Instagram Algorithm 2026: Watch Time, Likes and Sends

In 2026, Instagram distributes reach according to signals the head of the platform has named himself. That is unusual: on most networks the weighting stays guesswork. Adam Mosseri made three signals public in January 2025, and everything that has circulated as Instagram tactics since then can be checked against them. Most of it does not survive the check.

The Instagram algorithm is not a single algorithm

Instagram points this out itself: there is no central ranking system. Feed, Stories, Explore, Reels and search each use their own model, cut to the way people use that particular surface. This explains something many business accounts experience as a contradiction. A Reel performs well in the Reels tab among strangers and barely surfaces in the feed of its own followers. Both surfaces judge the same video, but with different priorities.

The same raw signals apply across every surface. Watch time, likes and sends are the three most important ones, Mosseri said on 22 January 2025. In the operational version he named the ratios, not the absolute values: average watch time, likes per reach, sends per reach. The difference matters. A worked example: a post with 200 likes across 40,000 people reached is algorithmically weaker than one with 60 likes across 1,500 people reached.

Instagram measures watch time more generously than many assume. According to the Instagram help centre, play time covers the total time including replays of a Reel. Loops count. Short videos with a high replay rate can therefore produce the same total number of seconds as a much longer video watched through once.

Connected and unconnected reach: two audiences, two weightings

Mosseri's second statement from January 2025 is rarely quoted, yet it is the more useful one in practice: likes matter slightly more for connected content, sends slightly more for unconnected content. Connected means people who already follow you. Unconnected means recommendations to everyone else.

Only the first row of this table is documented. Mapping the surfaces is a rule of thumb from practice: Mosseri comments on the weighting of likes and sends, not on which surface delivers which audience.

What follows from this is a decision, not an optimisation checklist. If you want to hold on to existing followers, you build on recognition: the same formats, the same narrative patterns, continuity with earlier posts. If you want to grow, every post has to work without prior knowledge, because it reaches people who do not know your account. You can do both, just not in the same post. Ignore the difference and you produce content that is too generic for followers and too demanding for strangers.

Sends per reach: private forwards beat public likes

The weighting itself comes from Mosseri's explanation in January 2025: with non-followers, sends per reach count slightly more than likes. He described the usage behaviour behind it in May 2025 to Business Insider: on Instagram, far more photos and videos are shared into DMs than into Stories, and far more into Stories than into the feed. That is a statement about user behaviour, not about ranking, but it fits the logic of the signal. Forwarding something to a specific person is an act of recommendation with social risk; a like costs nothing.

The shift can be measured. For its 2026 benchmarks, Socialinsider analysed around 70 million posts published between January 2024 and December 2025: shares per post rose by 12% on Instagram and by 45% on TikTok, while comments per post fell by 16% on Instagram and by 24% on TikTok. These are international figures from brand accounts, not DACH data, and they show correlation, not cause. The direction is unambiguous all the same: discussion under the post is losing weight, forwarding is gaining.

One widespread error belongs here. The claim that sends count three to five times as much as likes has been circulating in creator and tool blogs since 2025, but nobody names a source. Mosseri says “slightly more important”, nothing beyond that. Meta has never published a coefficient. Base your content planning on that figure and you are planning on an invention.

In operational terms, the post needs a reason to be forwarded. Usually that is a concrete benefit for a third person, a verifiable number, a set of instructions, or an observation that immediately brings someone in your own circle to mind. Prompts such as “share this with a colleague” work less well than content where forwarding is the obvious thing to do.

Saves and carousels: the second qualitative signal

Alongside sends, saves are the second strong quality indicator, because they express an intention to come back to something. By format the distribution is clear: according to the Metricool Instagram Study 2026, carousels generate nine times more saves than single images, measured across a good 24 million posts from around 375,000 accounts, comparing January and February 2025 with 2026. This too is a correlation drawn from observational data, not a causal statement about format decisions.

Which format wins when depends on the goal and is a calculation of its own; the format benchmarks in detail are in Reels vs. Carousel. For the algorithmic logic one rule is enough: Reels produce watch time and with it discovery, carousels produce saves and dwell time among people who are already there.

Hashtags are no longer a distribution mechanism

In November 2024 Instagram announced that it would remove the option to follow hashtags; the switch-off followed in December. As a search term and a post attribute, hashtags remained. Mosseri had already said in 2022 that he would not think of hashtags as a route to more distribution at all. That closes out a tactic still listed as compulsory in many social media guidelines across the DACH region.

Three to five relevant hashtags as thematic classification do no harm; that is experience, not a documented reach mechanism. The real lever sits elsewhere. Instagram is used as a search surface, and for that, keywords in the caption, on-screen text, subtitles and alt text count. How search works on Instagram, TikTok and LinkedIn is described in Social SEO.

Originality: the majority rule in the 30-day window

Instagram tightened its rules for unoriginal content in 2026. Accounts that mostly post other people's photos, carousels or Reels are no longer eligible for recommendations to non-followers; eligibility returns as soon as the majority of posts within a rolling 30-day period count as original. Instagram gives a single figure for this: 75% of recommendations in the US now come from original posts.

What Instagram does not say matters more. The circulating figures “10 reposts in 30 days” and “40 to 60% more distribution for originals” appear in no Instagram source, only in tool blogs without evidence. For practical purposes the majority rule is enough, and it is strict enough: an account that mostly curates instead of producing loses its recommendation reach entirely.

Plenty of false claims circulate about the role of AI content. In his year-end memo of 31 December 2025, according to the version documented by Om Malik, Mosseri announced: “Label AI-generated content and verify authentic content. Surface credibility signals about who's posting. Continue to improve ranking for originality.” That is a commitment to labelling, verification and originality ranking, not a promise to favour human content over AI content. The often quoted line “the polished, perfect aesthetic is dead” does not appear in the memo in that form; writing about the feed of square, smoothed-out photos, Mosseri said: “That feed is dead.” What makes sense in the use of generative tools and what gets penalised is covered in AI-assisted content creation. For DACH companies, the labelling obligation under Article 50 of the EU AI Act applies from 2 August 2026, regardless of how Instagram ranks.

GEM and Andromeda: organic and paid move closer together

Meta rebuilt its advertising infrastructure between the end of 2024 and 2025. Andromeda handles retrieval and filters a few thousand candidates out of tens of millions of ads, then GEM takes over the ranking. According to Meta Engineering, the rollout in the second quarter of 2025 delivered 5% more ad conversions on Instagram and 3% on the Facebook feed.

The consequence for organic work is underrated: both systems learn from the same interaction patterns. A creative that produces high watch time and plenty of sends organically is a reliable candidate for paid amplification, and you know that before any budget moves. Organic-first testing becomes the cheapest form of creative research. The generic mechanics behind both systems, from candidate generation through to re-ranking, are described in Recommendation algorithm: how social media algorithms rank.

Instagram and Facebook in DACH: what the numbers show

Instagram is the most used social media platform in Germany, but it is no longer growing. For the end of 2025, DataReportal reports 31.3 million addressable users in Germany, a decline of 2.3% compared with October 2024. The ARD/ZDF Medienstudie 2025 puts weekly usage at 40% among the population aged 14 and over and notes that Instagram is losing younger users for the first time. The two values measure different things: one comes from Meta's advertising tools, the other from a survey.

On self-reported usage the two Meta networks sit close together. In the German country report Digital 2026, 56.2% of internet users aged 16 and over say they used Instagram in the previous month, against 58.9% for Facebook; the figures were collected by GWI in the second quarter of 2025.

On self-reported usage Facebook is more broadly anchored than Instagram, but organically it delivers only about half the reach. As a channel it stays sensible where older audiences, groups and local visibility count; that is a judgement from practice, not a documented ranking. Read the next point as a judgement too: for B2B distribution in the DACH region, Instagram is in most cases a brand and recruiting channel, not the lead channel.

Reading benchmarks without fooling yourself

Comparing Instagram figures usually means comparing denominators. Socialinsider divides interactions by follower count and arrives at 0.48% for 2025 and 0.45% for the first half of 2026. Emplifi divides by the reach achieved and reports a drop in median engagement from 16.9% in the first quarter of 2024 to 9.7% in the fourth quarter of 2025. Hootsuite works with the follower denominator again and gives 3.5% as the cross-industry average and 2.8% for Reels, data as of Q4 2024/2025.

The values do not contradict each other, they answer different questions. The engagement rate based on reach is the useful one, because distribution is decoupled from follower count anyway. The reach rate itself sits at 3.50% on average, down 12% year on year according to Socialinsider, measured on posts from May 2024 to May 2025. Which values are realistic for your segment is covered in What is a good engagement rate?. All of the studies named are international; DACH-specific engagement benchmarks barely exist.

What users can control themselves

Since January 2026 Instagram has been rolling out “Your Algorithm” worldwide to English-speaking users, initially for Reels and, over the course of the year, for Explore and the main feed as well. Users can view topics, add them and remove them. For business accounts this means part of the interest mapping becomes explicit instead of implicit. Thematic consistency in an account becomes more important as a result, because an account that cannot be clearly assigned to a topic simply does not appear in a manually curated topic list.

What is documented as of August 2026

The most stable basis remains the three signals Instagram has made public itself: watch time, likes per reach, sends per reach, plus the mapping of likes to follower reach and of sends to reach among strangers. The majority rule for originality and the statement that hashtags bring no distribution are documented as well. Everything beyond that, everything working with exact factors, thresholds and golden hours, has no verifiable source. Point your Instagram work at watch time, sends per reach and originality and you are working with what is documented.