Your Data and Privacy

The short version: your reading is used to rank your own feed, and that is the only thing it is for. There is no advertiser, no data buyer and no second purpose.

This page is the long version — what is actually stored, because a privacy claim you cannot check is just a slogan.

What is recorded

Your explicit actions. Save, like, not-interested. You sent these deliberately and they are the strongest inputs to your ranking.

How long you stayed on an article. Measured only while the tab is visible and you have done something in the last fifteen seconds, and only counted past 30 seconds. An article left open in a background tab records nothing. This is a real ranking signal and it is the one people do not expect, which is why reading your feed says so too.

What arrived and whether you opened it. Ordinary reader state.

What you told the ranking — interests, muted topics, languages, weights — and every previous version of it.

That is the list. There is no cursor tracking, no scroll-depth profile, no cross-site anything, and no attempt to work out who you are from what you read.

What it is used for

Ranking your feed. That is the whole of it, and the ranking is not a black box you have to take on faith — the scoring code is open source, and Learned from your reading in Recommendation Settings shows you the picture it has built.

Three things it is deliberately not used for, covered in more depth under what the ranking never does:

• No engagement optimisation. Nothing is boosted for being provocative or for keeping you scrolling. There is no term in the scoring for how likely you are to react.
• No advertising. Nobody can pay to rank higher. There is no inventory to sell.
• No profile for anyone else. Your reading is not a product, and it is not shared between accounts.

Who else can reach it

Agents you connect, and only to the extent you allowed. A connection carries an explicit list of permissions shown on the consent screen, and you can revoke any key instantly. Agent access and tokens covers the nine permissions and what each reaches.

Two things worth knowing about that boundary:

• No permission lets anything message anyone on your behalf. Digests are sent by the scheduler, not under a user grant.
• Erasing the account needs a signed-in session. An agent holding a key with every permission still cannot delete the account that issued it.

Anyone you share a persona with, minus the private parts. Sharing carries sources and ranking. It never carries paid-subscription feed URLs or newsletter inbox addresses — those are skipped, and the result tells you how many. The optional reading-behaviour summary is per-source totals only: no individual articles, no timestamps. See personas.

Nobody else. Your feed, saved list and settings are not visible to other readers, and personas are private until you publish one.

Taking it out

Two exports, for two purposes:

Export data, in Recommendation Settings, gives you your profile, interests and ranking as JSON. This is the one to take if you want a record of what Feedpik thinks about you.

Export persona, on the Algorithm & Agents page, gives you a reading space — sources, folders and algorithm — in a file that imports back into any account. This is the one to take if you want your setup somewhere else.

Both are plain JSON you can open and read. That is the point: an export you can inspect is a claim you can check.

Deleting it

Delete account, in Recommendation Settings, is self-service and permanent. It asks you to type your own email address as confirmation.

It refuses while a subscription is still active — cancel first, then delete. That order is deliberate: deleting an account with live billing is how someone ends up paying for something that no longer exists.

There is no recovery afterwards and no grace period. Export first if there is anything you want.

Account security

Sign-in is by email. Password reset goes to the address on the account.

Access keys are shown once. If one is lost, revoke it and issue another — there is no retrieval.

The audit log records every change, including which actor made it, so an agent's changes are distinguishable from your own. It is the first place to look if something changed and you did not change it.

If something looks wrong

I want to stop the dwell-time measurement — it cannot be switched off separately, but Learn from saved and liked articles and Learn from hidden or skipped articles in Recommendation Settings control how much any of your reading history counts. Turned down, it is recorded and barely used.

An agent changed something I did not want — revoke its key first, read the audit log second. Revoking is instant and total; diagnosing can wait.

I deleted a persona by mistake — it is gone, along with its sources and folders. Persona exports exist for this and take one click beforehand.

Cancel refuses, so I cannot delete — cancelling and deleting are separate steps in Manage Account and Recommendation Settings respectively. If cancel itself is failing, the error offers the payment provider's own portal.