Nearly 45% of Instagram's recommended Reels are less than a day old, according to figures Meta presented for the third quarter of 2026. The share increased by more than 11 percentage points from the previous quarter. The figures were disclosed on October 21, when the company detailed changes to its recommendation systems and developments in the use of its platforms.
The metric concerns the age of videos included in recommendations rather than all posts on Instagram. The increase in percentage points describes the difference between two shares, with the comparison made against the second quarter. Meta presents the figures as results from its own systems and measurements of the content distributed to users.
Instagram also evaluates the sequence of recommended videos
According to chief financial officer Susan Li, relevance changes introduced in the third quarter increased time spent with Instagram Reels by more than 4%. The company attributes part of the increase to an architecture that ranks sequences of videos rather than evaluating each Reel solely as an independent item in the recommendation ranking.
Meta also identifies richer signals about watch time. In the company's description, selection considers the experience of consuming a sequence of material and the responses of the person receiving it. Platform-level usage figures describe the aggregate effect of these changes on viewing, separately from the performance of a particular post or an individual account's content.
Globally, Instagram time spent per person reached a company-reported high, with double-digit growth from a year earlier in the third quarter. Meta links the development to Reels and Feed. The comparison concerns time spent using the platform, rather than an identical increase in followers for each account publishing on the service.
A new architecture selects posts on Facebook
On Facebook, Meta says it completed the global launch of a generative retrieval architecture for Feed and Reels. The system assigns semantic representations to content and proposes material for recommendations. The company explains that this mechanism can identify relevant new posts before they accumulate a large volume of engagement, broadening the signals available for initial selection.
The architecture handles roughly a quarter of unconnected Feed views and half of Reels views. Meta attributes a 1% increase in Reels watch time to it. The distribution percentages describe the surfaces served by the system, rather than the proportion of all Facebook accounts that publish videos or the number of creators using a new feature.
Separately, for Facebook Reels, the company doubled the amount of history its ranking model can consider. Meta reports a 1% increase in watch time in the United States from that change. The geographical scope is specified separately from the global figures about the new architecture and from the results presented for Instagram.
Content understanding and Instagram search
Large language models are also used to interpret public Instagram posts. According to Meta, the signals obtained through these systems improved engagement with search results by 1% through the end of the quarter. The metric concerns search within Instagram, separately from web search engines and the traffic received by business websites.
During the same quarter, the company began moving its content-understanding models to its own Muse family. In offline evaluations presented by Meta, the new models outperformed the previous production models and reduced per-unit processing costs by more than 80%. These results concern internal evaluation and content processing rather than the prices charged to advertisers for campaigns.
Threads also recorded an increase of more than 20% in time spent per user compared with a year earlier. That measurement is distinct from the number of active accounts. In its presentation, Meta tracks usage duration, views and engagement separately, according to the application and component under examination, rather than treating them as interchangeable measures.
A post's age and originality are separate criteria
The new figures about recent material continue a direction Meta described in January. At that time, the company reported that 75% of Instagram recommendations in the United States came from original posts. That indicator concerned content provenance in a defined market. The nearly 45% share reported now concerns the age of recommended Reels, meaning the two figures measure different things.
Instagram previously described a mechanism that initially shows eligible material to a small audience, including people who do not follow its creator. Content receiving favourable responses can then be distributed to larger groups. In its 2024 explanation of this approach, the platform included originality and compliance with its rules among the eligibility conditions for recommendation.
Recommendation rules differ from the rules governing whether a post can remain on the platform. Meta explains that material can stay published but be excluded from spaces where the system suggests content to people who have not chosen to follow its author. This distinction concerns distribution rather than the automatic removal of the post itself.
For professional accounts, Instagram introduced its Best practices area in 2024, organised around content creation, engagement, reach, monetisation and guidelines. It combines general information with personalised recommendations about account activity. This is a separate tool from the aggregate figures Meta has now presented, rather than a new feature released alongside the quarterly report.
The new figures describe changes to selection, content freshness and app usage during the third quarter. For commercial publishing, organic content, paid advertising and actions subsequently taken on a website are different categories of activity. The percentages Meta disclosed in this presentation concern the specific metrics and surfaces explicitly identified by the company.
