The AI Assistant

Feedpik has an assistant inside the reader. It is not the same thing as connecting your own ChatGPT or Claude — that gives an outside agent access to your account. This one is already here, on every plan, and it is the fastest route to a feed that behaves.

Open it from AI Assistant at the bottom of the sidebar. On a wide screen it takes a fourth column rather than covering what you were reading.

Three places it opens, three things it knows

The button is the same; the context is not.

Opened from	What it has in front of it
AI Assistant in the sidebar	Your ranking, interests and sources
Ask AI on an open article	That article's full text
Inside a draft	The draft and its attached materials

So "what is the actual claim here" works on an article without pasting anything, and "less crypto" works in the sidebar without naming a setting.

Each message costs 2 credits, from the same monthly allowance as the article actions. See plans and AI credits.

Say what you want, in your own words

The assistant's real job is the ranking. You do not need to know what a weight is:

more AI infrastructure and chip supply, much less crypto and celebrity gossip

I keep seeing the same story from five sites. Fewer duplicates.

nothing about sport, ever

It answers with a proposal card — the change spelled out as chips (block "crypto", +interest "chip supply", recencyWeight → 0.4) and a one-line summary. Nothing is written until you press Apply. A proposal sitting unapplied costs you nothing, which is why there is no Cancel button next to Apply: ignoring it is already the way to refuse.

Why one answer has several buttons

A proposal can contain two quite different things, and they are deliberately not one decision.

The settings half — interests, muted topics, languages, weights — is one unit with one Apply and one Revert. It only reorders what you already receive, and undoing it is a click.

The sources half is a list of separate offers, each with its own button. Subscribing changes what arrives every day, and it comes undone one feed at a time.

Forcing those together meant agreeing to a ranking change also meant accepting ten subscriptions, and disliking one suggested feed meant rejecting the ranking change too. So take the ranking, take three of the five sources, leave the rest. Each source reports its own outcome — a URL that fails to resolve says so rather than being counted as added.

Undo

Revert on the card puts the settings back. Beyond that, every applied change is a new version of your ranking, and Algorithm & Agents lists them — so a change you accepted last week is still reversible after the card is long gone. See tuning your ranking.

This is why the assistant proposes rather than asks permission at every step. The safety comes from reversibility, not from interrogation.

What it will not do

It does not read your mind between messages. It changes what you ask it to change. If the feed is still wrong, say what is still wrong.

It cannot spend money. Upgrading, buying credits and cancelling are yours alone.

It is not a search engine. Asking it for news it has not been given produces the failure every chatbot has; ask about an article you have open, or about your own ranking, and it is on solid ground.

If something looks wrong

I applied a change and the feed looks the same — weights apply to the next draw. Refresh Recommended.

The card says applied but a source is missing — subscribing fetches each URL, and one can fail while the others land. The card names which failed and why.

It proposed something odd — do not apply it. An unapplied proposal has changed nothing, and rewording the request usually beats arguing with the answer.

I am out of credits — every message costs 2, and the allowance is shared with summaries and translations. Plans and AI credits covers topping up.