GA4 Task Assistant: What It Recommends and What It Cannot See

At the bottom of the left menu in Google Analytics there is an item called Tasks. Every property has it. Almost nobody opens it. It is the property's own checklist: a list of setup jobs Google believes are unfinished, grouped by what they affect. It shipped on 29 April 2026 as Task Assistant, and the changelog entry calls it "tailored recommendations to help you optimize your configuration and improve data collection".

Read that sentence closely and it tells you what you have. The recommendations are about configuration. They are generated from what the property believes about its own settings. That is not the same thing as an audit of your measurement, and treating it as one is how an agency ends up telling a client their tracking is finished while the numbers are still wrong.

The dashboard breaks into six categories and 21 tasks. Some deserve your time on a UK property. Several are noise. And there is a whole class of tracking failure that never appears in it at all.

Where it lives, and who can open it

Sign in to Analytics, look at the bottom of the left navigation, click Tasks. The help page describes it in two steps, which is a fair signal of how little surface area the feature has.

Permissions decide who sees it. You need the Administrator, Editor, or Marketer role. Analyst and Viewer cannot view or access the dashboard at all. That has a practical consequence on a client property: the person most likely to ask "is our GA4 set up properly?" is often sitting on Viewer access, so they cannot see the panel you are about to discuss with them.

There is a second consequence, and it is the one that catches agencies out. Task status is shared. Google repeats the same line for every task: once marked complete, "anyone with access will be aware of the task's updated status". Ticking a box is not a private note. It changes what the client sees the next time they open their property. Marking something complete to clear noise off the list is a visible act with an audience.

Skipping fails in the other direction. Collapsing a task removes it from view and records nothing. There is no field for "not applicable, and here is why". The dashboard keeps no memory of your reasoning, and neither will the next analyst unless you write it down somewhere the property does not control.

The six categories and what sits in them

Twenty-one tasks across six headings, counted from the current documentation:

• Get started (6): Set up data collection, Turn on Google signals, Set up key events, Create audiences, Enable enhanced measurement, Confirm privacy settings.
• Connect your accounts (1): Link to Google Ads.
• Enhance your reporting (2): Create custom insights, Create audiences.
• Optimize your advertising (3): Create a conversion action, Target ads to Analytics audiences, Link to Google Ads.
• Add first-party data (6): Set up user-ID, Set up and use measurement protocol, Set up user-provided data, Import campaign data, Import item data, Import offline event data.
• Fix data issues (3): Resolve missing deep links, Fix misconfigured deep links, Exclude spammy users.

Every task offers the same three options. Take action deep-links you into the relevant settings screen, usually the exact tab. Watch video opens a walkthrough. Skip collapses the entry. Google Ads linking shows up twice, once as a plain connection and once under advertising, which is duplication rather than a bug.

The task UK properties should not skip

Confirm privacy settings carries a line worth quoting in full: "Data missing consent signals may impact your measurement capabilities in regions with consent regulations." That sentence is doing real work. In the UK, consent regulations are the whole game, and the task exists because a property can look healthy in every other report while quietly losing consent state.

Take action drops you into the Consent settings tab under Data collection and modification. What you find there is a switch, not a diagnosis. The task checks that a consent setting is configured. It does not check that your consent management platform fires before the Google tag, that the consent update lands before the first page view, or that the default state inside your container is set to denied in the first place. A property can show this task as complete and still be discarding consent signals on every single session.

If a UK property shows this one as outstanding, treat it as the highest priority item on the list. If it shows complete, verify it anyway. The UK Consent Mode v2 guide covers the checks that matter, and the GDPR consent checklist covers the paperwork around them.

Which tasks are worth your time

Working down a client property, this is roughly the order things are worth attention:

• Fix data issues, first. Spam referrals and missing session_start data corrupt everything downstream, including the reports you will use to judge whether the other fixes worked. Sort these before anything else. GA4 not collecting data covers the diagnostic path when the problem is deeper than referral spam.
• Confirm privacy settings. Covered above. Non-negotiable on UK properties.
• Set up data collection. The task is marked complete once one tag is firing somewhere. If the client runs subdomains, a separate checkout domain, or a booking system on a third-party host, the tick tells you nothing about whether those are covered. Check data streams and recent traffic per hostname yourself.
• Enable enhanced measurement, then trim it. Turning it on collects page views, scrolls, outbound clicks, site search, video engagement, file downloads, and form interactions with no extra code. That is useful, and it is also how a property ends up with a scroll event marked as a key event and a conversion count nobody can explain. Check the key events list against the business, not against the template. The event taxonomy guide covers naming and structure.
• Link to Google Ads. Worth doing when your key events actually feed bidding. Check the link points at the right advertising account rather than an old agency account nobody monitors, and that the imported conversions are the ones you intend to optimise against.
• Set up user-ID and user-provided data. Only with authenticated users, a lawful basis, and a reason. Sending hashed identifiers from a site with no login is scope creep, and it expands what your client has to defend in a data protection review.

What Task Assistant cannot see

The dashboard reads configuration state. It does not read implementation quality, and the gap between those two things is where most broken tracking actually lives. None of the following appear in Tasks:

• Duplicate events. A purchase event fired twice, once from a GTM tag and once from a hardcoded gtag call, is invisible here. Revenue doubles and nothing on the checklist notices.
• Malformed event payloads. A purchase event missing currency, or an item array with parameters that never got set, drops data silently. GA4 accepts the event and reports less than you sent.
• Cross-domain and subdomain splits. Traffic that crosses from a marketing site to a checkout host without proper configuration becomes two users. That failure sits in your domain settings, not in the task list.
• Container behaviour. Tag sequencing, consent checks inside individual tags, triggers attached to the wrong pages. Task Assistant does not read your GTM container at all.
• Campaign parameter problems. Missing aggregate identifiers (GBRAID and gad_) surface in a separate diagnostic, added on 30 July 2026, and campaign data import quality surfaces in a validation report added on 10 August 2026. Neither is a task. Both are earned by reading the property's diagnostics directly.
• Data retention, filtering, and internal traffic. Retention windows, data filters, and internal traffic definitions are configured elsewhere, and the checklist does not push you toward them.
• Whether anyone uses the reports. A property can complete every task and still have nobody who trusts the numbers.

The honest summary is that Task Assistant answers "is this property configured?" It does not answer "is this measurement correct?" Those are different questions, and only the second one decides whether a client's decisions are built on anything real.

How we run it on client properties

Four working rules, learned from properties where the checklist had been ticked and the tracking was still wrong:

1. Open it before you touch anything, and screenshot it. The panel is the client's own view of their gaps. It is useful evidence in a conversation about scope.
2. Treat a completed task as a claim, not a fact. A tick means somebody clicked Show on a notification. Verify the underlying state before you rely on it.
3. Record your skips in a document you own. The dashboard forgets. Your audit notes should not, because the skip decision usually comes back six months later as a question.
4. Re-run the pass after every migration, CMP change, or site launch. Property state changes, and the recommendations regenerate against it.

One more: do not hand the panel to a client contact with Viewer access and assume they can maintain it. They cannot open it. Either upgrade the role deliberately or keep the routine on your side of the fence.

A 20-minute pass

For a property you are meeting for the first time:

1. Open Tasks, count what is outstanding per category, take the screenshot.
2. Work Fix data issues first. Referral spam and missing session_start data distort every judgement you make afterwards.
3. Check Confirm privacy settings against the container's consent default, not against the toggle on the settings page.
4. Confirm each domain and subdomain shows traffic in the last 48 hours. The task will not tell you this.
5. Read the key events list and the enhanced measurement events side by side. Cut what does not map to a business outcome.
6. Decide on the first-party data tasks explicitly. User-ID and user-provided data are choices with a lawful basis attached, not defaults to switch on.
7. Write the skips and their reasons into your own audit document, then diarise a re-check in 30 days.

That pass takes about twenty minutes and it will surface more real problems than a year of ticking boxes. The checklist is a reasonable place to start. It is a poor place to stop, because it can only see the property's opinion of itself. Somebody still has to look at the data.