Fixing broken e-commerce attribution: the order to work in.
- “what order to fix GA4 attribution problems in”
- “how to prove an attribution fix worked”
- “GA4 attribution wrong where do I start”
- “sales credited to the wrong channel GA4 what to fix first”
- “ecommerce attribution fix checklist”
- “GA4 direct, referral and unassigned too high what to fix first”
- “fix broken e-commerce attribution step by step”
Baris Asa, tracking consultant, London · Last checked: 9 November 2026
If GA4 files paid sales under Direct, a payment provider, your own checkout domain, Organic Social or Unassigned, work in this order: first the redirects that strip campaign tags and click IDs, then visits split across your own domains or the payment return, then channel naming, declined cookies and server-side clicks for Meta. Change the attribution model last. A lost label usually can’t be repaired from the reports afterwards. Later fixes are hard to read until the earlier ones hold. Each step below has a test and a “done when”.Open Reports, Acquisition, Traffic acquisition now, with Session default channel group as the dimension. If Direct, Referral or Unassigned looks too big for paid traffic you are running, the table below says where to start. The attribution page lists fixes by symptom and who does them. This note answers one question: in what order, and how you know each one worked.
| What you see | Start at | What it usually means | Fault |
|---|---|---|---|
| Paid clicks under Direct or Organic, Direct up on a date | Step 1 | Campaign tags lost on a redirect | F-204 |
| Your own checkout domain among referrals | Step 2 | One visit split in two | F-204 |
| PayPal, Klarna or Stripe among referrals | Step 2 | The return from payment read as a source | F-205 |
Facebook ads in Organic Social, campaigns in Unassigned, Meta and meta as two rows | Step 3 | Tags GA4 doesn’t read as paid | none, naming |
| Paid sales fall when cookies are declined | Step 4 | Click ID not carried | F-201 |
| Meta sales with no ad attached | Step 5 | Server events without Meta’s click ID | F-401 |
If the channels look right but the totals don’t match the store, that is a different job: see why GA4 revenue differs from Shopify or the three-total comparison.
Before you start
Take four closed weeks of Traffic acquisition by channel, with your store’s web orders beside them, and write down the date of any redesign, new checkout, cookie banner or domain move. The store’s count is the ceiling: GA4 should never credit more sales than the store took.
Then leave the attribution settings alone while you work. Traffic acquisition and User acquisition use a fixed model, last click across paid and organic channels, so they stay comparable week to week (Google’s scopes of traffic-source dimensions).
The test visit you will reuse
Add this to your landing page address: ?utm_source=attribution-test&utm_medium=test&utm_campaign=route-check&gclid=test123. The distinctive names let you find the visit afterwards and ignore it in reports. Every step below uses it, and for each you check the address bar and, if you want proof, the page_view event’s page location in GA4 DebugView.
Step 1: do the campaign tags and click IDs reach the landing page?
If a UTM or a gclid (the ID Google Ads adds to each click) is stripped on the way, once the tag is gone from the landing address, GA4 has nothing left to read. GA4 then sees a visit with no source and files it under Direct or Organic. It is filed as F-204 .
Google’s advice is to make sure the gclid is passed to the final landing page when your site uses redirects (about auto-tagging). Domain moves, country redirects, link shorteners and “clean URL” rules are the usual culprits.
Test: open the test visit through every route your ads use: the old domain, the shortener, the country page. After the page loads, all four parameters should still be in the address bar.
Done when: every route keeps the full query string, and the weekly Direct line stops stepping up on the date you wrote down. A direct jump after a redesign is covered in this note.
Ask your developer: keep the whole query string through every redirect.
Step 2: does one visit stay one visit?
Two cases split a visit in two. Each starts a new session with a new source, usually Referral.
- Your own second domain, such as a checkout or booking step. Without cross-domain measurement each root domain gets its own cookies, so one buyer counts as two users. With it, the IDs travel in the
_glparameter, a short code added to the link. Google lists what breaks it: a redirect that drops_gl, navigation by script instead of a click, a script that stops the click event, and a domain missing from the list. Every page on every domain needs the same G- tag from one web data stream (cross-domain measurement). Full diagnosis: cross-domain tracking not working. - A payment provider you don’t own, such as PayPal or Klarna. You can’t tag their page, so cross-domain doesn’t apply. List the hostname as an unwanted referral: Admin, Data collection and modification, Data streams, your web stream, Configure tag settings, Show all, List unwanted referrals. It needs the Editor role and allows 50 per stream. Sessions already recorded keep the provider as their source (unwanted referrals). Walk-through: payment gateway showing as referral. This is F-205 .
Test: place a test order from a test visit tagged with the UTMs only, without the gclid (with a linked Google Ads account, auto-tagging can make a fake gclid change the source). For a payment provider, use its sandbox or a small real order, then refund it. Read it by transaction ID in an Exploration. Its source should be your test campaign, not your checkout domain or the gateway.
Done when: the test order keeps its campaign, and no domain you own or pay through is left among your referrals. Whether the Google Ads and Meta click IDs make the same trip is a separate check: do gclid and fbclid survive your checkout?
Step 3: does GA4 read your channel names as paid?
Naming comes third because it only relabels what arrives. It is worth doing once Steps 1 and 2 hold, since a mislabelled row and a lost tag look alike until then.
GA4 sorts sessions by fixed rules on source and medium. Paid Social needs a social source and a medium that contains cp (such as cpc or cpm), or is ppc or retargeting, or starts with paid, so paid_social passes. A social source with a medium that doesn’t match falls through to Organic Social, and a paid medium with a non-social source goes to a different paid channel. A session that matches no rule goes to Unassigned (default channel group).
Two more checks:
- Case. Channel rules ignore case, but the reports don’t. Google says parameter values are case sensitive, so
Metaandmetaare two rows (campaign data with custom URLs). - Missing tags. Without
utm_source,utm_mediumandutm_campaign, GA4 reports(not set).
One exception: with Google Ads auto-tagging on, Google’s help says GA4 gives auto-tagging priority over manual tags, with no override. Hand-built UTMs on Google ads don’t decide their channel.
Test: list every source and medium pair under your paid campaigns. Any paid pair sitting in Organic Social, Unassigned or Direct means the convention is wrong, not the tracking.
Done when: a new, correctly tagged campaign shows in the right channel once the data has processed. Check DebugView or Realtime first, then Traffic acquisition. Hand your marketers one written convention: lower case, one source per platform, one medium per channel.
Step 4: what happens when the visitor declines cookies?
Only now is a consent gap worth sizing, because it is what remains once the cheaper losses are gone.
With ad_storage denied (the setting for advertising cookies), the click ID isn’t stored in a cookie. URL passthrough can carry it between pages instead. Google lists the conditions: the tag is consent-aware, URL passthrough is switched on, consent mode is implemented, a GCLID or DCLID is in the URL, and the link stays on the same domain (consent mode guide). It doesn’t cross to another domain, so it doesn’t replace Step 2. Google’s page also notes the URL can gain other parameters (such as gclsrc, _gl and wbraid), so the Step 1 redirects must keep them.
With consent mode set up, GA4 can also model some conversions the banner hides. The volume threshold is on file at F-201 .
Test: with the banner declined, open the test visit, click to a second page, and check the click ID is still in the address bar.
Done when: the declined-cookie journey keeps the click ID within your domain, and DebugView shows the consent state on the test visit’s events. How that is wired: what basic and advanced Consent Mode change.
Step 5: does the click reach Meta from the server?
If Meta shows sales with no ad attached, check whether server-side purchases carry fbc, the value that holds Meta’s click ID. Meta recommends sending it with every Conversions API event, built as fb.<subdomainIndex>.<creation time in milliseconds>.<fbclid> from the fbclid on the landing URL when there is no _fbc cookie (fbp and fbc parameters). Don’t change the casing of the fbclid: Meta says it is case sensitive.
Test: place a test purchase and open it in Events Manager. Check fbc is listed and its last part matches the fbclid on your landing URL.
Done when: every server-sent purchase in a test batch carries fbc. What a good setup hands back: this note. When the browser alone can’t carry the click, the case for a server path is on file as F-401 , server-side collection that the browser can’t replace. Where else the click ID drops out between landing and order: do gclid and fbclid survive your checkout?
Step 6: only now, models, windows and tools
Everything above repairs the inputs. The reporting attribution model only changes how credit is shared between them. The reporting model defaults to data-driven (Google’s scopes page), and first-click, linear, time-decay and position-based models were removed in November 2023. Google says a model change applies to historical and future data. Lookback windows also differ by event: acquisition events default to 30 days (or 7), other key events to 90 (or 30, or 60) (attribution settings).
An attribution platform, a post-purchase survey or an incrementality test belongs here, and it is your marketers’ call. They answer how much credit a channel deserves. They can’t recover a UTM that never arrived. Close the inputs first. Then argue about credit.
Where does it need a specialist?
Steps 1 and 3 are often a developer and a spreadsheet. Steps 2, 4 and 5 are measurement work: domains, consent wiring, server-side events. If your Step 2 test order still shows the wrong source, send me the test address and a screenshot of your referral list at [email protected] and I’ll tell you which step it looks like it fails at. Details of the work are on the attribution page. I work with UK online shops and the agencies behind them.
Common questions
In what order should I fix broken e-commerce attribution?
Fix the losses that remove the campaign label before the ones that only relabel it. First the redirects that strip UTMs and click IDs, then the visits split across your own domains or the payment return, then channel naming, then declined cookies, then server-side clicks for Meta. Change the attribution model last. Each step has a test, so you know it worked before the next.
How do I know an attribution fix worked?
Use one tagged test visit and follow it to a test order. After the fix, the order’s session source in GA4 should be your test campaign, not Direct, your checkout domain or a payment provider. Date every change, because fixes work from the day they go live and old sessions keep their old source.
Should I change the GA4 attribution model to fix this?
Not first. The reporting attribution model only changes how credit is shared in reports that use event-scoped dimensions. Traffic acquisition and User acquisition use a fixed paid-and-organic last-click model whatever you choose. A model change relabels credit; it doesn’t bring back a UTM, a click ID or a session that was lost.
Will fixing attribution correct my old data?
Mostly no. Redirect, tag and domain fixes apply to new visits, and sessions already recorded keep their old source. The exception is the reporting attribution model, which Google says applies to historical and future data. Write down the date of every change so a before-and-after comparison is fair.
Do I need a consultant to fix e-commerce attribution?
Often not for naming. Redirects need a developer, and cross-domain set-up, consent wiring and server-side click data are measurement work, which is where to bring someone in. Synctag (Baris Asa, London) does that part. Campaigns and the model stay with your marketers.
Sources
Read on 9 November 2026:
- Google Analytics Help, default channel group (Paid Social, Organic Social and Unassigned rules; channel definitions are not case sensitive)
- Google Analytics Help, collect campaign data with custom URLs (parameter values are case sensitive;
utm_source,utm_medium,utm_campaign; missing parameters give(not set)) - Google Analytics Help, set up cross-domain measurement (
_gl; same G- tag; redirects, script navigation and stopped click events) - Google Analytics Help, identify unwanted referrals (Admin path; 50 per stream; Editor; sessions recorded before the exclusion keep the domain)
- Google Analytics Help, scopes of traffic-source dimensions (session and user scoped dimensions use paid and organic last click; event scoped follow the selected model)
- Google Analytics Help, attribution settings (reporting model applies to historical and future data; lookback defaults; older models removed November 2023)
- Google Ads Help, about auto-tagging (GCLID passed through redirects; auto-tagging prioritised over manual tagging in GA4)
- Google, consent mode guide (URL passthrough conditions, same domain only; parameters that may be added)
- Meta for Developers, fbp and fbc parameters (
fbcformat; recommended on every Conversions API event;fbclidcase sensitive)