Social Media Engagement Metrics by API: 2026 Study

Which social media APIs give an agent a save or share count?

X documents bookmark_count as a public post metric, and TikTok's API for Business documents favorites in its post metrics. TikTok's Display API documents no save field. Instagram documents saved for Feed posts and Reels only, LinkedIn's member post analytics added POST_SAVE in API version 202604, and Pinterest documents SAVE only for date ranges going back up to 90 days, with no lifetime save total. Bluesky's bookmarks are private, and its docs do not say whether bookmarkCount returns a count to the author. Share counts are widely documented: 8 of 9 platforms document one, and Pinterest is the exception.

Can an agent get follower count at the time of a post?

On most platforms, only if it stores the count itself. X, Threads, Bluesky and Pinterest document today's follower count and no history. TikTok, LinkedIn and YouTube document a follower history with no extra condition, and Facebook documents one for Pages with 100 or more likes. These are daily series, so they give a post's publish day, not its exact time. TikTok's covers the last 60 days and its docs describe the daily value both as a total and as a net gain, LinkedIn's do not say which, and YouTube's are daily gains and losses. Instagram's follows_and_unfollows metric returns a total for a date range, and not at all below 100 followers, and the docs do not establish a follower series. To get rates by post, store the count at publish time.

When should an agent read the numbers?

Each platform sets its own read limits. One rule per platform, labelled with its API:

• Instagram API: Media insights are stored for up to 2 years, with a delay of up to 48 hours. developers.facebook.com
• Facebook Pages API: Most metrics update once every 24 hours. developers.facebook.com
• TikTok API for Business (Accounts API): Post data stops updating 365 days after the post is published. business-api.tiktok.com
• LinkedIn member post analytics: Counts are described as best-effort. learn.microsoft.com
• X API v2: Non-public, organic and promoted metrics are only available for posts created within the last 30 days (public_metrics has no such limit stated). docs.x.com
• Threads API: followers_count cannot be queried by date, so it returns the current value only. developers.facebook.com
• Bluesky (AT Protocol): postView counts carry no stated window, retention limit or processing delay in the lexicon or reference pages checked. The lexicon documents no time semantics for likeCount, replyCount, repostCount, quoteCount or bookmarkCount (indexedAt is the post's index time only). github.com
• YouTube Data and Analytics APIs: Analytics API data is not real-time. Processing typically adds a 48 to 72 hour latency. EndDate is truncated to the last fully processed day. Docs recommend the Data API for real-time counts. developers.google.com
• Pinterest API v5: Pin analytics start_date cannot be more than 90 days back from today. developers.pinterest.com

Why do engagement rate benchmarks disagree?

They count different interactions and divide by different numbers. 3 of the 8 publishers we checked (Buffer, Dash Social and Socialinsider) use more than one engagement-rate definition for the same platform in their published benchmark figures, and most also give readers several formulas in their guides. Buffer's 2026 engagement report, which prompted this study, divides likes, comments, shares and saves by followers on one page, with variances for each platform, and divides engagements by reach, or the closest equivalent, on another. Its blog also gives two values for X's 2025 median. To score your own rate against a set of published benchmarks, each on its own basis, see the social media engagement rate calculator.

Does the API document the inputs for a published engagement formula?

This checks whether each input is documented, not whether a publisher's multi-brand figure can be reproduced. Of the 23 benchmark rows checked platform by platform, 13 of 23 use only inputs the platform's API documents (2 of those with conditions), and 10 of 23 count clicks or other interactions this study did not check. Buffer's report formula is applied with variances for each platform, so it is listed but not checked platform by platform. Where a benchmark divides by followers, the matching number is the count on its measurement date, which X, Threads, Bluesky and Pinterest document only as today's count, and Instagram's docs do not establish as a series.

Check a formula

Pick the interactions and the number you divide by. The table shows which platforms' APIs document every input for your own posts. It looks up fields only and computes no rate.

What this means for builders

If your agent tunes what it posts on saves, keep it to the 5 platforms above that document a save count, within their conditions, and do not rely on a save count from the other 4. Store a follower count with every post where no history is documented, record when each metric was read, and match post age when comparing counts. Related: API changes since October 2024 and what platform rules let an AI agent do.

Method

Every documented metric is quoted verbatim from the platform's own developer pages, every absence names the pages checked, and each is classified under rules written before collection, with disclosed clarifications.