Social media & AI

Deleted content moves to Recently deleted and stays 30 days, stories get a day, conversations never come back. And full deletion takes up to 90 days.

The limits are named outright, and so is the list of what doesn't work with scheduled content: product tags, sponsored posts and collabs.

Hidden comments stay in the total count and their authors never learn. The filter is on by default, and it has nothing to do with the distribution of your posts.

Tagged content isn't shared with your followers, a collaborative post is. Plus the primary creator's rights: delete, go private or block — and the post is gone.

The platform has no such term: it has two bars and an “Account Status” screen stating whether your content is eligible to be recommended. Recommendations switch off for the whole account at once.

90 days for the account, two years for a story and a reel. Word-for-word definitions of views, viewers and initial views — and a list of metrics the Help Centre doesn't describe at all.

The list closes off posts, notes, reels and stories, but the Help Centre says nothing about distribution. The closeness prediction is computed from messages instead.

All six Meta system cards read end to end: not one signal about the hour of the day. What is there instead is five windows of time and three buckets of freshness, counted for the reader.

Three predictions out of ten are computed by the card separately for the first post of the session. Plus the mechanics of the carousel, the reordering of its cards and what the platform calls informative content.

Selection in the Explore tab goes in three stages, and among the ten predictions one counts the probability that you will tap “Not interested”. Plus two numbers: five seconds and 95 percent.

A private channel isn't found by search and isn't shown in lists of similar channels. And an invite link in open access, by the FAQ's own explicit caveat, makes it equal to a public one in content disputes.

Recommendations have a system card of their own: ten predictions and thirty-one signals. One prediction is computed by a language model — that is what the document says. Plus windows from 12 hours to 84 days.

The platform calls borders, watermarks and crediting the author immaterial edits. Instead of percentages it offers a test in one phrase, and it counts the return to recommendations over 30 days.

At the start it's eligibility for recommendations and your angle that decide, not the number of videos. We break down what switches recommendations off entirely, how an angle differs from a topic, and which numbers to look at in the first weeks.

A channel has no reach machinery: a post goes to subscribers and to nobody else. From that follow different content, different numbers and three settings that decide more than any content plan.

Paid subscriptions require a professional account with 10,000 followers and an age of 18 and over, and the availability list consists of 48 countries, with no CIS country among them. Plus withdrawal thresholds and payout dates.

The help centre names nine languages that translated Reels work in, and Russian isn't among them. We go through what that means for an author from the CIS and what to do instead of waiting.

The comment button appears as a consequence of linking a discussion group — there is no separate setting for it in the platform's documents. Plus admin signatures, without which the reader doesn't know who wrote the post.

Telegram itself calls its views approximate: forwards count towards the total, and after about four days the platform forgets your view and counts you again.

The help centre describes the channel notification only for the first message. We go through how people get into a channel, what outsiders can see and why the launch happens once.

Instagram's help centre names three grounds for closing access to the music library: account type, post type and region. It also explains why an old video can lose its sound.

Instagram's help centre sets a limit of five hashtags per post, and across six prediction lists from Meta's system cards hashtags don't appear once. We go through what they actually do.

On Telegram the feed shows only what you're subscribed to. The one automatic showcase is similar channels, and they are selected by audience overlap, not by the channel description.

Reel length has two limits: 20 minutes is the technical ceiling and 3 minutes is the border past which the video stops being shown to new people. Length also picks the group you get compared against.

The reel reaches neither followers nor anyone else: in the Reels feed, follows and recommendations share one set of a hundred slots. Three insight numbers that diagnose it, and one Reels lacks.

Stories have no recommendations: they are gathered only from accounts you follow. Three places where reach is lost, five metrics, and a way to build the frame-by-frame drop-off curve by hand.

How to tell a real reach drop from normal fluctuation using Instagram's own stats, what the "Account status" section shows, and three reactions that make things worse.

The signals Instagram uses to decide whether to keep showing your reel, which of them you can see in your own stats, and why ads here treat the symptom instead of the cause.

How much to post without tanking your reach: a plan for 480 pieces of content cut lead flow by ten times. We break down why the feed selects instead of stacking.