Instagram's "Your Algorithm": Your Audience Can Now Switch You Off
- Your Algorithm rolled out from Reels testing in October 2025 to Explore in spring 2026 and the main feed in June 2026, syncing across all three surfaces.
- The shift is from inferred behavior to declared interest: users explicitly state topics they want more or less of, rather than the algorithm guessing.
- For creators this cuts both ways — you can be opted out of, but you can also be opted into by exactly the audience you want.
- The practical response is topic clarity. Ambiguous accounts lose in a system built on explicit topic matching.
For fifteen years, the deal on social platforms was this: the algorithm watched what you did and guessed what you wanted. Creators optimized against the guess.
Instagram has been quietly dismantling that arrangement. As of June 2026, its "Your Algorithm" controls reached the main feed, which means the average user can now state outright which topics they want more and less of. Including yours.
What it is and where it landed
Users tap the slider icon with two hearts in the top right of Reels, Explore, or the main feed. Instagram populates suggested topics based on their activity, and they adjust from there — more of this, less of that. On Explore there are also topic pills at the top for faster changes. Adjustments sync across all three surfaces.
| Date | Milestone |
|---|---|
| October 2025 | Initial testing on Reels |
| December 10, 2025 | Feature launch |
| Spring 2026 | Expanded to Explore |
| June 2026 | Extended to the main feed |
Adam Mosseri has outlined four further control surfaces in testing: a pull-down panel from the home screen, in-stream topic controls between Reels, quick signal buttons within the Reels function bar, and a conversational refinement interface.
One clarification worth making, because a lot of August coverage got it wrong: this is not a new ranking algorithm. Instagram did not ship a ranking change in July or August 2026. It restated existing signals and continued the originality enforcement from April. What changed is who supplies the input, and that is a big enough change on its own.

Declared interest beats inferred behavior
This is the actual story, and it is worth sitting with.
Your content used to compete for attention. Now it can be excluded from the competition before it starts.
Inferred behavior is forgiving. If someone scrolled past your last three posts, the algorithm might still show them a fourth — behavioral signals are noisy, and platforms hedge. Declared interest is not forgiving. If someone says "less of this topic," that is an unambiguous instruction, and there is no ambiguity for you to survive in.
The obvious reading is that this is bad for creators. The less obvious reading is that it is symmetrical. Users who declare "more of this topic" are telling Instagram to actively seek out content like yours — a level of intent that behavioral inference could never produce. If your topic is clearly defined and genuinely wanted, you just gained a mechanism you did not have before.
The people who lose in a declared-interest system are not the small accounts. They are the ambiguous accounts — the ones that cannot be filed under any topic a user would choose. If nobody can opt into you, the opt-out mechanism is all you get.
The second change nobody is talking about
Alongside the controls, Instagram's account classification reportedly works on a rolling set of recent posts, not on long-term positioning.
Read that again if you run an established account. It means your topic classification is determined by what you posted lately, not by what you have been for three years. A stretch of off-topic content does not dilute your classification gradually — it can rewrite it.
Combine the two mechanisms and the consequence is sharper than either alone: a few off-niche posts change how you are classified, and users have opted in or out at the topic level. Drift into a topic your audience opted out of and you can lose reach with people who never unfollowed you and never disliked anything you made.
This also explains a category of reach complaints that get misdiagnosed as shadowbans. If your reach fell after a content shift, and your posts are still visible and your account is in good standing, reclassification is a more likely explanation than a penalty. We drew that distinction for aggregator accounts in the Instagram aggregator problem, and the general version in shadowbanned even with a different IP.

What to do about it
Make your topic obvious to a machine, not just to a human. Topic classification reads text. Put your subject in your bio, your captions, your alt text, and your closed captions — hashtag influence has continued to decline and those fields have picked up the discovery role. If your niche is only conveyed by vibe and visuals, the classifier is guessing.
Audit your last twenty posts as a set, not individually. Ask what topic a stranger would assign to that group. If the answer is fuzzy, that is your classification problem, and no single post will fix it.
Find out what Instagram thinks you are about. There is no official readout, but there is a decent proxy: look at what Instagram recommends to you in Explore and Reels while logged into the creator account, and look at who appears in your recent non-follower reach. If neither resembles your intended audience, you and the classifier disagree — and the classifier wins.
Be deliberate about topic shifts. Not never — deliberately. A gradual, signposted pivot is survivable; alternating between two unrelated topics is the worst possible pattern, because it prevents any stable classification from forming.
If you need to cover multiple topics, use multiple accounts. This is the honest structural answer, and it is the same conclusion Douyin's tier system and Xiaohongshu's tag system push toward. One account per topic is not a growth hack in 2026; it is what the classification systems are built to reward. The practical guide to splitting is in how to build a social media matrix strategy.
That last option has an obvious operational cost — several accounts means several content streams, and copying the same posts between them defeats the entire purpose. This is the gap local multi-account tools like NoobClaw are built for: each account runs its own niche, persona and keywords so the output differs by design rather than by discipline. The tool keeps them distinct; deciding which topics deserve their own account is still your call.
FAQ
Can I see whether users have opted out of my content?
No. There is no creator-facing view of topic preferences, and there is unlikely to be one — it would expose individual users' declared interests. The closest proxy is watching reach among non-followers over time and correlating drops with content shifts on your side. If reach to existing followers holds while non-follower reach falls, classification or topic preference is a plausible explanation.
Does this replace the ranking signals I already optimize for?
No, it sits on top of them. Watch time, sends or DM shares, and likes still rank content, with early performance weighted heavily. Your Algorithm affects the candidate pool — whether your content is eligible to be ranked for that user at all. Signals decide the order; topic preferences decide who is in the room. See why sends outrank likes.
Should I post about trending topics outside my niche?
Under rolling classification, this is riskier than it used to be. Chasing an unrelated trend does not just underperform — it can contribute to reclassifying your account. If you do it, keep it rare and connect it explicitly to your usual subject in the caption so the text signals still point the right way. The generic reach a trend buys is rarely worth the classification cost.