GA4's Source Group dimension cleans up messy traffic sources — here's what it does, and the model that made it necessary
What is GA4's Source Group? It's a new dimension Google added on 11 June 2026 that automatically collapses the many raw source strings a single platform produces — facebook, fb, m.facebook.com, l.facebook.com, Meta-facebook — into one clean label: Facebook. It does the same for Instagram, TikTok, Pinterest, Amazon, YouTube, Google Search and Maps, and it names ChatGPT (OpenAI) and Perplexity explicitly. Best of all, it's retroactive — it re-labels your historical data, not just new sessions — and there's nothing to configure. If you've ever stared at a Traffic Acquisition report split five ways across the same platform, this is the fix.
That's the short answer, and it's genuinely good news. The longer answer is the interesting one: Source Group exists because GA4 spent years not doing this, and understanding why is the difference between treating clean traffic sources as a feature you switch on and treating them as something a tool either gets right by design or doesn't. This post covers what Source Group actually does, how to use it, where it stops, and the data model underneath that made a dedicated "un-fragment my sources" feature necessary in 2026.
What Source Group actually does
When someone arrives at your site from Facebook, the referrer string GA4 records depends on how they arrived: the mobile app sends m.facebook.com, a link-shim sends l.facebook.com, some integrations send fb or Meta-facebook, and the desktop site sends facebook.com. GA4 stores each of those verbatim as the Source. For years, your Traffic Acquisition report then showed Facebook as four or five separate rows, none of which told you the real total.
Source Group sits one layer above Source. It maps all the known variants of a platform to a single brand name, so "Facebook" becomes one row that adds up correctly. Google seeded it with the platforms that fragment the most — the major social networks, Amazon, YouTube, Google's own properties — and, notably, the AI assistants that started sending meaningful referral traffic in the last year.
It's worth being precise about the two related fields Google shipped, because they do different jobs:
| Dimension | What it answers | Example value |
|---|---|---|
| Source | The exact referrer string, verbatim | m.facebook.com |
| Source Group | Which brand/platform is that? | Facebook |
| Source Platform | Which ad ecosystem handled it? | Meta Ads |
Source Group is the brand label. Source Platform is a separate, coarser field naming the advertising ecosystem a step above it (for example, "Meta Ads" or "Google Ads"), which helps you tell paid ad-platform traffic from organic without hand-writing regex. In the same June update Google also added admin-level hostname filtering, so you can exclude hits claiming to come from domains you don't own.
How to use it
There's almost nothing to do, which is the point:
- Pick it as a dimension. Wherever GA4 lets you choose Source, you can now choose Source Group instead. Google is rolling it out across GA4's reporting surfaces, so exactly where it shows up first may depend on when your property gets it — but the dimension itself needs no enabling.
- No setup, no tag changes. It's populated automatically. There are no rules to write and nothing to add to your site.
- No backfill wait. Because it's retroactive, historical sessions are already grouped — you don't need to reprocess anything or wait for new data to accumulate.
That's a real improvement over the old workaround, which was to build a custom channel group with regex rules — ^(facebook|fb|m\.facebook|l\.facebook) and so on — for every platform you cared about, maintain it as platforms invented new source strings, and accept that it only applied going forward, never to history. Source Group replaces most of that busywork. Credit where it's due.
Why GA4 needed this in the first place
Here's the part the how-to guides skip. Source Group is a patch for a problem GA4's own architecture creates.
GA4 records the raw referrer string per session and defers all normalization to report time. Nothing decides "these five strings are all Facebook" at the moment the hit is collected — the strings are stored as-is, and it's on you (or, now, on Source Group) to reconcile them later when you read the report. That design is flexible, but it means fragmentation is the default state of your data. Every new platform, every new mobile-app referrer format, every link-wrapper starts life as its own uncategorized row until something upstream teaches the report to merge it.
A first-party analytics tool can make the opposite choice: normalize referrers into a small, fixed set of channels at aggregation time, before anything is stored as a summary. That's how Simplytics works. Its referrer classifier folds m.facebook.com, l.facebook.com, lm.facebook.com and any *.facebook.com subdomain into Social; old.reddit.com, out.reddit.com and the rest into Social; every google.* country domain into Search; the AI assistants — including Perplexity, which GA4's own native AI channel still files under Referral — into AI. There are six channels, they're fixed, and there is no regex to write and no custom group to maintain, because the fragmentation never happens: variants collapse the moment the numbers are rolled up.
In other words, the reason Google shipped Source Group in 2026 is the same reason a cookie-less first-party tool never needed it. Fragmented sources aren't a law of nature; they're a consequence of storing raw strings and normalizing at read time. Change that one decision and the problem disappears — along with the feature built to solve it.
Where the honesty line is
It would be easy to overclaim here, so let me be exact about what Source Group does that Simplytics does not.
- GA4 is more granular at the platform level. Source Group gives you Facebook, Instagram, and TikTok as three distinct labels. Simplytics collapses all of them into one Social channel. If you need per-platform social breakdowns inside a single view, GA4's Source Group is genuinely the better tool for that job, and it isn't close — a fixed six-channel model is a deliberate simplification, not a secret superpower.
- GA4 splits paid from organic; Simplytics doesn't attribute ad spend. That's what Source Platform is for. Simplytics is not a paid-attribution tool and makes no attempt to be one. If measuring ad conversions is your reason for using analytics, that's GA4's territory.
- The clean-up is at the channel layer, not the raw list. Simplytics still stores and shows individual referrer hostnames, so at the raw-referrer level you'll still see
m.facebook.comandfacebook.comas separate rows — the automatic normalization is specifically the Channels view, not a magic dedupe of every source. Source Group, by contrast, cleans up the source list itself.
None of that changes the structural point. Source Group is a good fix, GA4 is more capable on platform granularity and paid attribution, and it still took a dedicated 2026 feature to give you a clean "which platforms send me traffic" answer that a first-party channel model produces for free. Both things are true.
The bottom line
If you run GA4, turn to Source Group — it's retroactive, it's automatic, and it retires a pile of custom-channel-group regex you never enjoyed maintaining. It's one of the better GA4 changes this year.
But if you're choosing a tool and "I just want to see which platforms send me traffic, cleanly, without configuring anything" is high on your list, notice what the story is really telling you: GA4 needed a special dimension in mid-2026 to un-fragment data its own model fragments by default. A tool that classifies referrers into fixed channels at aggregation time answers that question on day one, at $1/month — versus the $9–15/month other privacy-focused tools charge — a lot of functionality for a lot less money, and one less thing that needed a patch.
For the neighboring GA4 stories this ties into — how ChatGPT, Perplexity and Gemini show up as a referral channel, and why GA4's hostname and report filters only act after data is already ingested — the pattern is the same one Source Group fits: GA4 collects everything raw and sorts it out later, and "later" is where the work lives.