A Reel with 50 likes and 1,000 views can now out-rank one with 100 likes and 10,000 views. That's not a typo, and it's not a bug. It's the entire point of how Instagram and TikTok rank content today. If your team is still watching the like counter and wondering why reach has flattened, you're reading the wrong number.
The like was never the goal, it was a proxy
Likes were always a stand-in for something the platforms actually wanted to measure: did this content matter to someone. For a long time, a like was the easiest available signal, so the algorithm leaned on it. But a like is also the cheapest possible action a person can take. It costs nothing, it can be tapped without watching past the first two seconds, and it tells the platform almost nothing about whether the content actually landed.
Once platforms had the machine learning infrastructure to track richer behavioral signals, likes stopped being useful as a primary ranking input. What replaced them isn't one single metric. It's a cluster of higher-effort, higher-intent actions that are much harder to fake and much more predictive of whether a piece of content deserves a bigger audience. For a breakdown on launch mechanics, check out our guide on the first 60 minutes after posting.
What actually ranks content now
Instagram has been the most explicit about this shift. Head of Instagram Adam Mosseri has publicly confirmed that the platform ranks content on three signals: watch time, sends per reach (how often people DM your content to a friend), and likes per reach. Sends carry roughly 3-5x more weight than likes when Instagram decides whether to push content to non-followers, and that gap exists for a specific reason: a DM share is a person personally vouching for something to one other human being. A like is a passive reflex. Instagram's own data backs the scale of this — reportedly hundreds of thousands of Reels get sent via DM every minute across the platform.
Saves tell a similar story. Industry tracking puts saves at roughly 3x the ranking weight of a like, because saving something signals the person expects to come back to it. That's a stronger vote of long-term value than a like ever was.
TikTok runs on a comparable logic, even though its cold-start mechanic looks different on the surface. A new video is shown to a small test pool, usually somewhere in the 200-500 viewer range, and the platform watches completion rate alongside share, save, and comment-depth signals within the first 30-90 minutes. If a video clears roughly 35% completion and at least one engagement signal fires above about 1.5% of that test pool, it graduates to a much larger expansion pool. DM shares on TikTok are weighted at around 3x a like, and the completion bar for genuinely viral distribution has climbed to roughly 70%, up from about 50% in 2024.
YouTube never counted likes as a dominant signal in the first place, but it's worth including here because it shows where the other platforms are heading. YouTube's own Director of Growth and Discovery has said the platform weights watch time by satisfaction, not raw minutes, and comments plus likes are now described internally as comparatively weak ranking signals, correlated with good videos but not the cause of their distribution. What actually gates recommendation on YouTube is click-through rate relative to other videos shown in the same slot, average view duration as a percentage of the video, and increasingly, whether the viewer stuck around on YouTube after your video ended.
The pattern across all three platforms is the same. Passive metrics are out. Behaviors that require the viewer to do something costly — watch to the end, send it to a specific person, save it for later — are in.
| Signal | TikTok | YouTube | |
|---|---|---|---|
| Primary ranking driver | Watch time | Completion rate & watch time | CTR + average view duration |
| Like weighting | Low, de-prioritized | Low, minor signal | Low, correlated not causal |
| Strongest reach signal | Sends per reach (DM shares, 3-5x a like) | Shares/saves clearing test pool | Session contribution |
| Save behavior | ~3x a like in weight | Strong intent signal | Not a primary factor |
| Cold-start mechanic | Trials feature tests with non-followers | 200-500 viewer test pool | Small-audience test |
What this actually means for what you make
Here's where a lot of advice oversells itself. Nobody can hand you a formula that guarantees a DM share. What you can do is design content that makes the specific behaviors the algorithm is watching for more likely to happen naturally.
For sends and shares, the content needs a reason to exist in someone's DM thread with a specific other person in mind. That's usually a strong opinion, a relatable and slightly uncomfortable truth, or a piece of practical value specific enough that someone thinks "my coworker needs this," not "this was fine." Generic, safe content rarely gets sent anywhere, because there's no specific person it makes you think of.
For saves, the content needs to function as a reference the viewer plans to use later: a framework, a checklist, a before/after, a step-by-step breakdown. If someone can absorb everything in a three-second glance, there's nothing to save.
For completion and watch time, the fix is almost always structural, not cosmetic. Front-load the actual value instead of building up to it. Cut anything that doesn't earn its place. This is where a professional edit genuinely outperforms a rushed one, not because of filters or transitions, but because pacing decisions — where to cut, where to hold, where to add a visual beat — are the difference between someone finishing a video and someone scrolling past it at second four.
Where AI genuinely helps, and where it doesn't
There's a lot of noise right now claiming AI can auto-generate content that hits these signals. Be skeptical of that framing. AI tools like Veo 3.1, Kling, and Runway Gen-4 can generate raw footage or visual assets fast, and tools like ElevenLabs can produce clean narration in minutes instead of a studio booking. That's a genuine speed advantage, and it's real.
What AI cannot currently do is judge whether a specific cut will make a specific audience want to send it to a specific friend. That judgment call — what to trim, where the emotional beat lands, whether the hook actually earns the next three seconds — still comes from a human editor who understands the platform and the audience. The studios and creators seeing real gains in 2026 aren't the ones replacing editorial judgment with AI. They're the ones using AI to compress production time so the human editor can spend more of their time on the part that actually moves these ranking signals: pacing, hook strength, and whether the finished piece earns a share.
A quick framework for auditing your own content
Before your next post goes out, run it through four questions instead of just previewing the caption:
- Would I personally send this to one specific person I know? If you can't name that person, the algorithm's audience probably can't either.
- Does the first three seconds justify staying past three seconds? This is the single most tested window on every platform in this article.
- Is there anything in here worth saving for later? A framework, a number, a step-by-step. If the entire value is delivered in one glance, there's no reason to save it.
- Are we still measuring success by likes in our own reporting? If yes, that's a reporting problem before it's a content problem.
The takeaway
Likes didn't disappear from the app, they disappeared from the math. What replaced them — sends, saves, completion rate, and watch time — all share one property: they cost the viewer something, even if it's small, and that cost is exactly what makes them trustworthy signals. A tap is free. A DM to a specific friend, a save for later, a full watch-through — those all require the viewer to actually mean it.
Chasing the like counter in 2026 means optimizing for a number the platforms have already stopped weighting. The teams pulling ahead right now are the ones who've rebuilt their content process, and their reporting, around what's actually earning distribution.
If your current content isn't converting reach into shares and saves, that's usually a pacing and hook problem more than a topic problem, and it's exactly the kind of thing worth a second set of eyes on before you burn another month guessing. VizEdits' Video Editing team builds every cut around retention curves and share triggers, not just a polished timeline, and our Social Media Management service can help rebuild the reporting layer so you're tracking the signals that actually matter. Get in touch for a free consultation if you want a read on where your current content is leaking reach.