tube search bars route you instead of finding things

the search bar on a tube isn’t a search bar. it’s a router. you type something, and it hands you a page built to keep you on the site as long as possible, with relevance as one input among several.

once you see that, “porn site search not relevant” stops being a bug report and starts being a business model.

what a tube search bar is optimizing for

a tube makes money on impressions and sessions. every page load is ad inventory, and every extra click is another pop-under chance. the number that matters internally is how many pages you view before you leave.

a search that puts the exact clip you wanted in slot one ends the session in a single click. a search that returns something adjacent, plus a strip of related searches, plus a row of promoted tiles, keeps you moving. so the ranker gets tuned on engagement, and engagement is a different thing from what you asked for.

this is also why porn search results are all the same. the ranker leans on views and recency, and the clips at the top get more clicks because they’re at the top. type ten different queries in one category and you’ll watch the same dozen thumbnails rotate through. the popularity loop eats the query.

the same query on xhamster, spankbang and eporner

pick something boring and specific, like “amateur couple”. not a tag with a million entries, just a phrase with a clear meaning, and run it on all three.

what you notice first is that the pools barely overlap. each site is searching its own inventory, so some of the split is just different catalogs. the rest is what each one wants you to see first.

xhamster mixes creator and channel content in near the top, which makes sense for a site pushing its uploader ecosystem. spankbang sorts hard on popularity, so the first screen is long-lived clips with big view counts. eporner is the closest to plain keyword matching with a quality sort, and it has the least chrome around the results.

all three re-tune constantly, so your week-to-week results will drift. the pattern holds though: the word you typed is matched against the pool, and then the site’s own priorities reorder it.

the unpopular take: filters beat keywords

if you want to know how to search porn sites better, stop rewriting the query and open the filter drawer.

keywords change how the ranker scores things inside a pool. filters change the pool. adding a fourth adjective to your query gets you a slightly different ordering of the same popular clips, while a duration floor of 15 minutes deletes the entire economy of 3-minute teasers that exist to push you toward a paysite.

a date range does the same to the ten-year-old reupload with a rewritten title. a quality floor kills the upscaled 480p. none of that can be expressed in words, and all of it changes what you see more than your vocabulary does.

almost nobody uses this. i don’t have any site’s analytics, but the design tells you what they expect: with roughly 87% of adult traffic on mobile, the tube site search filters live behind a small pill or icon that sits away from where thumbs land. the default sort is the site’s own “relevance”. a drawer that stays closed is a drawer that never gets in the way of the session-length metric.

here’s the pushback i expect. a duration floor will drop a great 8-minute clip, and a date filter will hide a good older upload. true. i’ll take that trade every time, because the cost of a miss is one more search, and the cost of no filter is ten minutes scrolling the same pool the ranker wanted you in.

why sponsored slots sit inside the organic results

because the grid is the inventory. an ad drawn as a thumbnail, at thumbnail size, in the thumbnail grid, gets clicked far more than a banner in a sidebar. that’s why the sponsored tile and the organic tile are usually built from the same component, and it isn’t an accident.

promoted tags work the same way. the chips above your results, or the “trending searches” strip, often lead somewhere other than a category. some are paid placement pointing at a cam site or a paysite, dressed up as a tag.

the practical fix is boring. before you click something that looks like a result, check where it goes. on desktop, hover and read the domain in the status bar. on mobile, long-press and read the link preview. anything that leaves the tube’s own domain is an ad.

this is also why “why won’t the tube just give me what i typed” has an uncomfortable answer. the people who control the search box are paid for the near miss.

what a search across sources has to solve first

people call the thing they want an adult video search engine, meaning one box that searches everywhere. the obvious way to build it is to query six sources, merge the results and rank them. that’s the wrong order.

in our index the same scene shows up on 4-7 platforms. rank first and dedup later, and page one becomes two scenes repeated five times each. the ranking looks fine right up until a person reads it.

so dedup comes first, and it’s harder than it sounds. titles get rewritten per tube. thumbnails get recropped or watermarked, so exact image matching fails. the most stable signal is duration, matched within a couple of seconds, combined with loose title overlap and a perceptual hash of the thumbnail.

once you’ve collapsed those into one scene with a list of sources, ranking gets interesting. a scene that exists on six platforms is a better signal of demand than any one tube’s view counter, because no single site’s algorithm inflated it. that’s a popularity score nobody can push with a promoted slot.

and the source list becomes a choice you hand back to the reader. same scene, six places to watch it, and you get to pick the one that treats you like an adult.

that’s the shape we’re building at prowl, and it can’t find anything a tube doesn’t already have. it just stops the tube from deciding what you were really looking for.

the search bar isn’t going to fix itself, because the people who own it get paid when it misses.