What Does a "Like" Mean to the Platform? — Checking the Official Documents of Facebook, Instagram, and X

To the person tapping it, a “like” on social media is a light expression of sympathy. But seen from the platform company that operates the service, what kind of data is a like? The common belief that “the more likes a post gets, the more people it reaches” is often repeated, yet rarely discussed on the basis of the platforms’ own explanations. In this article, drawing solely on documents that Meta (Facebook and Instagram) and X (formerly Twitter) publish themselves — the ranking explanations in the Meta Transparency Center, the ranking explanation on the official Instagram Blog, the recommendations explanation in the X Help Center, and the actual algorithm code that X open-sourced in 2023 — we organize the meaning each platform officially assigns to likes (all quotations were verified on the original pages on July 25, 2026).

The Conclusion First — To the Platform, a Like Is “Material for Prediction”

Comparing the official documents of the platforms side by side, the answer to the question in the title is clear. Seen from the platform, a like is, before being a point awarded to the poster, input data (a signal) for predicting how users will react. Its uses come down to the following two.

Because of ②, the common belief mentioned at the outset — that accumulating likes helps a post surface — is supported by the official documents. However, it is not a direct scoring scheme. None of these platforms orders posts by “most likes first.” For each viewer, they predict “the probability that this person will react to this post with a like, a reply, and so on,” and decide the display order from those predictions. Likes already accumulated take effect indirectly, as input data that pushes those predictions up.

One more point common to Facebook and X: a like by itself is weighted more lightly than reactions that generate conversation, such as replies and comments. X makes this explicit through the weights in its published code, and Meta through its policy documents. We will look at each in turn.

Facebook (Meta) — Likes as a Representative Example of “Thousands of Signals”

The Relationship Between Signals and Predictions

“Our approach to Facebook Feed ranking” in the Meta Transparency Center explains the mechanism that decides the order of the feed as a structure that “makes predictions from signals (input information).”

Then, for each of these posts, we consider thousands of “signals” to make predictions about what you will find most interesting.

(Then, for each of these posts, thousands of "signals" are considered in order to predict what you will find most interesting.)

And likes are named explicitly as a representative example of these signals.

Many of these signals are pieces of information you give us directly when you like or share a post, you connect with a friend or Group, or you comment on a Page’s post

(Many of these signals are pieces of information you provide directly when you like or share a post, connect with a friend or Group, or comment on a Page's post.)

“Friends’ Likes” Surface a Post

Meta publishes explanatory documents called “system cards” for its major AI systems, and the system card for Facebook Feed contains a description that corresponds, from the post’s side, to “the effect of likes”: the friends’ likes attached to a post are listed as an example of input signals.

These signals might include who created the post and how you previously interacted with them, whether the post is a photo, a video or a link, or how many of your friends liked the post.

(These signals may include who created the post and how you have previously interacted with that person, whether the post is a photo, a video, or a link, and how many of your friends liked the post.)

In other words, when a post accumulates likes, the value of the signal “how many of your friends liked the post” rises in the feeds of the likers’ friends, making the post more likely to surface there. Likes are designed to influence a post’s distribution not only on the liker’s own screen but also through the liker’s social connections.

However, the Weight of a Like Differs from Person to Person

The same document also states explicitly that the weight of the like signal is not uniform.

Another example is that for some people, “liking” a post is a strong indicator that they found that post valuable, whereas for others (such as people who don’t use the “Like” button), spending time reading the post may be a more useful prediction.

(As another example, for some people, "liking" a post is a strong indicator that they found the post valuable, whereas for others — such as people who do not use the "Like" button — the time spent reading the post may be a more useful basis for prediction.)

As a Matter of Policy, “Posts That Generate Conversation” Come First

In a 2018 Newsroom post, Meta declared that it would shift its ranking away from the sheer quantity of engagement, such as reaction counts, toward “meaningful interactions” between people.

we will predict which posts you might want to interact with your friends about, and show these posts higher in feed. These are posts that inspire back-and-forth discussion in the comments and posts that you might want to share and react to

(We will predict which posts you might want to discuss with your friends, and show those posts higher in the feed. These are posts that inspire back-and-forth discussion in the comments, and posts that you might want to share and react to.)

Likes work, but posts that generate back-and-forth in the comments rank higher — this ordering appears as concrete numbers in X’s published code, discussed below.

Instagram — Separate Rankings per Surface; in Explore and Reels, Likes Are Among the Top Predictions

Instagram is operated by the same Meta, but Instagram has no single algorithm. According to the explanation “Instagram Ranking Explained” (2023) on the official Instagram Blog, separate rankings run for each surface — Feed, Stories, Explore, and Reels. Within that, five interactions are listed as the predictions weighted most heavily in Feed, and “how likely you are to like it” is one of them.

In Feed, the five interactions we look at most closely are how likely you are to spend a few seconds on a post, comment on it, like it, share it, and tap on the profile photo.

(In Feed, the five interactions watched most closely are the probability that you will spend a few seconds on a post, comment on it, like it, share it, and tap on the profile photo.)

The side where accumulated likes surface a post is also described clearly. The same explanation lists the number of likes and how quickly they arrive as a measure of popularity, among the “information about a post” used in Feed ranking.

These are signals both about how popular a post is – think how many people have liked it and how quickly people are liking, commenting, sharing and saving a post – and more mundane information about the content itself, like when it was posted, and what location, if any, was attached to it.

(These are both signals about how popular a post is — think of how many people have liked it, and how quickly people are liking, commenting on, sharing, and saving it — and more basic information about the content itself, such as when it was posted and what location, if any, was attached to it.)

On the surfaces that expose posts beyond one’s followed accounts, the like prediction comes even more to the fore. For Explore and Reels, the same explanation names the most important predictions as follows.

The most important actions we predict in Explore include likes, saves, and shares.

(The most important actions predicted in Explore include likes, saves, and shares.)

The most important predictions we make are how likely you are to reshare a reel, watch a reel all the way through, like it, and go to the audio page

(The most important predictions made are the probability that you will reshare a reel, watch it all the way through, like it, and go to the audio page.)

The Instagram Feed system card in the Meta Transparency Center likewise lists positive actions, including likes, as material for predicting “the probability that you will enjoy a post.”

Whether you take a positive action on the post, such as liking, commenting or resharing

(Whether you take a positive action on the post, such as liking, commenting, or resharing.)

The structure — likes are material for predicting reaction probabilities, and the count and speed of accumulated likes become popularity signals — is the same as Facebook’s. The structure is no different on X, but because X’s actual code is public, we can verify it there down to concrete numbers.

X — The Help Center’s Explanation, and the Weights Written in the Published Code

The Official Explanation: Likes Are Training Data and a Recommendation Signal

“For You Home Timeline Recommendations” in the X Help Center explains the mechanism that decides the order of the “For You” timeline as follows.

𝕏 ranks the relevance of posts using a neural network that is continuously trained on your interactions (e.g. Likes, Reposts, and Replies) to give you a positive experience.

(X ranks the relevance of posts using a neural network that is continuously trained on your interactions — such as Likes, Reposts, and Replies — in order to give you a positive experience.)

The same document goes on to name “other people’s likes” among the examples of signals used for recommendations.

We recommend posts to you based on a variety of signals including, but not limited to, accounts you follow, Topics you follow, posts you like, posts liked by those in your network, and accounts followed by those in your network.

(Posts are recommended to you based on a variety of signals — including, but not limited to, the accounts you follow, the Topics you follow, the posts you like, the posts liked by those in your network, and the accounts followed by those in your network.)

This is the same structure as Facebook’s “friends’ likes.” When a post receives a like, it becomes a candidate for recommendation in the “For You” timelines of the liker’s followers. One of the main routes by which posts from outside your followed accounts appear in your timeline is this “like within your network.”

The Weights Written in the Published Code: Like 0.5, Reply 13.5

In March 2023, X published the actual code of its recommendation algorithm on GitHub. The README for “Heavy Ranker,” the model responsible for the final stage of ranking, states the score formula directly: predict the probability of each engagement (like, repost, reply, and so on), multiply each by a weight, and sum them.

score = sum_i { (weight of engagement i) * (probability of engagement i) }

(score = Σ (weight of engagement i × probability that engagement i occurs))

The actual weight values recorded in the same README (as of April 5, 2023) are as follows.

Among the positive weights, the like is in the lightest class — 1/27 of a reply, and 1/150 of “a reply that the author engages with.” Negative reactions, by contrast, carry large weights: by simple arithmetic, one predicted “show less often” or similar cancels out 148 likes.

However, reading these numbers as “to X, a reply is worth 27 times a like” would be hasty. The README states that the weights were originally set so that each engagement — which occurs at very different frequencies — would contribute roughly the same amount on average. Likes occur far more frequently than replies, so the design balanced their contributions despite the small weight. On top of that, it also states explicitly that the weights have since been adjusted periodically to optimize platform metrics, so there is no guarantee that the values at publication time are still in use today.

Note that the explanatory article posted on the engineering blog of X (then Twitter) when the algorithm was published had become a dead link as of July 25, 2026, and its content could not be verified. The official materials still available for reference are the Help Center page and the GitHub repository cited above.

When Likes Accumulate, How Does a Post Surface?

Bringing together the documents of the three platforms, the routes by which likes push up a post’s exposure can be organized into two.

The first is as input to the prediction models. Display order is decided by predictions, for each viewer, of “the probability of reacting,” and the number of likes a post has already received — and who they came from — is among the material used to compute those predictions. The very fact that likes have accumulated strengthens the prediction that the next viewer will react too.

The second is as a route that generates distribution candidates. On Facebook via “how many of your friends liked the post,” and on X via “posts liked by those in your network,” a like works to send the post into the feeds of the liker’s connections. On Instagram, in addition to the popularity signal of “how many people have liked it and how quickly,” the top predictions on Explore and Reels — the surfaces that expose posts beyond one’s followed accounts — include “how likely you are to like it.” This is the core of the mechanism by which posts spread beyond followers and friends, and from the poster’s point of view, this effect is structurally the larger one.

At the same time, none of the platforms adopts the simple mechanism of “putting the posts with the most likes on top.” Meta states explicitly that the weight of a like differs from person to person, and since 2018 Facebook has held up a policy of prioritizing posts that generate conversation. In X’s published code, the like was weighted more than an order of magnitude below the reply-type engagements.

Summary

• Seen from the platform, a “like” is — according to the official documents — input data (a signal) for predicting how users will react. It has two roles: teaching the platform the liker’s own interests, and surfacing the post to other users. The latter is stated explicitly as “how many of your friends liked the post” on Facebook, as the popularity signal “how many people have liked it and how quickly” on Instagram, and as “posts liked by those in your network” on X
• Facebook, Instagram, and X all officially explain that they decide display order not by “most likes first” but by predicted reaction probabilities for each viewer
• Therefore, “more likes make a post reach more people” is supported by the official documents — but the effect is not a direct score bonus; it works indirectly, by pushing predictions up
• Instagram has a separate ranking per surface, and on Explore and Reels — the surfaces that expose posts beyond one’s followed accounts — “how likely you are to like it” is among the most important predictions
• Facebook and X weight reactions that generate conversation (comments, replies, and their back-and-forth) more heavily than a like by itself. Meta shows this in its policy documents, and X in its published code (as of April 2023), where the like’s weight was 0.5 against 13.5 for a reply and 75.0 for a reply that the author engages with
• X’s weights were set with the intent of correcting for frequency differences, and are stated to be adjusted periodically, so the ratios of the numbers cannot be read directly as ratios of value

References

• Our approach to Facebook Feed ranking — Meta Transparency Center
• Facebook Feed AI system — Meta Transparency Center
• How AI Influences What You See on Facebook and Instagram — Meta Newsroom (2023)
• News Feed FYI: Bringing People Closer Together — Meta Newsroom (2018)
• Instagram Ranking Explained — Instagram Blog (2023)
• Instagram Feed AI system — Meta Transparency Center
• For You Home Timeline Recommendations — X Help Center
• twitter/the-algorithm-ml projects/home/recap/README.md (Heavy Ranker explanation and weights) — GitHub