The five-part structure of every street poll that works
Street-interview ads look improvised. The good ones are not. Underneath almost every one that performs is the same five-part structure, and once you can see it you can write to it deliberately instead of hoping a shoot produces it by accident.
The five parts are hook, reaction, turn, reveal and close. Here is what each one has to do, how long it gets, and how it fails.
1. The hook (0–2 seconds)
The hook is the question, asked. Not an introduction, not a title card, not the interviewer explaining what they are doing. The first thing a viewer hears should be the question itself, mid-exchange, as though they arrived a second late.
The question has to have a trap in it. The respondent should be able to answer honestly and be slightly embarrassed by the honest answer, or answer confidently and be wrong. "How many hours a day do you think you're on your phone?" qualifies on both counts. "What do you look for in a supplement?" qualifies on neither, which is why that version of the ad dies at two seconds.
How it fails: starting with the brand, starting with the interviewer introducing themselves, or asking a question with no wrong answer.
2. The reaction (2–8 seconds)
This is the respondent answering, and it is the part that proves the exchange is real. What sells it is not the words but the delay — the beat where they think, the eyes going up, the "uh", the laugh at their own answer.
The instinct when editing is to cut the pause because it feels slow. Do not. The pause is the evidence. A respondent who answers instantly and cleanly reads as an actor, whether they are one or not.
Two respondents is usually better than one here. The first gives a confident wrong answer, the second gives an honest one that is worse. The contrast does the work.
How it fails: trimming the thinking time, using only articulate answers, or letting the respondent talk long enough to become boring.
3. The turn (8–14 seconds)
The turn is where the interviewer introduces the information that reframes the answer. It is a fact, a number, or a comparison that makes the respondent's answer land differently in retrospect.
"The average is closer to seven." "Most people guess two. It's nine." The turn should come from the interviewer, not from a voiceover or a caption card, because it keeps the whole thing inside one conversation.
The best turns produce a visible second reaction — the respondent's face when they hear the real number. If you can get that in the same take, you have the strongest three seconds in the ad.
How it fails: the turn is a claim rather than a fact, the turn arrives from nowhere with no source, or the turn is delivered by a graphic instead of a person.
4. The reveal (14–24 seconds)
Now, and not before, the product appears. The reveal connects the gap the turn just opened to the thing you are selling.
The mistake is to make the reveal an advertisement — to change the register, brighten the lighting, cut to a clean product shot, and start listing benefits. Keep it in the same conversation. The interviewer mentions what they use, or hands the respondent something, or the respondent asks what they are supposed to do about it.
The reveal should be one claim, not three. You have roughly eight seconds and the viewer has already decided whether they are still watching. One specific, defensible statement about what the product does will outperform a list every time.
How it fails: the tonal break into commercial mode, too many claims, or a product shot so polished it announces that the preceding twenty seconds were setup.
5. The close (24–30 seconds)
The close is the instruction, and it should be short and physically specific: tap the link, it's in the bio, first order gets a code. One sentence.
The strongest closes come back to the respondent — their reaction to the product, a last line from them, a laugh — rather than ending on the interviewer or on a logo. Ending on the stranger keeps the ad feeling like content to the last frame.
How it fails: a long CTA, multiple CTAs, or a branded end card that adds three seconds of nothing.
Timing and total length
Thirty seconds is the working target. The structure compresses to twenty and stretches to forty-five, but below twenty the turn has no room and above forty-five the reaction section starts repeating itself.
Proportionally: about a fifth of the runtime on hook and reaction combined, a fifth on the turn, a third on the reveal, and the remainder on the close. If your reveal is running longer than that, you have written a commercial with a street-interview opening rather than a street interview.
Writing it as beats
The practical way to build one is to write each part as a separate beat with its own direction — who speaks, what they do, what the camera does — rather than writing a continuous script. Beats let you re-render the part that is not working without touching the parts that are.
In practice the turn is the beat you will rewrite most. It is the hinge of the whole ad, and small changes to the fact you use move performance more than anything else in the structure.
Variations worth testing
Once the five parts are in place, the cheap tests are: a different question with the same turn, a different turn with the same question, and a different respondent order. Those three produce meaningfully different ads from the same underlying concept, and they are far more productive than changing the music or the caption style.
What is not worth testing early: the interviewer. Keep them fixed. The recognizability across videos is worth more than any individual performance difference.