Magik Install free

How to make a product video that gets watched

The three-second rule. Ninety per cent of recall in the first six seconds. The optimal length of twenty-one to thirty-four seconds. We went looking for the source of each and couldn't find one. What the platforms do publish is a very short list — and a usable creative brief falls straight out of it.

Published 2026-09-11 · 10 min read

The short answer

If you only read one paragraph

Instagram states that Reels ranking is driven by predictions about whether you will "reshare a reel, watch a reel all the way through, like it, and go to the audio page." That's four things — and it's the closest thing to a published creative brief either platform has given anyone. TikTok's version is compatible: finishing a longer video is "a strong indicator of interest" and outweighs weak ones.

So: make it finishable, make it worth sending, and choose the sound deliberately — because one of the four things Instagram says it predicts is whether you'll tap through to the audio, which makes music a ranking decision rather than decoration. Almost nothing else about "how to make a good video" is documented by anybody.

Below: what's documented versus what merely circulates, the four predictions turned into four design decisions, the only three things Instagram says it actively de-prioritises, and how to judge your own videos without a benchmark.

Documented, versus merely repeated

This distinction is the whole reason to read this rather than the hundredth listicle. The left column below comes from pages the platforms publish. The right column is what everyone says, and we could not find a first-party source for any of it.

What the platforms document versus what circulates unsourced Two columns. Documented: Instagram's four Reels ranking predictions — reshare, watch all the way through, like, go to the audio page; TikTok weighting finishing a longer video as a strong indicator; TikTok naming captions, sounds and hashtags as video information; and Instagram de-prioritising low-resolution, watermarked or already-posted reels. Unsourced: the three-second rule, ninety per cent of recall in six seconds, an optimal length of twenty-one to thirty-four seconds, and a best number of on-screen words. DOCUMENTED Reels ranking predicts four things: reshare · watch through · like · tap audio Finishing a longer video = "a strong indicator of interest" (TikTok) Video information = captions, sounds, hashtags (TikTok) Low-res / watermarked / reposted: less visible NO SOURCE WE COULD FIND "The 3-second rule" "90% of recall in 6 seconds" "Optimal length: 21–34 seconds" "Ideal number of on-screen words" Attributed loosely to the platforms, never linked. The right column isn't necessarily wrong. It's unverifiable, which is a different problem and a reason not to build a process on it.
Worth saying plainly: the unsourced rules may well be directionally fine. Front-loading a video is sensible whether or not "three seconds" is a real threshold. The objection is to treating invented precision as a specification.

Four predictions, four decisions

Instagram's four are unusually actionable if you read them as a brief rather than as metrics. Each one implies something specific about how the video should be built.

Instagram's four Reels ranking predictions turned into design decisions Four rows. Will you watch it all the way through implies cutting until nothing is skippable and putting the payoff before the end. Will you reshare it implies giving the video a reason to be sent, since product beauty alone is rarely sendable. Will you like it implies basic craft and taste. Will you go to the audio page implies choosing the sound deliberately, because audio is one of the four things being predicted. What Instagram says it predicts → what that asks of the video "Watch it all the way through" the heavily weighted one Cut until nothing is skippable. Payoff before the end, not at it. "Reshare a reel" the hardest to earn Give it a reason to be sent. A nice product shot is rarely sendable. "Like it" the lowest bar Ordinary craft. Lighting, framing, not looking careless. "Go to the audio page" the one nobody mentions The sound is a ranking decision, not decoration. Choose it on purpose.
The fourth row is the one worth stealing. Instagram is telling you the audio is part of what it's evaluating, and most ecommerce video treats music as something you drop on at the end because silence feels wrong.

TikTok corroborates the audio point from the other direction: its three ranking signal categories include "video information," which it defines as "captions, sounds, and hashtags." Sound is named in both platforms' published accounts of what they use. That's a stronger evidence base than almost any other creative advice you'll be given.

One catch if you're a store

A TikTok Business account restricts the in-app Add Sound picker to TikTok's Commercial Music Library rather than the general catalogue of trending tracks — which is the trade you made for a clickable link in your bio. So "choose the sound deliberately" means choosing well from a smaller set, and it's a real cost of that account type rather than a detail.

The only documented negatives

Instagram publishes three specific things it works to make less visible, and because documented negatives are so much rarer than confident positives, they deserve more of your attention than any hook advice.

Instagram states it makes certain reels less visible for reasons including these. The fixes are ours.
What Instagram namesHow a store trips itFix
Watermarked reels Saving a video out of TikTok and re-uploading it to Reels with the logo bouncing in the corner Export clean from your editor; upload natively to each platform
Low-resolution reels Footage that's been through a download-and-reupload cycle, or exported at a low bitrate Upload the original export, in-app, over wifi
Already posted on Instagram Recycling the same clip across a Reel, a Feed post and a Story, or reposting last month's video unchanged Re-cut: new opening, new on-screen text, same footage

There's a fourth, upstream of the video itself: on the API side Meta states that JPEG is the only image format supported and that filters aren't supported. So a catalogue full of PNG or WebP product images will silently fail to publish through any posting app unless it converts them — covered in full here.

A brief you can actually shoot

Everything from here is our judgement derived from the documented signals above, not something a platform published. Treating it as anything more than reasoning would be doing the thing this article objects to.

  1. Show the product in the first frame. Not a logo animation, not a title card, not a slow pan across an empty surface. The documented signal is completion, and you can't complete what you've already scrolled past.
  2. Pick one thing the video is about. "Does linen actually stop creasing" is a video. "New collection" is a caption pretending to be one. A single answerable question gives people a reason to stay to the end, which is the whole game.
  3. Cut it until nothing is skippable. Not to a target length — neither platform publishes one — but until every second is doing work. If there's nothing left to say after twelve seconds, stop at twelve.
  4. Choose the sound before you edit, not after. It's one of four things Instagram says it predicts. Pick it deliberately from what's available to your account type, and cut to it.
  5. Put the key words on screen and say them aloud. Both platforms read on-screen text and audio as evidence of what a video is about, and it makes the video work with the sound off.
  6. Ask whether anyone would send it. This is the test most product videos fail. If the honest answer is no, you've built something people might watch but won't spread — which caps it.

And how to tell whether it worked

Against your own average, not a benchmark. No trustworthy public benchmark exists for completion, sends or saves, and the figures in circulation come from tool vendors without disclosed samples. Because outcomes are very uneven, you're looking for a video that clearly beat your previous ones — and you need enough videos for that comparison to mean anything.

When one does clearly win, that's not just a good day — it's tested creative, and both platforms sell you a way to put money behind exactly that post. The handoff is here.

Where this leaves the tooling question

Honest placement: nothing on this page is solved by automation. The documented signals are all about whether a specific video holds a specific person's attention, and no scheduler improves that.

What Magik does is produce creator-style videos from your Shopify catalogue at a rate you couldn't hand-film, published from your own source file so the watermark and low-resolution problems above don't arise. That helps with volume and with two of the documented negatives. It does not decide what your video should be about, and it can't make a dull idea finishable.

Which is roughly the honest summary of this whole category: tools raise the number of attempts, craft raises the hit rate, and the two multiply. If you're comparing options, the comparison page covers Outfy, Xyla AI, IDEQO, Minta and Buffer alongside us. If you're deciding what to film rather than how to publish it, the content ideas page is the more useful one.

Frequently asked

Is the 3-second rule real?

We couldn't find a first-party source for it, or for the "90% of recall in six seconds" figure usually quoted with it. Both circulate through agency blogs attributed loosely to TikTok for Business. What is documented: Instagram predicts whether you'll watch a reel all the way through, reshare it, like it and tap the audio; TikTok calls finishing a longer video a strong indicator of interest. Front-loading matters because completion matters — not because three is a threshold.

How long should a product video be?

Short enough that people finish it. Neither platform publishes an optimal length and the ranges in circulation have no traceable source. Completion is the documented signal, which argues for cutting until nothing is skippable rather than padding to a recommended duration.

Does the sound actually matter?

Yes, and this one is documented rather than inferred. Instagram names "go to the audio page" as one of four Reels ranking predictions, and TikTok names sounds within its "video information" signal category. Caveat for stores: a TikTok Business account limits the in-app Add Sound picker to the Commercial Music Library.

What makes a video less visible?

Instagram names three: low-resolution reels, watermarked reels, and reels already posted on Instagram. These are the only documented negatives we're aware of, which makes them worth more attention than the undocumented positives. Export clean, upload at full quality, re-cut rather than repost.

Do I have to be on camera?

No — neither platform's published ranking documentation mentions faces. The named signals are behavioural: completion, resharing, liking, tapping the audio. Hands using a product can generate all four. A person on camera helps hold attention through a spoken explanation, which is a craft advantage rather than a ranking one.

Why do photos get no reach?

Because three of Instagram's four predictions are about watching, and a still generates almost none of it — nothing to complete, no attention held, no audio page. Photos aren't penalised; they simply can't produce the signals that earn recommendation reach. They still work on the surfaces served to existing followers.

How do I know if a video worked?

Compare completion, sends and saves against your own previous videos, not a published benchmark — none of the public ones disclose a sample. Outcomes are uneven enough that you're looking for something that clearly beat your own average.

What should I do with a video that wins?

Put money behind that specific post. TikTok's Spark Ads and Meta's "Use existing post" both exist to promote an organic post you've already tested — which makes your feed a library of proven ad creative rather than just a feed.

Sources

Unusually for an article about creative, almost everything asserted as fact here comes from two pages. Everything else is labelled as our reasoning.

  1. Instagram — Instagram Ranking Explained, 31 May 2023. Source of the four Reels ranking predictions — "reshare a reel, watch a reel all the way through, like it, and go to the audio page" — and of the statement that it works to make certain reels less visible "such as low-resolution or watermarked reels… or reels that have already been posted on Instagram".
  2. TikTok Newsroom — How TikTok recommends videos #ForYou. Source of the three ranking signal categories, the definition of video information as "captions, sounds, and hashtags", and the weighting of "a strong indicator of interest, such as whether a user finishes watching a longer video from beginning to end".
  3. Meta for Developers — Instagram Platform, Content Publishing. Source of "JPEG is the only image format supported" and that filters are not supported for API-published content.
  4. Deliberately not cited, because we could not source them: the three-second rule; "90% of ad recall happens in the first six seconds"; an optimal video length of 21–34 seconds; and prescribed counts of on-screen words. Each is widely attributed to TikTok for Business, and we could not reach a first-party page stating any of them — TikTok's creative guidance pages did not resolve for us. They may be accurate. They are not verifiable, and this article does not build on them.
  5. The six-point shooting brief and the fixes in the de-prioritisation table are our judgement derived from the documented signals, and are labelled as such in the text rather than presented as platform guidance.

Keep reading

Why do my videos get no views?

Before blaming the craft — the diagnostic that separates a content problem from an eligibility one.

Content ideas beyond product posts

What the video should be about, once you know how it should be built.

Organic or ads?

What to do with the one video that clearly beat your average.

How long before social works?

Why you need a run of videos before "my content isn't working" is a supportable conclusion.