Instagram algorithm in 2026: what changed

Instagram changed two things in 2026. It kept telling people which signals rank content, and it started letting people edit their own recommendations. A lot of what you read about the first one is invented. This post separates what Instagram confirmed from what marketing blogs made up.

What Instagram actually confirmed

Adam Mosseri, the head of Instagram, named the three signals that matter most across the app. They are watch time, sends per reach, and likes per reach.

Read those carefully, because two of the three are ratios.

• Watch time. How long people watch, not how many saw it.
• Sends per reach. The share of viewers who send your post to someone in a DM.
• Likes per reach. Likes divided by how many people saw it.

The ratio part is the useful part. A post seen by 1,000 people with 50 likes has a better like rate than one seen by 10,000 with 100 likes. Instagram weighs the rate, not the total. This is why follower count alone stopped explaining reach.

Instagram says these signals apply to Feed, Reels, Stories, and Explore, but carry different weight on each surface. That is the whole of the official position. It is less specific than most articles claim.

What is claimed but not confirmed

Search for the Instagram algorithm and you will meet very precise numbers. One DM share is worth 15 likes. A save is worth 10. Reach for normal accounts fell from 28-42% of followers to 10-20%. Saves were promoted to the second-highest signal in February 2026.

None of that comes from Instagram. These numbers appear in marketing blogs, without a source, and they contradict each other. The same search returns "sends carry 3-5x more weight than likes" and "one send is worth 15 likes" on the same day. Both cannot be right.

We are not saying the direction is wrong. Sends probably do count for more than likes, and Instagram's own framing supports that. We are saying the exact multiplier is not public. Nobody outside Instagram can measure it, because nobody outside Instagram sees the ranking model.

This matters when you plan work. A made-up number invites a made-up target. If you set a goal around "15 likes per send", you are optimising against fiction.

Your Algorithm: from a setting to a surface

This is the real 2026 story, and it gets far less coverage than the fake multipliers.

Your Algorithm gives each person a dashboard of the topics Instagram thinks they care about. You can add topics you want more of, and remove ones you do not. Instagram pre-fills suggested tags based on your activity.

The rollout has been steady. Instagram tested it on Reels in late 2025 and launched it in December 2025. It reached Explore in April 2026, and the main feed by mid-2026.

In June 2026, Instagram went further. It began testing ways to reach the controls without opening settings at all. Pulling down on the feed can surface the Your Algorithm menu. Swiping up from a Reel can open a similar prompt.

Mosseri described the goal directly: "We want to evolve Your Algorithm from a setting to something that feels central to your experience on Instagram." He also added a caveat worth keeping: "Some of this is testing now, some is coming soon, some might not work."

One detail from the reaction is worth noting. The most popular replies to Mosseri's post asked for something simpler: show the accounts people already follow.

Why user-controlled ranking changes your measurement

Every ranking signal discussed above is behavioural. Instagram watches what people do and infers what they want. Your Algorithm is different. It asks them.

If stated preference becomes a real input, two things follow for anyone tracking reach.

First, your audience gets a filter you cannot influence with content quality. Someone who removes your topic stops seeing you, regardless of watch time.

Second, reach becomes harder to compare over time. A drop in July against May may reflect people editing their own topics, not your posts getting worse. Keep that in mind before you rebuild a strategy around one bad month.

Original content and reposts

Instagram has said it gives original photos and carousel posts more priority in recommendations. Accounts that mostly repost other people's content are not recommended in Explore.

This is a policy direction rather than a number, which makes it easier to act on. If your account is mostly aggregated content, Explore is not a realistic source of reach.

What shipped, May to July 2026

Beyond ranking, Instagram shipped a long list. The ones with practical weight:

• Photo comments. People can reply with an image under posts and Reels, not only text or emoji.
• Instants. The standalone app launched globally in May, and gained video recording in July.
• Series. A June test for episodic Reels. Selected creators can group Reels into episodes, with a hub on the profile.
• Grid reorder. Everyone can now rearrange posts on their profile.
• Repost removal. You can remove other people's reposts of your posts.
• Screenshot prevention for Stories. Rolling out through June.
• AI Creator label. A new account label for AI-generated content.
• Carousel captions. A test for per-image captions inside a carousel.
• Business hours and contact banners. A July test in the Banners section.
• Trial Reels. Continued global rollout.
• Edits. Steady updates, including bilingual captions in 15 languages, Reels search in the Inspiration tab, 15-minute exports, and a desktop version in development.

Two of these are quietly useful for a business account. Trial Reels let you test with non-followers before committing. Series gives serialized content a structure it never had.

What to do with all this

Three plain reads.

Track rates, not totals. Instagram's confirmed signals are ratios. If your reporting shows likes and views as raw counts, it does not match how you are ranked. Sends per reach is the one most people do not track and should.

Distrust precise claims about the model. If an article gives an exact weight for a signal, ask where the number came from. In almost every case the answer is nowhere.

Sends and saves are not something to fake. Both signals exist because they are hard to game. They measure whether one person thought a post was worth sending to another. Bought engagement does not produce that, and buying it puts the account at risk for a signal that was designed to resist it.

If you are picking tools to schedule or measure any of this, our SMM tools playbook covers what to check first. For how the wider tool market is moving, see SMM tools in 2026.