Splitting a Shared GA4 Measurement ID
Two of our sites were firing the same Google Analytics measurement ID.
One is a blog. The other sells managed IT. Same tag on both, which meant every report blended two audiences with nothing in common — someone reading a technical article and someone pricing a support contract counted as the same kind of visit.
It was the highest-priority open item on our own marketing plan, and the fix took about twenty minutes once we understood one option we had to refuse.
What a shared measurement ID actually costs
The reports still work. That is the difficulty — nothing looks broken, the numbers are plausible, and the conclusions drawn from them are wrong in a specific direction.
Conversion rate becomes meaningless. Blog traffic converts at a fraction of commercial traffic, and it is usually the larger number. Blend them and your commercial conversion rate looks terrible while your blog looks like it is failing to sell, which is not its job.
Bounce and engagement figures describe neither site. Someone who reads one article thoroughly and leaves is a success on a blog and a failure on a service page. Averaged together they produce a number that means nothing about either.
Channel attribution mixes two funnels. Organic search to the blog and organic search to a service page are entirely different intents, arriving in one row.
You cannot answer the only question that matters: does the blog actually send people to the commercial site, and do those people buy? That is a cross-property question, and you need two identities before you can ask it.
The split
Both sites sit under the same brand, so the right structure was one property with two data streams rather than two properties. That keeps cross-site reporting possible while separating the numbers.
Three steps:
- Create a second web stream under the existing property, for the blog. It gets its own measurement ID.
- Point the blog at the new ID. One value in the site’s configuration.
- Leave the commercial site on the original ID, untouched. Its history stays continuous, which matters — the site with the longer commercial record should not be the one that gets a new identity.
The option to decline, and why
When you create a new stream, GA4 offers to use an auto-detected existing tag it has found on your domain.
Decline it.
Accepting means the new stream adopts the configuration of the tag it detected — which is the old shared tag. You end up with a new measurement ID that behaves like the old one, having done the work of separating and none of the separating. It is the most plausible-looking wrong button in the flow, because it appears helpful and is phrased as a convenience.
★ Insight ─────────────────────────────────────
This is worth generalising beyond analytics. The dangerous option in a setup flow is rarely the one labelled “advanced”. It is the one that offers to carry over your existing configuration, because it looks like it saves work and it silently reproduces the state you were trying to leave. Any time a tool detects your current setup and offers to apply it, ask what you would lose by saying no — the answer is usually five minutes.
─────────────────────────────────────────────────
The related mistake: one property, many brands
The inverse arrangement causes the same damage and is more common in groups.
Several distinct brands, each with its own site, all reporting into one property because it was set up once and extended. Reports then average across businesses with different customers, different price points and different conversion definitions, and every brand’s numbers are diluted by the others.
The tell is a conversion rate that has been stubbornly flat for years while individual brands have visibly changed. An average of several divergent series moves very little, and the flatness is read as a plateau rather than as an artefact.
Do you have this problem?
Three checks, in order of speed.
Count the sites, count the measurement IDs. If you have more than one domain and one ID, you have this. Businesses running a main site plus a blog on a subdomain are the most common case, because the blog was set up by copying the main site’s tag.
Look at your landing page report. If pages from two different domains appear in one list, they are on the same stream.
Check for a second tag manager container doing the same job. Occasionally both a hardcoded tag and a container tag fire, which produces double-counted pageviews — a different problem with the same cause, and the symptom is a sudden historical doubling that nobody can explain.
What to do about the history
The honest answer: you cannot retroactively separate it.
Data collected under one ID stays blended. Going forward you have clean numbers; going backward you have a blend, and any year-on-year comparison spans the change.
Two things help. Annotate the date of the change in GA4, so whoever reads a report in eight months knows why the shape changed. And accept that the first month after a split looks like a collapse on one property and a jump on the other — that is the separation working, not traffic moving.
Two streams or two properties?
Worth getting right first time, because moving later means another discontinuity.
Two streams in one property is correct when the sites are one business and you want combined reporting available. A user who reads the blog and then visits the commercial site can be followed across both, which is the question most content-led businesses actually want answered.
Two separate properties is correct when the sites are genuinely different businesses, when different people need access to each without seeing the other, or when a site may be sold or spun out.
The common mistake is choosing separate properties for separation’s sake, then discovering that the cross-site journey — the one that justifies the blog’s existence — has become unanswerable, because the two properties cannot see each other.
We chose one property with two streams. Both sites are the same brand, and the question we most wanted back was whether the blog feeds the commercial site.
The broader point about measurement
The reason this sat unfixed for so long is that nothing was broken. Reports generated, dashboards populated, numbers moved. Everything about the system said it was working.
What it could not do was distinguish two things that needed distinguishing, and no report will ever tell you that — because a report can only show you the categories it has, never the category it is missing.
That is the same shape as a total that does not know what it excludes, or a check that reads a file that does not exist. The system is not wrong. It is answering a narrower question than the one being asked of it, and the gap is invisible from the output.
The practical habit: once a quarter, take one number your business acts on and ask what it is combining. Usually the answer is nothing. Occasionally it is two audiences, two products or two regions, and finding one of those is worth the quarter’s checking.
Cloud Geeks sets up analytics, tracking and cloud infrastructure for Australian small businesses. Websites come from Cosmos Web Tech, apps from Awesome Apps, and the group is Ganda Tech Services.