How to build a Google Analytics dashboard

Search for how to build a Google Analytics dashboard and you'll land in a confusing place, because GA4 contains three different answers and the word dashboard fits all of them badly. There's the Reports snapshot - the overview screen GA4 opens with, which editors can rearrange. There are custom overview reports - your own card layouts, built inside GA4's report library. And there's Looker Studio, Google's separate free tool, which is what most people picturing "a dashboard" actually want: a canvas of charts you design, share and schedule.

This guide walks up through the three levels in order of effort, because each one is enough for somebody and overkill for somebody else. What a dashboard is *for* - the always-on view of your standing questions, as opposed to the periodic report you send people - is covered in the general dashboard post; this one is the GA4 mechanics.

One piece of advice before any mechanics: write down the three to six questions your dashboard should answer before you open a single settings menu. "Is traffic normal, which sources move, are conversions happening" is a dashboard. "Every metric I can find" is wallpaper. Every decision below gets easier with the questions in hand.

The cheapest dashboard is the one GA4 already shows you. The Reports snapshot - the first screen under Reports - is a stack of cards, and with the Editor or Administrator role you can reorder them, remove them, and add new ones: drag to rearrange, the x to drop a card, Add cards for more. Changes apply for everyone who uses the property, which cuts both ways - your improvements help the whole team, and your experiments confuse them.

Don't confuse the snapshot with the Home page, which looks similar and isn't yours to control. Home is personalized by Google - it watches which reports you open and picks cards accordingly, and your adjustments there are session-deep. It's a lobby, not a dashboard. If you've ever "fixed" your dashboard and watched it revert, you were on Home.

For a small site, ten minutes of snapshot editing is often the whole job: put traffic, sources and key events on top, delete the cards you never read, done. The level above exists for when one screen has to serve several distinct questions.

GA4 lets editors build custom overview reports - your own card layouts that live in the left-hand navigation like the built-in screens. The flow runs through the report Library (bottom of the Reports navigation): create an overview report, then fill it with summary cards. Cards themselves are created from detail reports - open any detail report, Customize, create a card from the dimensions and metrics on screen - and each card links back to its source report, so the overview doubles as a menu.

Two limits worth knowing before you invest an afternoon. An overview report holds at most 16 summary cards - a real cap, easy to hit if you're building one screen for a whole team. And a property holds at most 150 custom reports in total, which sounds distant until an agency shares a template pack. Design within the 16: one card per question, and if a question needs three cards, it's probably two questions.

You can even set your custom overview as the default Reports snapshot, which is the closest GA4 gets to "my dashboard when I log in". For a team living inside GA4, level 2 is usually the right stopping point.

When the requirements outgrow cards-in-GA4 - a client wants their logo on it, you want Search Console and analytics side by side, someone asks for a TV screen - the answer is Looker Studio. It's free, connects to GA4 natively, and is a genuine design surface: drag charts anywhere, mix data sources on one page, share by link, and schedule the finished page as an email. The template gallery means you rarely start from blank.

The catch nobody's template mentions: Looker Studio talks to GA4 through the Analytics Data API, and the API has quotas. Busy dashboards - many charts, many viewers, aggressive date ranges - can exhaust them, and the symptom is charts replaced by quota errors at exactly the moment someone important is looking. It's a managed-scarcity problem rather than a dealbreaker: fewer charts per page, extracted or cached data where possible, and no thirty-chart pages. But go in knowing the free canvas has a meter running behind it.

A fair test for whether you need level 3 at all: does anyone outside the analytics tool need to see this? If the audience all has GA4 access, level 2 does the job with less to maintain. Looker Studio earns its keep when the dashboard is itself a deliverable.

Whatever level you build at, the data underneath obeys GA4's rules, and three of them regularly get misread as broken dashboards. Freshness first: GA4 processes data in tiers - realtime cards within minutes, intraday data over a few hours, and fully settled daily data on a longer lag. A dashboard checked at 9am describes yesterday-ish, not this morning, and a number that shifts between morning and evening isn't a bug.

Second, thresholding: with Google Signals enabled, cards slicing small segments can withhold rows entirely to protect privacy. On a modest site, a demographics card is often a blank card. Third, sampling - GA4's built-in reports are generally unsampled, but anything built on Explorations samples once a query crosses 10 million events, so a card fed by an exploration on a busy property is an estimate. For most small sites this never triggers; for the sites where it does, it explains a lot of "why don't these two numbers match".

Resist adding a seventh until one of the six has earned retirement. The full menu of metrics exists when a real question calls for it - the dashboard is for the standing questions, and standing questions are few.

Six cards, one per question - a layout that fits comfortably inside any of the three levels:

• Visitors over time, this period against the last - the is-anything-unusual card, and the one that catches a spike or drop early.
• Traffic by channel - which of the source buckets moved. This is where marketing effort becomes visible.
• Top pages - what's carrying the load this week, and whether anything new entered the list.
• Key events - conversions, assuming you've defined them; an empty card here is your to-do list, not a data problem.
• One engagement number - engagement rate or its inverse, as a health check, not a goal.
• Realtime actives - honestly optional, but it's the card that makes people open the dashboard, and there are worse tricks.

Build more than one of the levels above and you'll eventually catch them disagreeing - the Looker Studio page says 4,120 sessions, the GA4 card says 4,187, someone asks which is broken. Usually neither. The surfaces read data at different moments in the freshness pipeline, apply date boundaries in different time zone contexts, and an API-fed chart can lag the interface by hours. Add thresholding hitting one surface and not another, and small disagreements are the steady state, not an incident.

The working rule: treat differences under a few percent as noise, pick one surface as the number of record for anything people will quote, and investigate only when the gap is large or growing. Chasing every two-percent discrepancy between your own dashboards is the analytics equivalent of setting two watches and being upset.

A number alone answers nothing - 4,200 visitors this week means one thing after a 3,000-visitor week and another after 6,000. Every level above supports comparison, and it's the single setting that most improves a dashboard: date-over-date on the snapshot cards, a previous-period line in overview report charts, side-by-side date ranges in Looker Studio. Set each card to show this period against the last comparable one, and pick the comparison honestly - a seasonal business compares to last year, not last month.

The same goes for segments. A card can carry a comparison between mobile and desktop, or new and returning visitors, and the split often *is* the story - flat total traffic hiding a mobile rise and a desktop decline is three cards' worth of insight in one. Use segment comparisons sparingly though: one per card, or the card stops being glanceable, and glanceable is the whole contract.

Who sees what differs sharply by level, and it's worth knowing before someone asks for access. Inside GA4, dashboards are shared by giving people access to the property - snapshot changes and published overview reports are visible to everyone on it, which is clean for a team and wrong for a client you'd rather not hand the whole property. Looker Studio flips this: the dashboard is a document with its own sharing, viewable by link without any GA4 access at all - the reason it wins every client-facing scenario.

For the people who won't open anything, scheduled delivery closes the loop: GA4 can email standard reports on a schedule, and Looker Studio can send the finished dashboard as a PDF. A dashboard someone receives beats a better one they'd have to remember to visit.

Dashboards decay socially before they decay technically. The card added for a campaign outlives the campaign; the metric someone important requested outlives their interest; within a year the screen answers questions nobody asks. The maintenance is a calendar habit, not a technical one: once a quarter, ask of each card "did this change a decision since last quarter?", and retire the ones that didn't. A dashboard that shrinks over time is being used; one that only grows is being performed.

And notice whether the dashboard gets opened at all. If it doesn't, the fix is rarely more cards - it's usually a scheduled report delivering the three numbers people actually want to where they already are.

Start with questions, then pick the cheapest level that answers them: a rearranged Reports snapshot for one person, a custom overview report inside the 16-card cap for a team in GA4, Looker Studio when the dashboard is a deliverable of its own. Respect the freshness lag, expect thresholding on small segments, and prune quarterly. And if the honest problem is that GA4 makes even the simple dashboard feel like work - a common conclusion - tools built around one good screen exist: Bigdelta's opens on live visitors, sources and pages with the spike alerts built in, no assembly required. The best dashboard is still the one that gets looked at.