The short answer

In the first 3 seconds, winning Meta video ads rarely explain anything. They point at something on screen, react, name a character or give an order. On our board, openers that start with a command ("Don't", "Stop", "Listen") reach the top tier 21.4% of the time and openers that start with a number 20.5%, against about 10% for the board. Openers that start with "you" sit at 3.9%.

Three layers in three seconds

The opening of a video ad happens on three channels at once: what the viewer sees, what they hear, and what they read on screen. Most viewers get the first and third. Many never get the second, because they scroll with the sound off.

We looked at the 1,752 video ads on the Ad Radar board that have a transcript, all live on October 9, 2026, all with $50k or more in estimated spend. The spoken opener is the part we can measure directly. The visual side we can read from each ad's production format tag. Overlay text isn't in our data, so that section is practice, not statistics.

Layer 1: the first words

We classified each opener by how it starts. "Reach top tier" is the share of ads in the top 10% of the board by winner score (about 10% of all ads).

How the opener startsAdsShareReach top tierMedian est. spend
A command ("Don't", "Stop", "Listen", "Give me")422.4%21.4%$253K
A number ("Four days.", "$2,300...")392.2%20.5%$318K
A reaction ("Oh", "Wait", "Okay", "What")754.3%13.3%$236K
A greeting ("Hi", "Hey", "Good morning")482.7%12.5%$309K
Pointing at the screen ("This", "Look", "Watch", "Here")30517.4%8.5%$224K
First person ("I", "My", "We")29817.0%8.1%$202K
Second person ("You", "Your", "If you")18110.3%3.9%$237K

Command and number openers are rare (about 40 ads each), so the rates are leads, not laws. But the direction is consistent with what we see elsewhere on the board: openers that act on the viewer or hand them a concrete fact beat openers that describe the viewer to themselves. "You" openers are the weakest group we measured. Most of them describe a symptom the viewer already knows they have.

WuffesMeta ad · winning
Don't wait until your pet gets to that point.
Est. spend
$4.5M
Days live
152
Format
Video, 54s
Variants
16
  • Unaware
  • Expert Explanation

A command with a vague "that point" the picture has to fill in. Sixteen variants on one opener is unusual even on a board of winners.

Layer 2: the first frame

The production format tells you what the viewer sees at frame one: a face talking to camera, footage under a voiceover, the product in use, an animation. On the 610 ads with a production format tag:

First frame (production format)AdsReach top tierMedian est. spend
UGC selfie (a person talking to their phone camera)7812.8%$592K
Footage + voiceover17714.1%$345K
Founder or expert on camera14512.4%$296K
Product demo978.2%$301K
AI-generated772.6%$229K
Creator skit200%$371K

UGC selfie ads have by far the highest median estimated spend ($592K). Footage with voiceover has the best top-tier rate. AI-generated openers reach the board but rarely the top tier (2 of 77). Creator skits, with only 20 ads, haven't produced a top-tier ad at all, possibly because a skit's first seconds are a setup that needs sound and context.

The pattern behind the numbers: a human face or real footage at frame one, with the speaker already mid-thought. A greeting works when the face is the hook.

OmnipodMeta ad · winning
Hi, I'm Logan.
Est. spend
$3.7M
Days live
77
Format
Video, 59s
Variants
10
  • Solution-Aware

A child on camera introducing himself. The greeting adds nothing in words, so the frame does all the work. Line two gives his age and holds up the product.

Layer 3: the words on screen

We don't store overlay text, so this section is working practice rather than board statistics.

  • Don't transcribe the voice. If the spoken line is a reaction ("Oh, no way!"), the overlay is where the claim or the product name goes. If the spoken line is a claim, use the overlay for the audience ("Women over 50") or the number.
  • One line, top or center. Two short lines at most, placed clear of the interface at the bottom of Reels.
  • On screen from frame one. Text that fades in at second two misses the viewers who decide at second one.

Text overlay hooks covers sound-off openers in more detail.

What the strongest openings combine

The ads that stay live longest stack the three layers so each one does a different job. A reaction line plus a visible product plus a text claim. A number in the voice plus a person on camera plus the audience in text.

Based SuppliesMeta ad · winning
Yes, we make skincare from beef fat.
Est. spend
$1.4M
Days live
250
Format
Video, 86s
  • Solution-Aware
  • Footage + Voiceover
  • Story Narration

Opens by answering an objection nobody has voiced yet. "Yes" implies the viewer is already surprised, and the image of the product makes the surprise concrete.

Cosmody shows the same idea with dialogue: "Okay, come see." (est. $846K, 417 days, UGC selfie). The line is an instruction to the person on camera, and by extension to the viewer, to look at something just out of frame. The second line is "Have you ever seen them?" Neither line explains what the product is. The frame and the curiosity do that.

The brand name can wait

Only 7% of winning video openers mention the brand in the first sentence (120 of 1,752). Those that do reach the top tier at about the board rate (10.8%), so naming the brand doesn't hurt. It just isn't how most winners spend their first three seconds. The brand typically arrives in the second or third line, after the viewer has a reason to care. Read hold rate for where the brand mention tends to land in long-running scripts.

A three-second checklist for your next edit

  1. Frame one shows the subject. Product, problem or person. No logo cards, no black frames, no slow fade-in.
  2. The first spoken words give an order, a number, a reaction or a name. Avoid opening with "you" or "if you" describing a symptom.
  3. Four words or fewer if the visual is strong. Openers this short reach the top tier 11.9% of the time on our board.
  4. The overlay adds a second piece of information. Not the voiceover transcribed.
  5. Start mid-thought. Cut "Hey guys", "So", "Today I want to" unless the greeting itself is the hook, as with a child or a known face.
  6. Make three versions of the first three seconds. Keep the body identical and measure each with hook rate.

The 60 real openers and the hook formulas give you material for step six. To see the first frame and the first line of competitor ads side by side, filter the Ad Radar board by niche and production format.

Figures marked as estimated spend come from Ad Radar's model of engagement on public Meta Ad Library ads. They are estimates, labeled as such, and are best used to rank ads against each other.

Questions

Why do the first 3 seconds of a video ad matter?

Meta reports 3-second video plays as a standard metric, and it is the window in which a viewer decides whether to keep scrolling. Hook rate (3-second plays divided by impressions) is built on it. If the first three seconds fail, nothing later in the ad gets seen.

What should be in the first 3 seconds of a video ad?

One clear visual subject (the product, the problem or a person), a first spoken line that makes the viewer look at that subject, and on-screen text for people watching with the sound off. On our board, openers that give a command or start with a number do best.

Should the brand name or logo appear in the first 3 seconds?

Not as the hook. Only 7% of winning video openers on our board mention the brand in the first sentence. Most name it after a short setup, which keeps the opening about the viewer's problem or curiosity rather than about the company.

How many words fit in the first 3 seconds?

Fewer than most scripts use. The median first sentence on our board is 11 words, which runs past three seconds at a normal pace. Openers of 4 words or fewer are 11% of winners and reach the top tier 11.9% of the time.

Keep reading

See the ads that are already winning.

Membership is $49 every 4 weeks at the launch price. Cancel anytime.