How search ranks results, and what the result count means

Why is this the top result, and what is the number next to each tab counting?

The order of your results is set by two searches running at the same time: one for your exact words, one for your meaning. Everything found is scored by how high it placed in each of those two lists, with meaning weighted a little heavier than exact words. Nothing in the ranking looks at how recent a piece of work is, how many likes it has, or who made it.

Where each of the two searches stops

The word search finds every entry containing the words you typed. The meaning search turns your query into a position on a map of meanings and returns whatever sits closest to it, whether or not those entries contain your words. Each side puts forward its best 200 entries, and only those get ranked.

Two limits follow from that, and they explain most odd-looking results.

  1. Very common words are ranked on meaning alone. Once your words appear in more than 5,000 entries, scoring each one for how well it matches is too slow to do while you wait, so the word side drops out and the order comes entirely from the meaning search. A single broad word is well past that line. Add a second, more specific word and the word search comes back into play.
  2. Past the first 200 results, ranking stops. Keep scrolling and what arrives next is the remaining entries containing your words, newest first. That is an order, not a judgement about relevance.

The number on a tab counts matches, not good results

The figure beside Projects, People or Media is how many entries contain your words. It is not a claim that all of them are worth opening, and it does not shrink as the matches get weaker. A large number against a common word is expected.

Filtering image results by content type

The content type filter on image results (Photos, Floor Plans, Sections and the rest) works from a list capped at 50,000 images per type. Photos is far past that cap, so the filter draws on a portion of them, and it is not the same portion on every request. The same filter can return a different set from one search to the next. Adding words to your query narrows results more dependably than the filter does.

Two more things worth knowing

Related