Captions on UGC ads are the hook

On a UGC ad the first three words on screen are the ad. The face, the kitchen, the product in hand — those are the plate. The mute scroller reads type before they decide to stay. Caption style is part of the creative test, not a post-process you run after the winner is picked.

The hook is type

UGC ads are watched muted in a feed that already looks like UGC. Native-feeling picture is the point. Native-feeling picture with no words is a silent mime. Kapwing’s 2026 round-up of UGC benchmarks, with Billo, archive.com and creator.co in the same research set, is the usual stack: on the order of 4× CTR and about 50% lower CPC versus non-UGC, 86% of brands saying UGC lifts performance, 92% of people trusting peers over brands. None of those figures is a caption study. They are why the format is in the account. The caption is how a muted impression of that format still has a claim.

The first three words have to be true, finishable, and in a band the chrome does not own. If those three words are a wrong SKU, a shouted paragraph, or sitting under a CTA, the rest of the creative test is noise. You are testing a miss.

That is why you caption before you scale a cut, not after. Burn-in is pixels. A “we’ll add captions in the next version” note on a winning silent ad is a different ad. The winner was mute-illegible. The captioned follow-up is a new test.

AI UGC does not change the hook. A generated creator still needs type the mute viewer can finish. Photoreal generates that run as YouTube Shorts also pick up an on-video AI overlay as of the 27 May 2026 YouTube Blog post. Keep the three words out of that badge as well as out of the bottom chrome.

Speech captions follow the mouth. If the first three words are a line the talent did not say — a price, a “stop scrolling,” an offer — that is an overlay, and it is still the hook. Do not ask ASR to invent it. Do not hide it in a caption track that starts three seconds late.

Style is a variant, not a polish pass

Teams treat captions as finishing: pick Glass, burn, ship. On UGC ads the look is a creative lever.

Test at least two drawings of the same edited transcript:

• BASIC two-line block (simple, plain, corpo — 21 looks at 1×). Quiet. Reviewable. Often the paid control.
• DYNAMIC word-timed (glass, whisper, fusion — 9 looks at 2×) or Snap/Reveal from caption styles. Hook performance. Only if the mute pass still finishes.

Do not test ten fonts on a broken transcript. Fix names, prices, SKUs first. Then change one look at a time. Karaoke versus block is a real fork. Snap Gold versus Snap Cyan is not a strategy; it is an accent.

Paid versus organic is a second axis. The paid file usually wants the BASIC control plus a legal overlay. The organic file can carry the DYNAMIC variant. Same words. Two burns. Score them separately. A Glass organic winner pasted into the ad account without the disclosure is not a caption test. It is a compliance miss.

CPS is the kill switch. If the talent talks a 90-character clause, you are not testing style. You are testing whether anyone can read. Split the cue. Sentence case. No emoji on the offer. Then test look.

Contrast is a kill switch too. White-on-window kitchen UGC is the house failure. Outline or a backdrop. If the plate is hostile, the “native” look that drops the stroke is a fail with better vibes.

The first three words

Write them on purpose.

• They should be speech the mouth actually produced, or an overlay you authored as the claim. Not both saying the same sentence.
• They should include the reason to stay: the problem, the product, the number. “Um so I” is not a hook. Trim until the type starts on the claim.
• They should survive a 1-second glance at phone size, in the safe hole, with paid chrome if this is the paid file.

That is a cut problem as much as a caption problem. Captioning cannot rescue a cold open that spends two seconds on a hallway. Trim, then burn.

Speaker IDs and SDH brackets almost never belong in those three words. [music] is not a hook unless the ad is the music gag. A name super is not a hook unless identity is the product. Keep the lower third for speech. Hold the card high or not at all on the opening.

Language code matches the spoken language. 165 codes is a listen list. A US ad with a Spanish line in the middle is two ranges or an authored mixed cue. ASR will not code-switch the SKU for you.

Credits: BASIC is 1×, DYNAMIC is 2×, on the same transcription work. Testing four DYNAMIC looks on a 30-second ad is four 2× renders. Preview on the first seconds. Kill losers on the preview. What captions cost across a month of clips is what happens when Glass becomes the default instead of a variant.

Test, then lock a house

Once a look wins for a campaign, pin it. UGC is supposed to feel native, not like a caption-preset carousel. One BASIC house style for paid, one DYNAMIC house style for organic, same transcript pipeline, is a system. Random Fusion on Tuesday because last week’s Reel felt quiet is how the brand face disappears.

Pipeline:

1. Cut the UGC. Hook on the claim.
2. Clean the transcript. Prices match the landing page.
3. Burn BASIC control and, if the line is short enough, one word-timed challenger.
4. Place legal on the paid burn only.
5. Mute phone pass. First three words. Chrome. Contrast.
6. Score. Lock the winner for the rest of the set.

The AI caption generator and add captions to a video are the burn. The UGC video generator is the plate when the plate is generated. Neither one is a reason to caption after you scale.

Captions on UGC ads are the hook. Style is a test. The first three words are the brief.

Test the opening type in Versely.

FAQ

Should I pick a caption winner after the ad wins?

No. A silent winner is a different ad from a captioned one. Caption before you scale. Style is a variant in the test, not polish on the winner.

Is DYNAMIC always the UGC look?

No. It is the challenger when the spoken hook is short enough to read. BASIC two-line is the paid control and often the organic winner on dense offers. Score both.

Can the first three words be an overlay instead of speech captions?

Yes, if nobody said them. Price, “stop scrolling,” the offer — authored overlay. Speech captions still cover what the mouth said after that. Do not duplicate the sentence in both.