For fifteen years, the Facebook Feed has been a black box with my name on it. I never chose what showed up. An algorithm chose, trained on every like, pause, and doomscroll I'd ever committed, and I was just... along for the ride. Every few years someone would ask "why am I seeing this?" and the honest answer was always "nobody outside Menlo Park really knows."
On October 13, 2026, Meta did something I genuinely didn't expect: it handed me the steering wheel. Or at least, it started to.
The company announced "Your Algorithm" — a set of AI-powered tools that let Facebook users shape what appears in Feed and Reels by describing what they want in their own words. Not toggles. Not hidden settings menus. Words. You literally tell Facebook what you're into, and the ranking changes in real time.
I've spent the last day digging into how it works, what Meta isn't saying, and what it means for anyone who builds for the platform — creators, pages, and developers. Because if you run a page like I do, this changes the game under your feet.
What "Your Algorithm" Actually Is
Let's get the mechanics straight, because Meta's announcement — titled "Your Algorithm, Your Words: New Ways to Shape What You See on Facebook" and posted to the Meta Newsroom on October 13 — is short and light on numbers. Here's what it describes, in three parts.
Part one: a dedicated space called "Your algorithm." This is the headline feature. It's a place inside Facebook where you can directly add and manage the topics you want to see in Feed and Reels. Meta describes it as "a dedicated space to tell us exactly what you're into." You open it from the three-dot menu on any post, and you type — in natural language — what you want more or less of.
Part two: conversational feedback on individual posts. The existing "Interested" and "Not interested" buttons got smarter. After you tap one, Facebook may ask follow-up questions to get more detail on what you liked or disliked — and you answer in your own words. "You can easily tell us, in your own words, what you want to see more or less of, and your recommendations will update in real time," Meta wrote. They call it "a more direct, conversational way to shape your experience."
Part three: the three-dot menu redesign. That little "..." on every post is now the front door to all of this. Meta rebuilt it to make the customization tools easier to reach.
The technical bit that matters: large language models now sit inside the ranking loop. Meta says its ranking systems already use thousands of signals — likes, comments, videos watched, all the behavioral exhaust. The new tools add explicit, written intent on top. An LLM interprets the context and intent behind your words, so "I want more tech news but fewer sponsored gadget posts" becomes something the ranker can actually act on. A like button could never express that distinction. A sentence can.
And the kicker: it works fast. Meta says recommendations can respond in real time, including within the same session. You type what you want, and what appears next changes. No waiting weeks for the algorithm to "learn" you. You just... tell it.
This Started on Instagram
One detail worth knowing: Facebook isn't first. Meta says it launched this feature on Instagram earlier, and October 13 was the Facebook rollout. That matters because it means the system has already been battle-tested on one of the world's most aggressive recommendation surfaces. Instagram Reels is where Meta's ranking is most ruthless, and if natural-language controls survived there, they're probably more than a demo.
It also tells you where Meta thinks this is going. The pattern is: prove it on Instagram, then bring it to the bigger, older, more complicated surface. Facebook Feed has fifteen years of legacy behavior baked in. If "Your Algorithm" can tame that, it can go anywhere — including, presumably, Threads and WhatsApp surfaces down the line. I'm speculating on that last part, but the trajectory is hard to miss.
What Meta Isn't Telling Us
Now the part where I put on my skeptical hat, because Meta's announcement is conspicuously thin in all the places that matter.
No numbers. The Newsroom post gives no user counts, no test results, no A/B data showing that written feedback actually improves anything. For a company that A/B tests the color of buttons, shipping a ranking change with zero published metrics is... a choice.
No rollout details. Meta hasn't said which countries get it first or on what schedule. "Rolling out" is doing a lot of heavy lifting in every article about this. If you're reading this outside the US, assume you don't have it yet.
No ad clarity. This is the big one. Meta did not say whether your typed preferences are used for ad targeting. Think about that. You're handing Meta clean, structured, natural-language statements of your interests — "I'm into marathon training and sourdough" — which is vastly more valuable than inferred signals. Does that flow into the ad auction? Meta's silence here is deafening, and I'd bet good money the answer is yes, eventually, if not already.
No weighting transparency. How much does a written instruction count versus three years of watch history? Meta hasn't said. "Your Algorithm" might be a steering wheel connected to the wheels, or a placebo button. I can't prove which one it is, and neither can you.
I'm not saying it's fake. The LLM-in-the-ranking-loop architecture is real and technically coherent. I'm saying: trust, but verify — and Meta hasn't given us much to verify with yet.
What This Means for Creators and Pages
Here's where it gets personal for me, because I run Facebook pages, and this announcement rearranged my mental model of distribution.
Until now, the creator's job was: make content the algorithm's behavioral model will reward. Optimize for watch time, comments, shares — the inferred signals. It was a game of reading tea leaves.
"Your Algorithm" adds a second game: be the thing people explicitly ask for. If a user types "more woodworking videos," and you make woodworking videos, you don't need to game engagement metrics — you need to be legible to the intent-matching layer. That means clear topics, clear titles, clear descriptions. The LLM needs to understand what your content is about to match it to someone's stated interests.
This is actually good news for niche creators. The old algorithm favored broad-appeal content because behavioral signals aggregate toward the mainstream. Explicit intent favors the specific. A user who writes "I want deep-dive videos about mechanical keyboards" is a high-intent audience of one, and matching is deterministic in a way virality never was. The long tail just got a search box.
But there's a darker flip side. Users can now explicitly say "less of this." Before, if someone was tired of your content, they'd just scroll past, and the algorithm would slowly get the hint. Now they can type "fewer prank videos" and you're gone from their feed in the same session. The punishment for being mildly annoying got faster and more final. Creators who relied on algorithmic inertia — content people watched but didn't love — are the most exposed.
My practical advice: audit your page's topic legibility. If an LLM read your last twenty posts, could it describe in one sentence what you're about? If not, you're invisible to the new layer. Clarity is the new SEO for the feed.
The Developer Angle: LLMs in the Ranking Loop
As a developer, the architecture here is the most interesting part. Meta has put large language models inside a production ranking system serving billions of requests — not as a chatbot bolted on the side, but as an interpreter between human intent and the ranker.
Think about what that implies. Somewhere in Meta's stack, there's now a component that takes "fewer celebrity gossip posts, more science explainers" and converts it into ranking adjustments in real time, blended with thousands of behavioral signals. That's a genuinely hard systems problem: low latency, massive scale, and the output has to be a better feed, not just a plausible one.
It also represents a philosophical shift in how Meta thinks about personalization. For a decade, the doctrine was: never ask the user, infer everything. Explicit preferences were considered unreliable — people don't know what they want, the data knows better. "Your Algorithm" is Meta publicly walking that back. The new doctrine seems to be: inference for the subconscious, conversation for the conscious. Both signals, one feed.
If you're building recommendation systems, watch this closely. The "explicit + implicit" hybrid with an LLM as the translator is going to become the industry template. Whoever figures out the weighting — how much to trust what users say versus what they do — wins the next generation of feeds. Meta just made its bet public.
My Take
I've been cynical about "user control" announcements from social platforms for years, because they've historically been theater — a settings page nobody visits, a toggle that changes nothing. So I approached this one with my guard up.
Here's what lowered it: the real-time part. A placebo control doesn't need to work in the same session. The fact that Meta is promising immediate, visible changes means users can test it themselves in about thirty seconds. Type "less sports," scroll, check. If nothing changes, the internet will know by dinner. That's a falsifiable claim, and companies don't make falsifiable claims about features that don't work. Usually.
What raised my guard again: the ads question. Until Meta says plainly whether typed interests feed the ad machine, I'll assume the product here is still you — just with better labeling. "Your Algorithm" might mean "the algorithm, configured by you, still optimizing for Meta's revenue." Those two readings coexist just fine, and I suspect both are true.
The optimistic read: this is the beginning of the end of the black-box feed. Once users expect to be able to talk to the algorithm, every platform will have to offer it, and the era of "the algorithm decided, don't ask questions" starts to close. The pessimistic read: it's a pressure valve — give people the feeling of control so they stop demanding actual control, like data portability or chronological feeds.
I land somewhere in the middle. This is a real technical capability, shipping to real users, with real implications for creators. It's also, conveniently, a great story for Meta to tell regulators. Both things can be true. My job is to tell you how to use it, not to resolve Meta's motives.
How to Use It (When You Get It)
When it reaches your account: open the three-dot menu on any post, find "Your algorithm," and add specific topics ("AI coding tools," not just "technology"). Use Interested/Not interested aggressively for the first week and answer the follow-up questions in full sentences — you're training it, so give it good data. Then scroll and check whether the mix actually shifted in the same session. If it didn't, say so publicly; user reports are the only QA this feature has. Revisit monthly, because interests drift.
Conclusion
"Your Algorithm" is Meta's most honest feature in years — not because the company suddenly loves transparency, but because LLMs finally let it ask what you want instead of guessing. The black box isn't open, but for the first time, it has a suggestion slot.
For users: try it, test it, don't trust it blindly. For creators: become legible — the feed is learning to read. For developers: explicit intent plus implicit behavior, translated by LLMs in real time, is the new template. Learn it before your competitors do.
