Imagine opening your analytics in January and seeing revenue down by two thirds, while your store records show a much smaller dip. That happened to one UK retailer in a recent audit. Between Christmas Day 2025 and January 4, 2026, Google Analytics 4 told the retailer its revenue had fallen 68% per day compared with the year before. The store’s own order records showed a fall of 13%. The gap was not a slow season. It was broken tracking, and it had been quietly getting worse for a year.
The audit, published by Search Engine Journal on October 28, 2026, compared GA4 with store orders at eight UK ecommerce brands. Six brands could be matched order by order. At three of them, GA4 told a very different story from the till. At one retailer, GA4 recorded 81% of web orders in the fourth quarter of 2024, then 41% in the fourth quarter of 2025, then only 32% over the Christmas period. If you make budget decisions on numbers like that, you are not reacting to demand. You are reacting to a measurement fault.
What the Audit Actually Compared
The auditor lined up GA4 purchases against web orders, week by week where possible, for the fourth quarters of 2024 and 2025. The useful number here is the capture rate: GA4 orders as a share of store orders. A healthy setup will not reach 100%, because some shoppers decline tracking and some use ad blockers, so a steady gap is normal. What should worry you is a capture rate that steps down sharply, drops close to zero for a stretch, or jumps above 100%. Those shapes point to a specific fault.
Five Ways GA4 Goes Wrong
First, a steady gap. One brand recorded between 72% and 84% of web orders every week of the fourth quarter of 2025, and another held steady at 76% in both years. That is the shape of cookie consent doing its job. A similar share of shoppers declines tracking each week, so GA4 undercounts by a similar share each week. This gap is not a bug to fix. It is a limit to respect. Use GA4 for trends and for comparing channels, not as your sales total.
Second, a cookie banner that switches itself off. Two retailers moved to a new consent setup in 2025, but their consent plans also covered a fixed number of page views per month. Once the allowance ran out, the banner stopped showing until the next billing date, and with it, much of the tracking stopped too. Retailer A saw sessions drop out in blocks that always ended on the 14th of the month. Retailer B saw blocks ending on the 20th. In the Black Friday billing cycle, Retailer A’s allowance ran out on December 1 and GA4 did not recover until December 15. That is the heart of peak season, lost from the reports. GA4 kept receiving some data, so nothing looked broken at a glance: Retailer B recorded up to 68% of orders in weeks when the banner stayed up, and only 10% to 22% when it switched off.
Third, a blackout followed by a late backfill. Retailer C recorded 2,215 store orders between October 28 and December 4, 2025, and GA4 recorded no purchases at all in that window. Then, on December 9, GA4 recorded 1,261 orders in a single day, including 1,000 order numbers from inside the gap. Almost all of them landed in the Unassigned channel. So the quarter looked roughly plausible in total, while the weekly story, and every channel report built on it, was wrong.
Fourth, the same order counted twice. Google’s guidance says GA4 deduplicates purchase events with the same transaction ID, but only from the same user. At Retailer C, 657 of the 3,365 order numbers GA4 recorded in the fourth quarter of 2025 had more than one purchase event, adding 735 purchases to the total. Across the quarter, the blackout and the double counting almost cancelled each other out: GA4 showed 4,100 purchases against 4,501 store orders. Check only the quarterly total and the setup looks fine. Check order by order and it is not.
Fifth, visitors who were never shoppers. Retailer D recorded 23,187 sessions on September 27, 2026 and 25,622 the next day, against a normal range of 300 to 900 a day in mid September. Almost all were direct, desktop Chrome visits averaging about a second of engagement, while Search Console clicks did not move. GA4 filters known bots and spiders, but other bots that run a site’s code like a browser still count as visitors. Left in the reports, they inflate sessions, crush conversion rates, and make real traffic look weaker than it is.
Use Search Console as an Early Warning
One of the most practical ideas in the audit is also the cheapest. Search Console counts a click whenever a result sends someone to a page outside Google, whatever your site’s tags are doing. Divide GA4’s Google organic sessions by Search Console clicks each week, and a tracking failure shows up as a sudden drop in that ratio.
During Retailer C’s blackout, Google sent 4,244 clicks in the week of November 3, while GA4 recorded only 57 Google organic sessions. You do not need a new tool to spot that. You need two numbers side by side, checked weekly. The same habit catches the other faults: session blocks lining up with a billing date, days with sessions but no purchases, a sudden run of purchases in Unassigned, or session spikes that orders do not follow.
What Bad Data Does to Budget Decisions
This is where a tracking fault stops being a reporting annoyance and starts costing money. Retailer A’s web orders actually rose from 5,874 to 7,686 year on year, while GA4 showed them falling by a third. GA4 also showed Google Ads return on ad spend falling from 8.07 to 4.79, and organic search revenue nearly halving. Matched to store orders in a separate attribution tool, Google Ads return rose from 4.42 to 7.93 on 20% less spend, and revenue attributed to organic and direct visits rose 58%. The attribution models differ, so the exact figures are not directly comparable, but they should not point in opposite directions.
Retailer C tells the same story in miniature. GA4 put revenue from Google, paid and organic, down 86%, while the store’s net sales rose 4.8%. Cut either channel on GA4’s numbers alone and you would have pulled budget from channels that were working. Agree the source of truth before the holidays: orders and revenue come from the store, GA4 explains behaviour and trends, and ad platforms report what they claim. Lead with the capture rate in plain language, for example that GA4 recorded 41% of web orders this quarter, down from 81% last year.
A Pre-Holiday Check You Can Run This Week
First, export weekly web orders for the fourth quarter of 2024 and the fourth quarter of 2025, leaving out marketplace, subscription, and manual orders so the comparison stays clean. Second, pull GA4 purchases for the same weeks as distinct transaction IDs on purchase events. Avoid the transactions metric, which counts refunds as well as purchases. Third, divide GA4 purchases by store orders for each week. A step change, a near zero week, or a week above 100% is a tracking problem, not a trading result.
While you have the data open, list the purchase event count for each transaction ID and flag anything above one as double counting, check for session dropouts lining up with a monthly billing date, and confirm your consent plan is sized for peak traffic with limit warnings reaching marketing as well as developers. If you find a fault, date every change and keep store and GA4 revenue side by side from that point on.
Move Purchase Tracking to the Server
Most GA4 setups still record a purchase in the shopper’s browser, where a checkout update, a script error, an ad blocker, or a page refresh can lose the event or count it twice. The audit’s longer term recommendation is to send the purchase from the store’s own system when the order is created, through the Measurement Protocol or a server side tagging container fed by the store, keyed on the order number. That way the event no longer depends on the confirmation page, a refresh cannot duplicate it, and every event matches a real store order.
Two cautions apply. Moving browser tags into a server container alone does not achieve this, because the purchase still starts in the browser. Pass the shopper’s GA4 client ID and session ID with the server event, captured at checkout, so GA4 can join the order to the visit. Server side tracking also does not change consent obligations, so the steady consent gap stays. If you are starting now, send server events to a separate GA4 property through the fourth quarter rather than switching over in November.
None of this makes GA4 a sales ledger. Before you cut or push spend this holiday season, count your orders where they happen: in the store. Then use GA4 for what it does well, explaining how shoppers behaved on the way to those orders. That division, checked weekly and agreed in advance, stops a tracking fault from becoming a budget mistake.

FAQ
What is the GA4 capture rate?
It is GA4 orders as a share of store orders for the same period. It tells you how much of your real trading GA4 is seeing. A steady rate below 100% is normal because of consent and ad blockers, while a sudden step change, a near zero week, or a week above 100% points to a fault.
Why would GA4 show revenue falling while store sales rise?
Because GA4 only records what its tags capture. If a consent banner stops showing, purchase tracking breaks, or a backfill lands late, GA4 can miss whole weeks of orders while the store keeps selling. The gap looks like falling demand in GA4 and steady or rising demand in the store.
How do I check GA4 against my store before the holidays?
Export weekly web orders for the fourth quarters of 2024 and 2025, pull GA4 purchases for the same weeks as distinct transaction IDs, and divide GA4 purchases by store orders each week. Investigate any step change, near zero week, or week above 100%.
Should I move GA4 purchase tracking server side?
For most ecommerce stores, yes, over time. Sending the purchase from the store system when the order is created, keyed on the order number, avoids losses and duplicates from browser refreshes and script errors. Start in a separate GA4 property through peak season rather than switching over in November, and keep the browser tag for behaviour.