Different Data, Differently

A number, a trend, a ranking, a process, a place and an argument are different kinds of knowing. The web hands every one of them back the same way.

Ask a search engine how tall Mount Everest is and you get ten links. Ask it how the price of eggs has moved since 2020 and you get ten links. Ask it how to change a tire, who the best point guards of the decade are, where the nearest pharmacy is open now, or whether a bill in Congress will pass — ten links, every time, in the same font, in the same column, ranked by the same machinery.

Those are not the same kind of question. They are not even the same kind of answer. A height is one number with a unit and a source. A price over time is a line, and the line is the answer — the shape of it, where it bends, what it was before. A tire change is a sequence where order matters and skipping step four ruins your afternoon. A ranking is a list with a stated measure, or it is an opinion pretending to be one. A pharmacy is a point on a map with hours. A bill is a path through a process, and the answer is where on the path it stands.

We have spent twenty-five years treating all of this as one thing: documents that might contain the answer. The reader is left to open them, find the part that matters, and assemble it into the shape the question needed all along.

The shape of an answer is part of the answer.

What each kind wants

Look closely at the questions people actually ask and they sort, fairly cleanly, into kinds. Each kind has a form it is naturally read in — the form a good reference librarian, a newspaper graphics desk, or a patient teacher would reach for without thinking.

A facthow tall, how many, when

One figure, its unit, its date and where it is from. Everything else is decoration, and a paragraph around it is a place to hide.

A trendhow has it changed

A line over time, with the points that matter called out. A sentence like "prices rose sharply" is a summary of a chart that should have been shown.

A rankingthe best, the biggest, the top ten

A list with its measure named. Without the measure it is a vote nobody held.

A processhow to

Numbered steps, in order, each one doable. Not an essay with the steps dissolved in it.

A placewhere, how far, open now

A map, a distance and the hours. A paragraph about a place is a map nobody drew.

A comparisonthis or that

Two columns and the same rows, so the difference is visible at a glance instead of buried in alternating paragraphs.

A recordwhat a thing actually is

The documents themselves — a product's ingredients and where it was made, a bill's text and its votes — laid out, with no claim added that the record does not make.

A storywhat happened, and why it matters

Events in order, people and their stakes, what is still open. This is the one kind prose was made for.

An argumentshould we, is it true

A claim, the strongest case against it, and the evidence both sides can inspect — not a verdict delivered with confidence.

None of this is new. Encyclopedias have infoboxes. Newspapers have graphics desks. Cookbooks number their steps. Atlases draw maps. Every mature medium for knowledge learned, over centuries, that different data wants to be shown differently. The search result page is the one place that forgot.

Why it matters now

Writing machines have made the forgetting worse, not better. A language model can now answer any question in fluent paragraphs, and that is exactly the trouble: a paragraph is the shape of a story and an argument, and almost nothing else. Ask for a trend and you get a description of a trend. Ask for a ranking and you get prose that mentions some names. Ask how far something is and you get a sentence that could have been a map. The answer reads well and shows almost nothing. It is the single-column search page all over again, only smoother.

The fix is not more words. It is choosing the form before writing anything — reading the question for the kind of thing it asks, and letting that pick the shape: the figure with its source, the line, the numbered list, the map, the two columns, the record, the story. Then — and only then — words, where words help.

Different data, too

There is a second sense of the title, and it matters as much as the first. Much of what people most need to know was never written as a web page at all. It lives in records — public filings, votes, price series, recall notices, ingredient lists, meeting agendas — and in what people know first-hand and were never asked to write down. A system that only reads web pages will only ever answer from what someone happened to publish as prose.

Different data deserves to be gathered differently, too: read from the records themselves, kept with its dates and sources, and shown in the shape it has. A product's supply chain is a route, so draw the route. A city council's week is an agenda, so show the agenda. A bill's life is a path, so mark where on the path it is today.

What this changes for the reader

The test of all this is simple, and anyone can apply it: after reading the answer, could you act on it, or did you have to go and open the documents anyway? A height you can quote. A line you can see bending. Steps you can follow with greasy hands. A map you can walk. A list whose measure you can dispute. A record you can check.

That is what it means to treat different data differently. Not a new kind of page for its own sake, but the old wisdom of every good reference work, brought back to the place most questions are now asked. Ask a different kind of question, get a different kind of answer — in the shape the answer already had before anyone wrapped it in a page.