Find out who a field actually listens to.

Bluepicks reads Bluesky's likes to work out which accounts a topic takes seriously. Those accounts are its Top Voices, and the feed is built from the posts they vouched for. Here's who Tech listens to right now.

Open the Tech feed →

One topic so far. More on the way.

How it works

1

The topic found itself

I take a running sample of everything posted to Bluesky, pool each account's words together, and group accounts by the vocabulary they actually use. Sixty groups fell out of that. One of them was AI and software development.

Nobody decided the topic in advance, and nobody wrote the keyword list. The longer answer ↓

2

Likes decide who's listened to

A like is a vote, and a vote from someone this topic already listens to counts for more. That makes the score circular by definition, so it gets solved as a random walk across the whole like graph. Follower counts and profile text play no part at all.

1,348 candidates in, 150 Top Voices out. The longer answer ↓

3

Every post needs a second

A post reaches the feed only if it clears two gates: it was written by a Top Voice, and at least two other Top Voices liked it. No single person can put anything in front of you.

What Bluepicks doesn't do

  • It doesn't count followers. A large audience earns nothing here. Neither does a keyword-stuffed bio.
  • It doesn't reward posting more. Volume is measured only so it can be divided out. Posting more can lower your rank; it can never raise it.
  • It doesn't know who you are. No account, no sign-in, no tracking. Everyone opening the feed sees the same posts.
  • It doesn't have a curator. Nobody adds a favourite or quietly removes someone they dislike. The list is whatever the like graph says it is.

Questions

Where did the "Tech" topic come from?

From the posts, not from me. I keep a rolling sample of English-language Bluesky posts, pool every account's words into one bag, and run k-means over 313,533 accounts that posted at least five times. That produced sixty groups. I read them afterwards and found one built out of mcp, codebase, refactor, npm, qwen3 and friends. That group holds 3,808 accounts and is clearly AI and software development. Naming it "Tech" was the only human decision involved.

Who chose the keywords?

A loop did. The words that most distinguish that group from everyone else become the first search terms. I search Bluesky with them, look at whoever turns up, work out what those people distinctively write about, add those words to the search, and go round again until the vocabulary stops moving.

I checked whether the starting point matters by running the same loop twice, once from words I picked by hand and once from words the machine picked. Both converged on nearly the same vocabulary. That's the property the whole project rests on: it can be pointed at a field I know nothing about.

How does an account become a Top Voice?

Accounts that keep turning up under those keywords go into the pool, 1,348 of them for Tech. I fetch their likes and build a graph: an edge runs from whoever liked a post to whoever wrote it, with repeat likes between the same pair compressed, so liking one person 200 times doesn't count as 200 endorsements.

Then a random walk crosses that graph. Landing on an account is the score, and each account splits its vote among the people it likes. So a vote from a well-regarded account is worth more, and someone who likes everything spreads their vote thin.

A few conditions sit on top of that: enough of the topic's vocabulary to show the account belongs there, and enough different people liked to show it reads the topic as well as posts into it. There are also defences against groups of accounts trying to like each other onto the list. The exact cut-offs, and how those defences work, are the one part I keep to myself. 224 accounts clear everything, and the top 150 are the Top Voices.

Does posting more help?

The opposite. The walk itself never sees post counts, but the final score is divided by how often the account posts. Someone who earns the same regard on two posts a day ranks above someone who needed twenty. It's the correction that keeps high-volume accounts from crowding out everyone else.

Why isn't an obvious name on the list?

Most likely because they don't like other people's posts. An account has to have liked a fair number of different people in the topic to be ranked. That is how news feeds, link bots and cross-posters drop out, since they broadcast without reading.

The cost is real and I accept it: genuine people who simply never press like are excluded too. I don't hand-restore them. The moment I start adding favourites by hand, the ranking stops meaning anything, and it stops working in fields I know nothing about.

What exactly is in the feed?

Posts written by one of the 150 Top Voices in the last 30 days, that at least two other Top Voices liked. They're ordered by how many Top Voices liked them, weighted towards recent posts so the feed keeps moving. How much arrives on a given day depends on how busy the topic is.

How often does the list change?

The feed updates continuously. The Top Voices list is recomputed periodically rather than live, so accounts join and leave it in steps, not by the hour.

What do you know about me?

Nothing. There's no account to make and no sign-in. The feed isn't personalised, so there's nothing to learn about you and nowhere to put it. Everyone reading the Tech feed is reading the same posts, in the same order.

I'm on the list and I'd rather not be.

Tell me at @bluepicks.app and you're removed, from the published list and from the feed both, no reason needed, and it stays that way. Bluepicks only ever reads posts and likes that are already public, but being singled out by name is a different thing from posting in public, and that's your call to make, not mine.

When are other topics coming?

Tech is the first because it's the one I could check by eye. The same pipeline has already been run over other communities, and topics will be added once each one holds up. If there's a field you want, say so at @bluepicks.app.

Open the Tech feed →

Free, and it works with whatever Bluesky client you already use.