Social Performance

What it measures

Every platform reports the same broad family of signals. They answer three questions: how many people saw the work, how they responded, and what they did next.

Reach

The number of unique accounts that saw a post. One person counts once, however many times they saw it.

Impressions

The total number of times a post was shown, repeat views included. Impressions are always equal to or greater than reach.

Likes

The lightest signal of approval. Cheap to give, easy to read, and weak on their own.

Comments

A deeper signal. A comment costs effort and often invites a reply, which platforms read as conversation.

Shares

The strongest signal. A share is a person spending their own reputation to put your work in front of their audience.

Saves

A signal of intent. A save says the content is worth returning to, which many platforms weight heavily.

Follows

The conversion of a single good post into a standing relationship.

Watch time

For video, how long people stay and whether they finish. Often the single most important ranking input.

Not all signals are equal

A like and a share are not worth the same. The signals sit on a ladder, ordered by how much they cost the person giving them and how much they tell you in return. A like is nearly free, so it means the least. A comment costs a sentence. A save is a quiet promise to come back. A share is the strongest signal of all: someone spending their own standing to put your work in front of the people who trust them.

Reach measures how far a post travelled; engagement measures how much the people it reached actually cared, and the two often move apart. A post can reach thousands and move no one, or reach a few hundred and move all of them. The numbers that flatter are rarely the ones that compound. A viral post with no saves, shares, or follows has rented attention it cannot keep. Native optimises for the signals that build something: saves, shares, follows, and viewers who return.

What good looks like

There is no single number that means good. Engagement rates fall as an account grows, so a village bakery with two thousand followers will usually out-engage a national brand with two million, and both can be perfectly healthy. Rates also shift by platform, by format, and by industry: a saved recipe, a shared property listing, and a liked landscape photo are not the same kind of event, and they do not compare.

This is why the only benchmark Native trusts is your own baseline. The question is never whether you beat a global average measured on someone else’s audience. It is whether this month beats last month, on your channel, with the people who actually follow you.

The variety paradox

The obvious way to use performance data is to make more of whatever did best. It is also a trap. In Native’s own production data, the theme a brand’s system generated most was accepted roughly half as often as the themes it starved. The more the system leaned on a proven winner, the faster the audience tired of it.

Generation share and acceptance moved in opposite directions. A model that optimises acceptance alone simply feeds a brand more of what it already liked, until the liking wears off, and audiences tire of repetition long before that model would notice. So Native holds a deliberate variety target and watches its own output for narrowing. It protects the range on purpose, because the data is clear that pure preference-following defeats itself.

How Native uses it

After a post is published, Native retrieves its reach, impressions, and engagement on each platform over time, and feeds those numbers back into what it generates next. The curator who approves a post is only a proxy for quality. The channel is the final judge.

Publish, measure, learn, generate, and publish again: the loop is the product. Every post is both an output and a small experiment, and the next one is a little better aimed than the last.

The research

Audiences judge content against a realistic alternative, not an ideal one (Kirk and Givi, 2025). For most small businesses the honest comparison is silence, not a commissioned campaign. Feeds also reward on a variable schedule, which is what makes them compulsive to check and unforgiving of sameness (Ferster and Skinner, 1957). And when many brands lean on the same models, their output converges (Doshi, Hauser and Mollick, 2024). Protecting variety is the defence against sounding like everyone else.