Problem
One dataset, four audiences. The editor wants to know which posts did well, the growth lead wants acquisition, the ad ops lead wants revenue, and the founder wants one number. Serving all four on one page serves none of them.
Solution
A small fixed set of named tabs, each a complete view for one kind of question. Overview, Content, Audience, Revenue. The tab names are the taxonomy, and they are chosen from the reader’s questions rather than from the data model.
The distinction from ordinary multi-page structure is what the split is along. Multi-page splits by subject or by volume. Tabs-as-genres splits by kind of question, which is why the tab names are so often verbs-in-disguise: Content means “what did we publish and how did it do”, Audience means “who are they and where from”. Naming them after data tables—Events, Sessions, Users—moves the burden of translation back onto the reader and is the most common way this pattern fails.
The set has to be small and stable. Four to six tabs is a set a person learns. Nine is a menu, and once it is a menu the tab strip is navigation and should be a sidebar instead. Stability matters because the value is muscle memory: a reader who knows Revenue is third gets there without reading.
The Overview tab is a special case and usually the one done worst. It is not a place for the panels that did not fit elsewhere. Its job is to answer the top question and to route people to the right tab, which makes it the only tab whose purpose is partly navigational.
Shared state is the same requirement as any page set: time range and filters must carry, or the tabs are four dashboards that happen to share a header.
Use when
One dataset genuinely answers several distinct kinds of question, the audiences differ, and the set of question-kinds is stable enough to name once.
Don’t use when
The tabs would be sequential steps rather than parallel questions, or the set keeps growing. A tab strip that gained three entries this year is a navigation problem being deferred.
Trade-offs
Tabs hide everything except the current one, so cross-genre comparison is impossible and the tab nobody opens rots invisibly. The names are load-bearing and hard to change once people have learned them, which means an early naming mistake persists for years. Analytics on tab usage usually reveal that one tab carries most of the traffic, which raises an uncomfortable question about the other three. And tabs imply equal weight, when in practice one is almost always primary.
Checklist
- Are the tabs kinds of question, or names of data tables?
- Would the intended reader recognise their question in a tab name?
- How many tabs, and is the set stable?
- Does the Overview tab route people, or is it a leftovers drawer?
- Do time range and filters carry across tabs?
- Does the URL identify the tab, so a link opens the right one?
- Which tab gets the traffic, and do the others justify their place?
- Can a reader tell what is on a tab without opening it?
- What is the process for adding a tab, and has it been used recently?
- Would a sidebar serve better at this count?
Compare
Google Analytics popularised this arrangement for web data and shows both sides of it: the genre names are recognisable to non-experts, and the set has grown across versions until the strip became a tree. Plausible deliberately refuses the pattern, keeping everything on one page, which works because it carries far fewer metrics and is a direct statement that tabs are a symptom of scope. Sentry uses top-level genres—Issues, Performance, Releases—that are genuinely different questions about the same events, which is the pattern applied correctly at product scale. Grafana has no tab primitive within a dashboard, so genre tabs are separate dashboards linked in a row, and keeping them consistent is entirely manual.
Related
Multi-page dashboard is the general form and covers the state-sharing requirements. Semantic grouping is the same instinct applied within one page. Sidebar and canvas is what a tab strip should become once it stops being a set. Single-column narrative is the alternative for an audience small enough to serve with one sequence. Saved view is what readers build when the tabs do not match their actual question.