What the ranking signals actually are

Instagram has been unusually public about what its systems weigh. The gap between that and what most advice repeats is large.

Sends are the strongest signal

Sharing a post with someone in a DM is the highest-value action. It is costly for the viewer — they are spending social capital — and it predicts value better than any passive signal. Instagram has said repeatedly that sends per reach is the metric it would choose if it could only see one.

This has a direct implication for what you make: content that gives someone a reason to send it to a specific person outperforms content that is merely enjoyable. "This is literally you" beats "this is nice".

Watch time and completion

For Reels, how long people watch, and whether they finish. Completion on a short video is easier than retention on a long one, which is why very short Reels can over-perform — but they also deliver less, so the follow-through is weaker.

Rewatches count for a lot, which is why loops and details-you-missed work.

Saves

Intent to return. Strong on carousels and anything reference-like. A save is a bet that this will be useful later.

Likes and comments

Real but weak, and weaker than the industry treats them. A like is nearly free. Comments are better because they cost more, but engagement-bait comments ("comment YES") are a recognised pattern and a diminishing trick.

Follows from a post

When a viewer follows after seeing a single piece, that is a strong quality signal. It is also the metric that best predicts whether recommended reach will translate into anything.

The negative signals

Skipping quickly, hiding, "not interested", unfollowing. These do more damage than the equivalent positive signals do good. Which is the real argument against posting high-volume low-quality content: you are not just failing to gain, you are teaching the system that people avoid you.

What trips people up

• Chasing likes because they are the visible number.
• Buying engagement. Bought signals come from accounts whose behaviour predicts nothing, and pollute your audience model.
• Reading a single post's numbers rather than the trend across a format.

You have got this when

You can look at your best-performing post and say which signal carried it — and design the next one to earn the same signal deliberately.