The Five Things New Accounts Get Wrong

Every one of these produces a feed that looks broken while the product is doing exactly what it was built to do. They are in the order people hit them.

1. Judging the feed before adding sources

A new account has no subscriptions, so Recommended is drawn entirely from the shared pool — everything Feedpik follows across all readers. It is genuinely interesting and it is not yours.

People then tune interests, watch the feed barely change, and conclude the ranking does not work. It works; it is reordering a pool rather than a library.

Interests reorder what is there. Sources decide what is there. Nothing you set will make the feed feel personal until you have subscribed to things. Fifteen or twenty is where it starts; importing OPML gets you there in one step.

2. Leaving languages unset

The pool defaults to English. Your own subscriptions are never filtered by language — only the pool is.

So a reader who works in Japanese or German subscribes to a handful of sources, sees a feed that is mostly English, and has no way to tell that a whole category of material exists and is being withheld. Nothing on screen says so. There is no error, because nothing is wrong.

Set your content languages in Manage Account or Recommendation Settings. It is a thirty-second fix for a problem that looks like the product being bad at your language.

3. Tuning everything on day one

The ranking learns from what you save, like and spend time on. On day one it has none of that, so the sliders are operating on nothing.

Worse, tuning six things at once makes the result unattributable — the feed changes, you cannot say which change did it, and the next adjustment is a guess on top of a guess.

Read normally for a week first. Then change one thing at a time. Every change is a version you can roll back, which makes single changes cheap and simultaneous ones needlessly expensive.

The exception is muted topics, which are worth writing immediately — they need no history to work.

4. Muting single short words

Muted topics match as substrings, and this catches everybody once.

Mute ai and it fires inside "said", "detail", "again", "maintain", "campaign". Since muting is the heaviest lever in the ranking, you have just pushed down a large and arbitrary fraction of everything.

The feed then looks broken in a way that seems unrelated to what you did, and the mute list is the last place anyone looks.

Use specific multi-word topics: ai infrastructure, celebrity gossip, crypto prices. Bylines are the opposite — they match exactly, so spell the name the way the publication does.

5. Expecting Radar and Tracking to work on a thin feed

Both measure agreement between your sources, so both need enough sources that agreement is possible.

Trending Radar requires at least two distinct sources covering the same story within 48 hours. With eight subscriptions across eight subjects, no two of them will ever converge, and the radar is correctly empty.

Tracking's cross-source pickup has the same shape — it counts other sources you follow covering a story, so a narrow feed reports zero pickup on everything.

Neither is broken. They are relative measures on a set that is too small. Fifteen or twenty sources in one subject area is roughly where both start producing signal.

The pattern behind all five

Four of these five are the same mistake wearing different clothes: treating a setting as the fix when the input is the problem. Sources are the input. Languages decide which sources are even eligible. Radar and Tracking measure relationships between sources.

Feedpik gives you unusually direct control over the ranking, and that makes it tempting to solve everything there. The ranking can only order what arrives.

If something still looks wrong

I did all five and it is still not right — say so to the assistant in plain words. It proposes a change and shows you what it will do; nothing is written until you press Apply.

I broke it and cannot tell how — roll back to a version you remember liking in Recommendation Settings rather than diagnosing the current one. Working forwards from a good state is faster every time.

A source I added has no articles — some publishers post rarely. Give it a day. See when a source stops updating.