Google Ads Conversion Tracking an Online Store Can Trust

Make the paid order your only primary conversion. Send it from the order confirmation page with the real order amount, the currency and a unique order ID. Keep add-to-carts, calls and imported copies of the same purchase as secondary, so each order counts once and Smart Bidding learns on sales. Then test the numbers, not just the tag: the right event, real values, no second goal counting the same order and no outlier amounts. Only then read revenue and ROAS in the report.

What does conversion tracking do in a store account?

Google describes the mechanism in How Google Ads tracks website conversions. When someone clicks your ad, Google Ads stores information about the click in a cookie. When that person converts, your tag reads the cookie and sends the conversion back to Google Ads. Each thing you count is a conversion action: a purchase, a call or a sign-up, each with its own value, counting and window settings.

For a store, these numbers do two jobs. First, they fill the reports you judge campaigns by: how many orders came in and their conversion value. Divide that value by ad spend and you get your ROAS.

Second, they are what Smart Bidding learns from. Google’s help on primary and secondary actions warns that misconfigured settings can stop Smart Bidding from optimising effectively, because it relies on these signals.

So a wrong number travels beyond the report: it changes what the campaigns buy next. When a store bids on order value rather than on the number of orders, bidding aims at that value, so the value has to be right. Why Google Ads, GA4 and your store still show different totals for the same week is a separate topic: attribution for online stores.

Which way should a store send purchases to Google Ads?

A store has four routes for sending purchases. They can run side by side, but only one of them should count the purchase that campaigns learn on.

The Google tag or Tag Manager. Google’s tracking guide recommends a site-wide Google tag. A conversion tag consists of that Google tag and an event snippet. With Tag Manager, Google asks for a Conversion Linker tag on all pages: it stores the ad click information in cookies on your domain. What the linker does and how to see that it works: Conversion Linker tag.

If both gtag.js and Tag Manager run on your site, Google suggests you check for conflicts and move to one method, preferably Tag Manager. How to send the purchase from the dataLayer with the transaction ID and the real value: the purchase event in Google Tag Manager.

GA4 import. According to About conversion measurement, you can import only events marked as key events in GA4. New import options can take 24 to 48 hours to appear. The same page recommends the Google tag if you need Ads-specific features such as enhanced conversions.

According to Google’s tracking guide, the GA4 tag does not keep the click ID in a cookie for Google Ads conversion tracking. A purchase measured in GA4 therefore reaches Google Ads through import.

Uploads. Orders taken by phone or confirmed offline never fire a web tag. They reach Google Ads through an upload, matched by click ID or by hashed customer data: phone and offline orders in Google Ads.

Google’s tracking guide lists four basics that every route needs:

1. Auto-tagging is on in every Google Ads account.
2. Click trackers and server-side redirects pass the click ID (GCLID) on to your landing pages.
3. If the order confirmation page sits on another domain, a cross-domain linker passes the click ID to it.
4. Tags do not fire from inside an iFrame.

What should count as a conversion for a store?

The paid order. Google splits conversion actions into primary and secondary in About primary and secondary conversion actions. Primary actions appear in the “Conversions” column and feed bidding, as long as bidding uses their goal. Secondary actions are for observation: they show in “All conversions” and stay out of bidding, unless they sit in a custom goal.

Google gives a store example in About conversion goals. If several goals track stages of the same funnel, set only one as the account-default goal. For an online shoe retailer, that goal is “Purchase”.

Add-to-cart, checkout start and page views stay secondary. Why a campaign that learns on add-to-cart buys carts instead of orders: primary and secondary goals for a store.

What a campaign bids on also depends on its goals. It uses either the account-default goals or goals of its own. How to keep calls, directions and store visits out of the goals your Shopping and Performance Max campaigns learn on: campaign-specific conversion goals.

Count every purchase. According to About conversion counting options, “Every” counts each conversion after an ad interaction and suits sales. It is the default for website actions. “One” counts a single conversion per click and suits leads.

Four actions sit next to the purchase, and each needs its own decision:

• Calls. By default, Google counts one conversion per ad click for calls from ads. A call that carries a made-up value mixes guesses into your revenue: call tracking in a store account.
• All conversions. This column adds secondary actions and cross-device data, so its value is not your revenue: Conversions vs All conversions.
• View-through conversions. These come from people who saw an ad but did not click it. Google keeps them out of the main “Conversions” column. Whether a store should count them for Performance Max, YouTube and Demand Gen: view-through conversions.
• Old and silent actions. Accounts collect conversion actions over the years. Some stay primary after the tag behind them stops firing: cleaning up conversion actions.

How should the order value reach Google Ads?

As the amount of each order, sent with every purchase. About conversion values offers two options: the same value for each conversion, or a different value for each conversion. A store needs the second, which Google recommends when you sell multiple products that vary in price. It needs a dynamic value in the tracking code. Without one, Google uses the default value for all conversions.

Watch one setting, which the Google Ads API calls “always use default value”. When it is on, Google replaces the value and currency your tag sends with the default ones. Google also uses the default when the tag sends no value or an invalid one.

The tag can be correct while the report still shows one number for every order. What this does to ROAS and Target ROAS bidding: static conversion values.

Decide what the value includes and keep it the same everywhere: with or without VAT, with or without shipping, before or after discounts. Each choice shifts the ROAS your targets aim for: what conversion value should include. Send the currency with the value too, as Google’s own snippet does.

The last risk is one oversized row. A wholesale order, a test purchase or a checkout glitch can carry many times your usual order value. The portal treats a product’s conversion value for a day as anomalous when it is more than 30 times the account’s median order value. How to find such orders and stop bidding from learning on them: outlier orders.

How do you stop one order counting twice?

With a transaction ID inside one conversion action, and with one primary action per purchase across the account.

Use a transaction ID to minimize duplicate conversions explains the first part. A customer can reload the order confirmation page, and the tag fires again.

If two conversions of the same conversion action carry the same transaction ID, Google Ads counts only the first and drops the second as a duplicate. The ID must be unique, generated for every order, up to 64 characters long and free of anything that identifies the customer. The same ID sent for different orders makes Google undercount.

The second part needs its own fix, because de-duplication works only within one conversion action. A Google Ads tag, a GA4 import and a goal installed by your shop platform’s app are three separate actions. If two of them are primary, both count the same purchase.

Conversions created from GA4 arrive as secondary by default, so here the double count starts when someone switches them to primary next to the tag. Google notes that this option is not yet available to every GA4 property. How to find which tag, import or app counts twice: duplicate conversions.

Example store, not client data.

A tableware store spends 40,000 a month and gets 300 orders worth 180,000, a ROAS of 4.5. Performance Max takes 38,000 of the spend and brings 280 orders worth 168,000. A search campaign takes 2,000 and brings 20 orders worth 12,000.

Someone adds the GA4 purchase import as a second primary action to the goal Performance Max bids on, next to the Google Ads tag. The search campaign bids on its own goal, which leaves the import out.

Performance Max now reports 560 orders and 336,000, a ROAS of 8.8 instead of 4.4. The account shows 580 orders, 348,000 and a ROAS of 8.7.

Count the campaigns as equals and one of the two looks wrong. Count by money and 95% of the spend runs through doubled numbers.

How long does Google wait for a purchase?

By default, 30 days after a click. According to About conversion windows, you can set the click-through window from 1 to 90 days, depending on the conversion source. Google recommends at least 7 days.

The view-through window defaults to 1 day and the engaged-view window to 3 days. A change applies only to conversions recorded from that day on.

According to Understand your conversion tracking data, the main conversion columns report conversions by the time of the click, not the time of the conversion. Orders keep arriving for days after the click, up to the end of the window, so a report on the most recent days is still incomplete. Judge last week’s ROAS or yesterday’s change too early and you see a drop that later fills in. Which window a store should use and how long to wait: conversion window and conversion lag.

How do you recover orders that cookies and consent hide?

Google says plainly that cookies are not always available, for example because of browser settings. When it can’t observe a conversion, it adds modeled conversions to the “Conversions” column as estimates. You can narrow the gap from your side in two ways.

Enhanced conversions add hashed first-party data to the conversion: an email address, a phone number or a mailing address. According to the Google Ads API guide for online conversions, the tag sends the click ID, and Google recommends adding the transaction ID as well to help matching. You must accept the customer data terms first.

The setup through Tag Manager is in our step-by-step enhanced conversions guide. When diagnostics show low coverage or errors, the order confirmation page usually holds the cause: enhanced conversions diagnostics.

The second way is consent mode, which passes each visitor’s cookie choice to Google tags. Google’s About consent mode describes two versions:

• Basic. Tags stay blocked until the visitor answers the banner. If the visitor declines, nothing reaches Google, not even the consent status. Conversion modeling then uses a general model.
• Advanced. Tags load straight away and send cookieless pings while consent is denied. Google builds an advertiser-specific model, which it calls more detailed.

Modeled conversions are estimates, so compare Google’s totals with the orders in your store regularly to see how much of the report is modeled.

How do you correct returns and days of broken tracking?

Google lets you correct a conversion after recording it. According to About conversion adjustments, a retraction removes a conversion from the count and sets its value to zero. It affects both CPA and ROAS bid strategies.

A restatement changes only the value, for example after a partial return. It affects ROAS bid strategies. You can adjust online conversions only if your tag sent them with a transaction ID. When adjusting pays off for a store: retracting conversions for returns.

A broken tag teaches bidding that clicks stopped converting. The Google Ads API guide to data exclusions describes them as a way to tell Smart Bidding to ignore dates with conversion tracking issues. It also warns that using them often or for long periods can hurt performance. How to tell a broken tag from a real fall in sales within an hour: conversions dropped overnight.

Fixing tracking is itself a change bidding has to absorb. Google’s help names a 7–14 day learning phase for Smart Bidding and advises against frequent changes to conversion goals during it. How to replace a broken purchase action without resetting learning or misreading the next weeks: switching your purchase conversion action.

Tracking that works can still count the wrong thing: what we found in 14 accounts

The usual check asks one question: is the data arriving? In June 2026 we ran the portal’s conversion trust method over 14 GetProfit client accounts, on 90 days of data each. This is a check of our method on real accounts, not a market survey. All 14 were recording conversions, from 366 to 17,865 in 90 days.

• 8 of the 14 passed real order values for the whole 90 days.
• In 5, between 32% and 57% of conversions came through actions that send a fixed default value instead of the order amount. In one of them, 97% of conversions were leads rather than purchases.
• In 1, the 90-day total pointed to default values, yet none of its conversions in the last 14 days carried a default. The store had fixed its tracking inside the window. A verdict on the 90 days alone would have flagged tracking that already worked.
• Long lists of primary actions were the norm: 12 in one account with clean purchase tracking, 40 in another, 75 in a third. The count alone said little. The risk is primary actions from different categories, such as purchases, calls and page views, firing at once.

These cases share one trait: nothing looked broken. Every number built on conversions inherits the error, from the conversion rate to cost per order and ROAS.

The portal measures trust by money, not by campaign count

The portal’s Conversion check asks four questions of every campaign:

1. Event: do the ads learn on a purchase or a lead, rather than on a page view or an add to cart?
2. Value: does the real order amount arrive with the order, or one and the same number every time?
3. Duplicate goals: do two goals count the same order, or do the ads learn on several goals at once?
4. Amounts: are there orders many times larger than your average order value?

The answers multiply rather than add, so one weak check pulls the whole result down. Campaigns then roll up into one number for the account, weighted by spend.

The weighting changes the verdict. In one account shown in the portal, over a 90-day window, two search campaigns scored 100% trust. They carried 2% of the spend.

The other 98% ran through 10 product campaigns, where the real order amount arrived only in part. Weighted by money, the account’s trust came to 55%. These are one store’s numbers, not a market average.

Trust is a scale, not a switch. If the purchase is real and only the value is wrong, the portal shows revenue in grey with a “not trusted” mark. It works with the number of orders and cost per order instead. Where the value can’t be trusted, it hides ROAS altogether.

In the portal’s store score, the data area carries 17 of 100 points and also acts as a gate. If the purchase event itself is wrong, the parts of the score that depend on conversions earn nothing, while the feed and the assortment keep their points.

What to check before you trust the numbers

Run these checks in this order before you change bids or budgets, or launch Performance Max. The full walk-through, with what each finding means, is in our conversion tracking audit.

1. List the primary actions. In Google Ads, open Goals → Conversions → Summary. Note every action set as primary and the goal it belongs to. For a store, one purchase action should be primary.
2. Move the rest to secondary. Click “Edit goal” on the goal, then choose “Secondary (observe only)” under “Conversion action optimization”. This keeps the data in reports and out of bidding.
3. Check what each product campaign bids on. See whether it uses account-default goals or its own. Keep calls, directions and store visits out of the goals that hold your purchases.
4. Check the value. Compare conversion value per conversion for the purchase action with your real average order value, week by week. A value that never moves points to a default. In that case, open the action’s value settings.
5. Set the count to “Every” for the purchase action. Google applies a count change only to future conversions.
6. Test one order end to end. Make a test purchase and check in Tag Assistant that the event carries the value, the currency and a unique transaction ID.
7. Look for a second counter. A GA4 import, an old tag or a shop platform app set to primary next to your purchase tag will count the same orders again.
8. Compare a settled month with your store. Use “Conversions (by conv. time)”, which Google suggests for comparisons with your own systems. Google Ads counts only orders that followed an ad, so its number should stay below your store’s total for the same days. A higher number points to something counting twice or counting the wrong event.
9. Check the click ID path. Make sure auto-tagging is on, redirects pass the GCLID, and a cross-domain linker runs if checkout sits on another domain.
10. Plan for returns and outages. Decide whether to retract returned orders, and know where to add a data exclusion if tracking breaks.
11. Change one thing at a time. Wait out the learning phase before judging, and read results over full weeks.

The portal’s Conversion check runs its four checks (event, value, duplicate goals and amounts) on a 90-day window and weights the result by spend. It lists anomalous conversions with the date, campaign, product and how many times the amount exceeds your average order value. It leaves the attribution model, conversion windows and the count setting to you: check those by hand.

What your ads are really learning on. An order counted twice. An amount substituted for the real one. A payment several times larger than your usual basket. The portal checks what stands behind them. The portal changes nothing without your consent.

Check your conversions →

Sources

• How Google Ads tracks website conversions — click information stored in cookies and read by the tag; Google tag plus event snippet; Conversion Linker on all pages in Tag Manager; consolidate gtag.js and Tag Manager; auto-tagging, GCLID through redirects, cross-domain linker, no iFrames; the GA4 tag does not keep the click ID for Google Ads tags; modeled conversions in the “Conversions” column. Checked 2 October 2026.
• About conversion measurement — only GA4 key events can be imported; 24–48 hours for import options; the Google tag for Ads-specific features such as enhanced conversions; cross-device data in “All conversions”; view-through conversions excluded from the “Conversions” column; 7–14 day learning phase. Checked 2 October 2026.
• Understand your conversion tracking data — the main conversion columns are reported by the time of the click; “by conv. time” columns for comparisons with your own systems. Checked 2 October 2026.
• About primary and secondary conversion actions — primary actions in “Conversions” and used for bidding; secondary in “All conversions”, except in custom goals; GA4-created conversions secondary by default; path to change the setting; misconfiguration hurts Smart Bidding; 7–14 day learning phase. Checked 2 October 2026.
• About conversion goals — one account-default goal per funnel; “Purchase” for an online retailer; account-default and campaign goals. Checked 2 October 2026.
• About conversion values — the same value or a different value for each conversion; different values for products that vary in price; without a dynamic value the default applies to all conversions. Checked 2 October 2026.
• About conversion counting options — “Every” for sales and as the website default; “One” for leads and for calls from ads; changes apply to future conversions; path Goals → Conversions → Summary. Checked 2 October 2026.
• Use a transaction ID to minimize duplicate conversions — duplicates of the same conversion action with the same ID are not counted; unique, dynamic IDs up to 64 characters without personal data; the same ID for different orders causes undercounting; verification with a test purchase and Tag Assistant. Checked 2 October 2026.
• About conversion windows — 30-day default click-through window, 1–90 days by source, at least 7 days recommended; view-through 1 day and engaged-view 3 days by default; changes apply going forward. Checked 2 October 2026.
• About consent mode — basic and advanced consent mode, data sent in each, general and advertiser-specific modeling. Checked 2 October 2026.
• About conversion adjustments — retraction and restatement and which bid strategies each affects; online adjustments require a transaction ID. Checked 2 October 2026.
• Google Ads API: Conversion management — a conversion is an action a user takes after clicking or viewing an ad; website conversions are measured with the Google tag. Checked 2 October 2026.
• Google Ads API: Conversion action fields — “always use default value” replaces the value and currency sent in conversion events; the default value applies to missing or invalid values; the primary flag. Checked 2 October 2026.
• Google Ads API: Manage online click conversions — enhanced conversions for web: hashed email, phone and address; the tag sends the click ID and order ID; transaction IDs recommended; customer data terms required. Checked 2 October 2026.
• Google Ads API: Create data exclusions — data exclusions for dates with conversion tracking issues; frequent or long use can hurt Smart Bidding. Checked 2 October 2026.
• GetProfit portal methodology, a check of the conversion trust method on 14 client accounts, 90 days each, June 2026 — real and default values, the share of default-value conversions, the number of primary actions.
• GetProfit portal, Conversion check — the four checks, spend weighting, the 55% example from one account over 90 days, the anomaly threshold of 30 times the median order value, no ROAS when the value can’t be trusted, the data area’s 17 of 100 points in the store score.