TikTok Saves vs Shares vs Likes: What 4,265 Posts Reveal
Try it on your own video
Paste a link. Get the clips.
When we measured 4,404 TikTok posts earlier this month, we reported plays and likes and said plainly that we were holding shares, saves, duration and follower counts for a follow-up. This is that follow-up, and it did not go where we expected.
Like rate behaves the way the growth advice says it should: the further a post travels, the harder people react. It climbs from 3.29% in the least-viewed decile to 8.14% in the most-viewed.
Saves and shares do the opposite. Both peak in the middle of the reach distribution and then fall away at the top. The most-viewed posts in our corpus are saved at roughly a third of the rate of posts with a fraction of their reach.
What we measured
4,404 TikTok posts sampled by hashtag across nine business verticals. We removed 132 posts flagged as paid ads, whose reach is bought rather than earned, and 7 with no usable follower count, leaving 4,265. Every post carries plays, likes, comments, shares and saves as TikTok reported them at capture, plus duration, follower count and whether speech was present.
The rules we hold ourselves to. Medians, never averages — a single 42-million-play post would rewrite any mean. Every comparison is stratified by follower band, because most apparent content effects are really account-size effects; we only believe a result that holds in the same direction across bands. Any group under n=40 is suppressed, not published with an asterisk. And where the honest answer is "no effect", we say so — there is one of those below.
Likes rise. Saves and shares peak and fall.
Splitting the corpus into ten equal groups by play count and taking the median engagement rate inside each:
| Decile | Median plays | Like rate | Save rate | Share rate |
|---|---|---|---|---|
| 1 (lowest reach) | 283 | 3.29% | 0.21% | — |
| 2 | 1,057 | 4.98% | 0.58% | 0.15% |
| 3 | 3,085 | 4.62% | 0.86% | 0.23% |
| 4 | 8,378 | 4.55% | 0.82% | 0.22% |
| 5 | 19,600 | 5.41% | 1.34% | 0.33% |
| 6 | 42,800 | 5.24% | 1.35% | 0.39% |
| 7 | 94,700 | 6.33% | 1.17% | 0.42% |
| 8 | 231,100 | 6.86% | 1.11% | 0.40% |
| 9 | 625,400 | 7.52% | 0.69% | 0.38% |
| 10 (highest reach) | 2,800,000 | 8.14% | 0.52% | 0.27% |
Like rate roughly doubles from bottom to top. Save rate peaks at decile 6 and then falls 61% by decile 10. Share rate peaks a decile later and falls 36%.
(The lowest decile medians zero shares — at a median of 283 plays, most posts in it were shared by nobody at all. We leave the cell blank rather than print a ratio against zero.)
The mechanism we find most plausible: saving is a private act with a cost. You save something because you intend to use it. Sharing is cheaper but still requires a specific person to send it to. Liking costs nothing. As a post escapes the audience it was made for and lands in general feeds, the marginal viewer is less and less likely to be someone with a use for it — so they tap the cheap signal and move on.
Reach dilutes intent. That is not an argument against reach; it is an argument against reading save counts on a viral post as though they mean the same thing they meant on a small one.
Most of the best-shared posts are not the best-saved
If saves and shares measured the same underlying thing, the posts at the top of one would sit at the top of the other. Taking the 3,657 posts with at least 1,000 plays and comparing the top quintile by share rate against the top quintile by save rate:
Only 42.4% overlap. 57.6% of the most-shared posts are not among the most-saved. Across the whole set the rank correlation between the two rates is 0.59 — related, but nowhere near interchangeable.
They also differ in scale. The median post is saved about three times as often as it is shared (0.91% against 0.30%). If you have been treating "engagement" as one number, these are two different behaviours pointing at two different content strategies.
Utility gets saved. Entertainment gets shared.
The split is clearest across verticals. Saves per share, for posts with at least 1,000 plays:
| Vertical | n | Save rate | Share rate | Saves per share |
|---|---|---|---|---|
| SaaS / B2B | 1,625 | 1.50% | 0.35% | 5.2 |
| E-commerce | 55 | 1.37% | 0.31% | 3.6 |
| General | 616 | 0.76% | 0.24% | 3.4 |
| Fitness | 76 | 0.69% | 0.24% | 3.1 |
| Beauty | 293 | 0.62% | 0.22% | 3.0 |
| Dental | 156 | 0.49% | 0.20% | 2.1 |
| Home services | 560 | 0.42% | 0.22% | 2.0 |
| Real estate | 125 | 0.77% | 0.44% | 1.9 |
| Food | 151 | 1.22% | 1.23% | 1.2 |
SaaS content is saved 5.2 times for every share. Food is saved and shared at almost exactly the same rate. That is the difference between "I will need this later" and "look at this".
The practical version: if you make reference-shaped content, a low share count is not a failure signal and chasing shareability will push you toward the wrong edits. If you make entertainment, a high save rate is not the win it looks like.
Speech drives saves at every account size
Only one content trait we tested held its direction in all four follower bands.
| Account size | Median saves, has speech | Silent | Effect |
|---|---|---|---|
| Under 1,000 followers | 26 | 4 | 6.38x |
| 1,000 - 10,000 | 243 | 187 | 1.30x |
| 10,000 - 100,000 | 717 | 501 | 1.43x |
| 100,000+ | 2,569 | 1,947 | 1.32x |
Shares show the same pattern, also 4 out of 4 bands. Combined with the reach result in our hooks analysis, spoken audio is now the only variable in this corpus that has moved plays, shares and saves in the same direction at every account size. If you change one thing, change that.
The 31-to-60 second dead zone shows up a third time
We first found that posts of 31 to 60 seconds underperform on plays at every account size. They underperform on shares at every account size (0.17x to 0.75x). And on saves:
| Account size | 31-60s | All other lengths | Effect |
|---|---|---|---|
| Under 1,000 followers | 4 | 14 | 0.29x |
| 1,000 - 10,000 | 109 | 251 | 0.43x |
| 10,000 - 100,000 | 550 | 628 | 0.88x |
| 100,000+ | 2,366 | 2,458 | 0.96x |
Three independent metrics, twelve band-level comparisons, not one of them favouring the 31-to-60 band. We are now fairly confident this is real rather than an artefact of any single measure. The penalty shrinks as accounts grow, which is the shape a genuine effect usually takes once other advantages pile up.
One thing that does flip with scale: posts over 60 seconds are saved less by small accounts (0.60x at 1k-10k) but more by large ones (1.87x at 100k+). Long-form appears to earn saves only once an audience already trusts the account.
The finding that was not there: trending sounds
"Use a trending sound" is close to universal advice, so we tested it. Comparing posts using original audio against posts using a licensed or trending track, stratified as everything else:
- Plays: original audio higher in 2 of 4 bands
- Shares: higher in 3 of 4 bands
- Saves: higher in 2 of 4 bands
There is a real-looking 1.41x advantage for original audio in the 1,000-10,000 follower band, and it does not replicate anywhere else. That is the signature of noise, not an effect. We cannot show that sound choice moves reach in either direction.
One honest caveat on our own null: TikTok's flag marks whether the audio is the creator's own, which is not quite the same question as whether a sound is trending. A creator talking to camera is "original audio" whether or not they are riding a trend. So read this as "we could not find the effect with the field available", not as proof that the advice is wrong.
Verified accounts were a similar dead end. Only 140 posts in the corpus come from verified creators and all but six sit in the 100,000+ band, where they actually underperform unverified accounts of similar size (200,950 median plays against 270,500). With one testable band we would not build a recommendation on it, and we mention it only so the number is not mistaken for something we hid.
What to do with this
- Pick the metric before you pick the edit. 57.6% of the most-shared posts are not among the most-saved. Optimising for both at once mostly means optimising for neither.
- Put speech in the video. The only trait that moved plays, shares and saves the same way in every follower band.
- Stop landing at 31 to 60 seconds. Three metrics, twelve comparisons, zero wins. Go under 30 or commit past 60.
- Judge save rate against posts of similar reach, not against your best post. Save rate falls as reach grows, so a viral post will always look worse on this measure than the mid-tier post that preceded it.
- Do not rebuild your audio strategy on trend-chasing. We could not find the effect.
If you want the reach-side companion to this, our analysis of what actually goes viral on TikTok covers the play and like distribution across the same corpus, and the hooks study covers which openings earn the first three seconds. To see what our own scoring model looks for in a clip, there is a free viral score checker.
What this cannot tell you
It is TikTok only. We have no Instagram Reels or YouTube Shorts corpus, and the platforms weight watch time differently enough that transferring these numbers would be a guess wearing a table.
It is hashtag-sampled, which biases toward posts that already received some distribution, and it is a snapshot rather than a tracked cohort — posts captured at different ages are not strictly comparable. SaaS is 38% of the corpus by design, because that is the content our customers make, so the pooled numbers lean commercial.
Most importantly, none of this is watch time. TikTok does not expose it, and watch time is the variable most likely to explain the patterns above. We are measuring the shadow it casts, not the thing itself.
Frequently asked questions
Are saves better than shares on TikTok?
They are different objectives rather than better or worse. Across 3,657 posts with at least 1,000 plays, 57.6% of the most-shared posts were not among the most-saved, and the rank correlation between the two rates is 0.59. Saves are also about three times more common (median 0.91% against 0.30%).
Does a viral TikTok get saved more?
Proportionally, no. Save rate peaks around 1.35% in the middle of the reach distribution and falls to 0.52% in the most-viewed decile. Like rate moves the other way, from 3.29% to 8.14%.
Does using a trending sound get more views?
We could not show that it does. Original audio beat licensed or trending sound in only 2 of 4 follower bands on plays and 2 of 4 on saves — no consistent direction once account size is controlled for.
What kind of TikTok gets saved?
Content with spoken audio, in every follower band we tested, from 1.30x to 6.38x more saves than silent posts. Utility content saves far more than entertainment: SaaS posts collect 5.2 saves per share against 1.2 for food.
How big is the sample?
4,404 collected posts, 4,265 after removing paid ads and posts without a usable follower count, and 3,657 for the rate comparisons, which require at least 1,000 plays. Every group reported has at least 40 posts.