The goal is simple: less time wading through noise, more time for what's worth reading. AI Filters read every new article as it arrives and decide what to do with it, based on a description you write in plain English — no keywords, no rules to wire up, you just describe the kind of article in your own words. The job is to thin out your list so you spend less time in the app, not more.
Describe the kinds of articles you consider noise. As each new article comes in, the AI judges whether it matches your description. If it does, the article is marked as read for you automatically — so it never clutters your unread list. Everything else stays unread as normal.
The same idea in reverse: describe what you most want to see — breaking news, topics you actively follow — and matching articles get starred so they stand out. Leave it blank to turn AI starring off.
Here's a real "auto mark as read" filter for someone who follows finance but doesn't want the day-to-day churn. You can write several categories at once — give each a short label, then describe it. Paste something like this into the box and tweak it to taste:
Daily market mechanics "Articles primarily about single-day movements in stock prices, indices, commodities, or cryptocurrencies — including 'X up/down Y% today,' index milestones, or daily trading recaps with no underlying news event driving the move." Earnings beat/miss theater "Articles focused on a company beating or missing quarterly estimates by small margins, pre-market or post-market reactions to earnings, or 'what to expect from earnings' previews, unless they describe a fundamental business change, guidance shift, or strategic announcement." Analyst noise "Articles whose main content is an analyst price target change, rating upgrade/downgrade, or 'Wall Street reacts' commentary, with no new underlying business development." Political horse-race coverage "Articles about daily polling fluctuations, hypothetical matchups, pundit reactions, or 'who won the news cycle' commentary, rather than substantive policy developments, legislation, or election outcomes."
What this filter does to your unread list:
Notice the "unless..." and "rather than..." clauses — those are escape hatches. They tell the AI when an article that looks like noise is actually worth keeping, so genuine news doesn't get marked read along with the churn. Being specific about both what to hide and what to spare is what makes a filter accurate.
You don't have to describe everything yourself. Beyond the filters you write, Pro builds a reading profile from what you star, open, and rescue. Over time it learns the topics you actually care about and the kinds of stories you skip — so it removes more noise on its own, while protecting the subjects you keep tabs on. Think of it like a trusted friend who only interrupts you for the things genuinely worth your time, and stays quiet about the rest. The profile is plain text that's yours to read and edit anytime in Settings, so you're never guessing what the AI thinks you like.
It stays honest, too: everything the AI marks read shows up in your Marked Read By AI list, newest first — nothing is hidden. If it ever sets aside something you wanted, mark it unread to rescue it, and that correction teaches your profile for next time.
Looking for something else? Visit Help & Support for the full guide and how to reach me.