary

Implementation

• Tags fire on a genuine success event, not a reloadable URL
• Phone conversions have a minimum call duration
• Enhanced conversions enabled and passing first-party data
• Consent Mode v2 configured, if you serve EEA or UK traffic
• GA4 linked, used for analysis rather than as the bidding source

Verification

• Last month's conversions reconcile with what the business received
• Reconciliation is scheduled quarterly, not done once
• Somebody checks tags still fire after every site release

Revenue loop

• GCLID captured and stored against each lead
• Qualified and closed outcomes uploaded back to Google Ads
• Bidding optimises toward qualified pipeline, not raw enquiries

None of this is difficult. Almost all of it is the kind of work that falls off the list when the week fills up — which is precisely why it is where the money is.