Around 2019, Instagram ran the hidden-likes experiment in Canada and slowly expanded it through Brazil, Ireland, Italy, Japan, Australia and New Zealand. The argument was mental health. The side effect, which Adam Mosseri later admitted in interviews, was that engagement on those accounts didn't really change in any dramatic way. Likes kept happening. People just couldn't compare them publicly.
That tells you something useful: the public number on a post is a social currency question, but the underlying like event is still one of the cleanest behavioral signals the platform has. Two different problems, often conflated.
What a like actually does inside the ranking system
Instagram's feed ranking, as Mosseri has described in his weekly Q&As and the 2023 transparency post, leans on a handful of predicted actions per post: probability you'll spend time on it, probability you'll comment, probability you'll like it, probability you'll share it to a friend in DMs, and probability you'll tap the profile. Each carries different weight depending on surface. Reels weights watch time and sends. Feed weights time spent and likes. Explore weights saves and shares.
So a like isn't decorative. It's one of five or six predicted-engagement variables the model uses to decide whether your next post deserves wider distribution. The interesting part is the order of operations. The platform doesn't wait for likes to come in and then push the post. It predicts whether a given viewer is likely to like it based on their past behavior, then ranks accordingly. Your historical like-rate trains that prediction.
This is why a sudden cold-start post often dies. The model has no recent data suggesting your followers want to engage, so it tests on a small slice, sees weak signal, and stops.
The 30-minute window everyone misreads
Creators repeat the line that the first hour decides a post's fate. The data is messier than that. Looking at posts across a niche like fitness coaching, where I've seen accounts in the 40k–120k range share analytics openly, the velocity that matters most is roughly the like-to-impression ratio in the first 30 to 90 minutes of organic distribution.
A post that gets 400 likes from 8,000 impressions in the first hour will usually keep being served. A post that gets 400 likes from 30,000 impressions will get throttled, even though the absolute number is identical. Ratio beats volume.
This is why timing advice that focuses on "post when your audience is online" is half-right. You're not chasing the audience being awake. You're chasing a moment when enough of your engaged followers can interact before the post is shown to a wider, colder pool.
Why the public like count still matters socially
Even though likes feed the algorithm regardless of visibility, the visible number does work on humans. There's a study from MIT's Media Lab on social proof in feed environments showing that posts visibly above a threshold (the study used ~30 likes for a mid-size network) saw a measurable bump in subsequent engagement from new viewers. The effect tapered above a few hundred.
In other words: going from 4 likes to 40 likes changes how the next viewer reads the post. Going from 4,000 to 40,000 mostly doesn't, unless the viewer is judging whether to follow.
This is the gap creators try to close in the first hours. Some do it by notifying their close-friends list. Some pin the post in a community Discord or Telegram group. Some send the link to three friends in DMs because shares-to-DM is also a heavily weighted signal. A few experiment with services that offer free Instagram likes on new posts to push past that social-proof threshold before the algorithmic test window closes, which is a tactic worth understanding even if you don't use it, because it tells you what the threshold actually feels like in practice.
The follower-to-like ratio creators get wrong
A common benchmark floating around is the 1–3% like rate as healthy. That number is from a 2019 Hootsuite report and it has aged badly. Reels changed everything because they pull in non-follower views at a much higher rate. An account with 50,000 followers can easily get 200,000 impressions on a single Reel from non-followers. The like rate on that Reel will look catastrophic against follower count, but the like rate against impressions might be perfectly healthy.
The more useful number is likes per reached account, broken down by content format. For static feed posts in 2024, anything above 4% reached-to-liked is strong. For Reels, 1–2% is normal because the reach is so much wider and colder. Carousels sit in between, usually 3–6% if the first slide earns the swipe.
If you're tracking this for your own account, the data is in the Insights panel under each post. Tap "View Insights," then look at Reach versus Likes, not Follower Count versus Likes. The follower number is a sunk cost. The reach number is what the algorithm gave you this time.
What to actually do with this
A few practical moves come out of all this:
Stop posting when analytics says your audience is online, and start posting 30 minutes before. You want the wave of engaged-follower likes to land during the cold test window.
Treat the first 10 comments as more valuable than the next 100 likes. Comment weight in Reels ranking, in particular, is higher per unit than like weight, because comments are harder to fake and require more time-spent.
If a post underperforms in the first hour, don't delete it. Deletion does nothing for the algorithm and forfeits the long-tail reach that some posts pick up days later through Explore or hashtag pages.
Look at your worst-performing post from the last month. Then look at your best. The difference is rarely the caption or the hook. It's usually that the best one happened to catch a moment when your engaged followers were ready, and the worst one didn't. That's not luck you can fully control, but it's the variable worth optimizing for.
Top comments (0)