17 November 2026 · 10 min read

Who still has access to your GA4 property?

Ask who can open a client's Google Analytics property and you usually get two or three names. Open Admin, then Account access management, and the list is longer. The difference between the two answers is where access problems live: the contractor from two years ago, the agency that was replaced, the intern who was added to a group, and the Google Ads account that quietly carried five people in with it.

Access is one of the few settings in analytics that gets worse by doing nothing. Every project adds a person and almost nobody removes one, and nothing breaks, so nobody looks. Then a client asks for a copy of their data, an auditor asks who could see it, or a former supplier still holds publish rights on a container.

Effective access is not the list you see

Google splits every member's permissions into two kinds. Direct permissions are the ones assigned to that person explicitly on that resource. Effective permissions are everything they hold through inheritance: a role on the account above the property, a user group, an organisation. The permission that applies is the most permissive one anywhere in the chain.

Google states the consequence plainly: a user with the Editor role at the account level has Editor on every property in that account, and you cannot revoke it at the property level. You can add permissions as you move down the hierarchy, but you cannot take away what a higher level granted.

Two things follow for anyone auditing. Reading the property-level list alone under-reports who can see the data. Removing a person from a property does nothing at all while their account-level role still grants the same access.

The five roles and what they can do

The names changed with Google Analytics 4. The current set:

For an audit, the deciding fact sits in the Editor row. Editor can view users but cannot change them, so a client or agency holding only Editor can read the list and do nothing about it. If nobody in your organisation holds Administrator, every offboarding request routes through whoever does.

The Google Ads door

Linking Google Ads to an Analytics property grants roles automatically, and this is the access path people forget because nobody typed an email address to create it. Google maps Ads access levels onto linked users: an Ads account administrator becomes Editor, a standard user becomes Marketer, read only, billing and email only all become Viewer. Those are Google's recommended assignments, and an Analytics Administrator can change them.

The shape matters more than the table. Each linked user stands for a group of people in the Ads account, and changing that linked user's role changes it for everyone in the group. The assignments persist until the link itself is deleted.

So when a paid media agency leaves, deleting their named users from Analytics is not enough. While their Google Ads account stays linked, their staff keep arriving through the link with the role the link assigns. The link lives under Admin, then Product links, then Google Ads links.

Two data restrictions worth understanding

Alongside the five roles sit two data restrictions: No Cost Metrics and No Revenue Metrics. They hide cost and revenue metrics, including any custom metric identified as a cost or revenue metric and anything derived from one. Restricted values read as zero rather than as an error, and the restriction is enforced in the interface and in the APIs alike, so a restricted user pulling through the Data API or building a Looker Studio report sees zeros.

That explains a class of dashboard ticket that looks like a broken connector. Before rebuilding a data source, check whether the person viewing the report is restricted, and whether the restriction arrived as an inherited permission. You can add a data restriction at a lower level, but you cannot remove one that was applied higher up.

One caveat from Google: restrictions may not apply to users whose Analytics permissions come from another linked Google product. Worth knowing before you rely on a restriction as a control.

The Tag Manager side runs a different model

Tag Manager has its own permissions, and it is the product that can take a site's measurement offline. At account level a user is either Admin or User. At container level the ladder is No access, Read, Edit, Approve and Publish. Edit allows workspaces and edits but not versions or publishing. Approve allows versions, workspaces and edits but still not publishing. Publish allows everything.

The inheritance trap has a twist here. Direct permissions persist even when an inherited permission is lost, so a member who holds Publish directly keeps Publish after the group that gave them Edit is removed. Cleaning up a group and assuming the access went with it is how an old contractor keeps publishing.

Google's guidance on administrators is blunt. Keep at least 2 active administrators, expect a Tag Diagnostics notification when an account has only one, and make sure the account is managed by someone in your organisation rather than an external agency. The failure case is worse than a lockout: an account or container with no admin is deleted automatically, and anyone with read access has 30 days to export it from the trash. If nobody has read access, the container is gone and the site gets retagged from scratch.

Listing access across every property

If you look after more than a handful of properties, clicking through Admin per property is not an audit, it is a sample. The Admin API exposes access bindings, and one detail catches people out: the resource lives in v1alpha, not v1beta, even though the rest of what you need is in beta.

The call for a single property is a GET to https://analyticsadmin.googleapis.com/v1alpha/properties/123456789/accessBindings. The same resource exists at account level as accounts.accessBindings, which is what makes a sweep across an account practical. pageSize defaults to 200 and caps at 500, so loop on the page token. Reading needs the analytics.manage.users.readonly scope. Changing anything needs analytics.manage.users.

To enumerate the properties first, v1beta properties.list filters on parent:accounts/123, ancestor:accounts/123 or firebase_project:, takes a pageSize that defaults to 50 with a maximum of 200, and accepts showDeleted if you want trashed properties in the results.

Plan for the quotas before writing the loop: 1,200 requests per minute per project, 600 per minute per user, 600 writes per minute per project, 180 writes per minute per user, and 500 user deletions per day per property. Exceeding a limit returns a 403 and the quota refills every 60 seconds, so a short wait and a retry is the correct handling. A tighter loop just spends the next minute failing.

Who actually read the data

Holding a role and reading data are separate questions, and Google exposes the second one too. runAccessReport returns a record each time a user reads reporting data, retained for up to 2 years. It covers reads through the Analytics interface, explorations, the Data API, and linked products such as Firebase and AdMob that pull data out. It does not cover configuration changes. For those, searchChangeHistoryEvents on the account is the endpoint.

Practical limits: the method is available to Administrators only. It can be requested at property or account level, and an account-level request includes every property underneath. Dimensions not related to quota are limited to 360 properties. A request allows up to 9 dimensions, 10 metrics and 2 date ranges, with a maximum of 100,000 rows per call and 10,000 by default.

Two flags decide what the report is for. includeAllUsers brings in users who have never made an API call, and expandGroups, which only works when includeAllUsers is true, returns the people inside a user group rather than the group itself. With both set, the access report becomes an access list that names the humans behind a group whose name tells you nothing.

A quarterly routine that takes an hour

  1. Enumerate. List every GA4 property and Tag Manager container you are responsible for, including trashed properties. A trashed property can still be restored, and the access attached to it comes back with it.
  2. Read both levels. Pull the account list and the property list. The account-level grant matters most because it applies everywhere and cannot be revoked further down.
  3. Reconcile against people. Compare names and addresses against current staff, current contractors and current suppliers. An account you cannot attach to a living person is the finding.
  4. Check the links. Google Ads, Firebase, BigQuery and Search Console links each carry an access path that no user list shows you directly.
  5. Fix at the highest level. Remove the account-level grant rather than patching individual properties. Then read that member's effective permissions back to confirm the change, because a removal you did not verify is a removal you do not know about.
  6. Count the administrators. In both products. An account with one Administrator is one departed employee away from a support request that Google will not action for you.

One behaviour makes this easier than it used to be: removing a user from a source property also removes them from its subproperties, while removing them from a subproperty affects only that subproperty. If your account has subproperties for separate brands or regions, that asymmetry decides which level you edit.

What the record should contain

An access audit that ends with a cleanup has a short life. An audit that ends with a document survives a change of staff. The fields worth keeping, one row per person and resource:

There is a compliance reason as well as a housekeeping one. Accountability under UK GDPR is about being able to show what you did and when, and access is one of the few areas where the evidence is a list you can export and date. Analytics data is personal data whenever it can be tied to a person, and the list of who could reach it is usually the part of the record nobody wrote down.

None of this is difficult. It is a decision that access gets reviewed rather than set once, taken by someone who holds Administrator and is willing to use it. The properties that hurt are the ones where nobody asked, while the answer sat in Admin the whole time.

Not sure who can still reach your analytics data? Book a free tracking audit and we will list the users, the inherited roles and the product links behind them, then tell you which ones should not be there.