VSL headline: how to write the line above your video and test which one wins

A VSL headline is the text above the video. It has three jobs: earn the play or the tap for sound, frame what the viewer will watch, and confirm they're in the right place after clicking your ad. Keep it to the ad's promise, short enough that the video stays on the first phone screen. Test variants on unmute or play rate; confirm with pitch reach and revenue per visitor.

What does a VSL headline need to do?

The headline is the first thing a visitor reads, often while the video is already playing muted. In a second or two it has three jobs to do.

Three jobs in a second or two

1. Earn the play or the unmute

With muted autoplay the video moves on its own, and the visitor decides whether to tap for sound. On a click-to-play page, they decide whether to press play at all. The headline supplies the reason. The VSL play rate guide covers the other things that move this number, such as load speed, the poster and the unmute prompt.

2. Set the frame

It tells the viewer what the video will give them, which is the promise your lead then pays off. A viewer who knows what's coming will sit through a slower first minute; one who doesn't has no reason to stay.

3. Match the ad

The visitor clicked a specific promise. Seeing it restated in similar words confirms they're in the right place. A headline about something else makes them wonder whether they clicked the wrong link.

One rule covers all three jobs: the headline is advertising, so its claims must be true and backed by evidence before it runs, including the claims it implies (FTC advertising FAQs). This isn't legal advice, but a headline your video can't support costs you twice: in early drop-off and later in refunds.

VSL headline formulas with examples

Every example below is for a fictional product. Borrow the formula, not the words, and promise only what your video and product deliver.

Two notes. Story headlines must be true stories about the real presenter. And "watch this short video" belongs only above a video that is actually short; calling a 40-minute VSL short is a false claim.

To pick a formula, start from the traffic. Ads that lead with a problem pair with call-out and "why" headlines. Ads that lead with a result pair with "how" headlines. A contrarian ad needs a contrarian headline. The headline should read like the next sentence of the ad.

How should you stack the pre-headline, headline and sub-headline?

A VSL page can carry up to four short pieces of text around the video, each with one job:

The text around the video

1. 1Pre-headline (optional)A short call-out above the headline, such as "For home espresso owners". It filters the audience and confirms it.
2. 2HeadlineThe promise or the hook.
3. 3Sub-headlineThe mechanism, a proof point or what happens in the video, such as "A barista trainer shows the two-minute fix in this video".
4. 4Unmute promptOn the player itself, such as "Tap for sound". It's a player element, not page copy, but visitors read it as part of the stack.

A sub-headline earns its place when it adds something the headline can't carry: a name, a specific detail, a reason to believe. If it only restates the headline, cut it and give the space to the video. The VSL landing page guide covers what goes below the video before and after the pitch.

How long should a VSL headline be on mobile?

Long enough to carry one promise, short enough that the video stays on the first phone screen. That second limit is the one that bites: a headline that pushes the player below the fold costs you plays before anyone finishes reading it.

Fonts and sizes differ, so measure your budget on a real device instead of borrowing a character count:

1. Open the page on a mid-size phone in portrait.
2. Count how many characters of your headline fit on one line at your headline size.
3. Multiply by three. That's your headline budget. If 24 characters fit per line, you have about 72.
4. Check that the pre-headline, headline, sub-headline and the whole video frame are visible without scrolling. If they aren't, cut words before you shrink the font.

Write the long version first, then cut. Remove adjectives that don't change the meaning, fold the call-out into the headline, and move proof into the sub-headline. Keep the type large and the contrast high; thin gray type over a busy background makes the headline work harder than it should.

How to run a VSL headline A/B test

Run it in seven steps:

A headline test in seven steps

1. 1Start with a hypothesis tied to a reason"Visitors from the leash-pulling ad need the problem named, so a call-out headline will lift unmute rate" beats "let's try a shorter one".
2. 2Write two or three variants that differ in ideaSwapping "simple" for "easy" isn't a difference you can measure on normal traffic.
3. 3Split traffic at random and keep each visitor on one versionChange nothing else: same video, poster, page and ads.
4. 4Pick the primary metricWith muted autoplay, use unmute rate; on click-to-play pages, use play rate. Set pitch reach and revenue per visitor (RPV) as guardrails.
5. 5Calculate the sample size before you startPlay and unmute rates are high, so they need far fewer visitors than sales do. The worked example below shows the math.
6. 6Run whole weeks and stop at the planned sampleAt 600 visitors a day split two ways, about 1,530 visitors per variant takes roughly five days, so you'd run a full week. Don't stop because one version leads on day two.
7. 7Decide with the decision tree in the next section, then log the result

The VSL A/B testing guide explains sample size, significance and the peeking problem in more depth. In RevvyScales (Growth plan and up), headline variants can run inside one VSL: each is shown by the weight you set and stays sticky per viewer, and analytics break the results down by variant (split tests inside one VSL).

Worked example: sizing a headline test

Say your unmute rate is 55% and you want to be able to detect a lift to 60%. At 95% confidence and 80% power, the standard calculation for two proportions gives about 1,530 visitors per variant. To detect a smaller lift, to 59%, you need about 2,400 per variant. The rule of thumb 16 × p × (1 − p) ÷ (difference)² lands close: 16 × 0.55 × 0.45 ÷ 0.05² ≈ 1,580.

Compare that with a test decided on sales. At a 2% conversion rate, detecting a 20% lift takes about 21,000 visitors per variant, roughly 14 times the 1,530 above. That's why headline tests are decided on unmute or play rate, with revenue as a guardrail rather than the deciding metric.

Here are the three sample sizes side by side:

Visitors per variant, by the metric that decides

The worked example at 95% confidence and 80% power

Unmute rate, 55% to 60%about 1,530

Unmute rate, 55% to 59%about 2,400

Sales: a 20% lift on a 2% conversion rateabout 21,000

Now the result. After a week at 600 visitors a day, A has 2,100 visitors and 1,155 unmutes (55%), and B has 2,100 visitors and 1,239 unmutes (59%). The pooled rate is 57%, so the standard error is √(0.57 × 0.43 × (1 ÷ 2,100 + 1 ÷ 2,100)) ≈ 1.53 percentage points, and z = 4 ÷ 1.53 ≈ 2.62. That clears the 1.96 bar for 95% confidence; the method is the two-proportion z-test (NIST's description). Next, check the guardrails.

What should you judge a headline test by?

Unmute or play rate decides the test. Pitch reach and RPV, both defined in the VSL metrics guide, guard against a headline that wins attention by attracting the wrong people:

Ship, rewrite, extend or keep?

Unmute or play rate decides; pitch reach and RPV are the guardrails

1What did unmute or play rate do?

Up, significantFlat, with pitch reach and RPV upDown, significant

2What did pitch reach and RPV do?

Flat or upDown

Best fit

Ship the new headline

Unmute or play rate up and significant, with pitch reach and RPV flat or up.

Best fit

Curiosity without fit: don't ship

Unmute or play rate up and significant, but pitch reach and RPV down. Rewrite to keep the frame.

Best fit

Better frame: extend the test

Unmute or play rate flat, with pitch reach and RPV up. Extend the test and judge on RPV.

Best fit

Keep the control

Unmute or play rate down and significant, whatever pitch reach and RPV did.

RPV rarely reaches significance in a headline test, because the sample is sized for unmute rate. Treat it as a direction check and ship only if RPV isn't clearly worse. In the example above, say pitch reach is 21.0% for A and 21.4% for B, and RPV is $1.62 against $1.68. The guardrails agree, so B ships.

How many headline variants should you test at once?

Two or three for most VSL budgets. Each extra variant takes its own share of traffic, and each extra comparison adds another chance of a false winner. A common, conservative correction (the Bonferroni method) divides the 5% error budget by the number of comparisons. With three challengers against one control, each comparison then needs z above about 2.39 instead of 1.96. That raises the visitors needed per variant by about a third, on top of splitting traffic four ways: roughly 2.7 times the total traffic of a simple A/B test.

How often should you test VSL headlines?

Tie headline tests to your ads. Each time you launch a new ad angle, write a headline that matches it and test it against the current control on that angle's traffic. Between launches, run one headline test at a time per VSL, back to back, and log each result with its hypothesis.

A rule of thumb: once a headline has beaten three challengers in a row, move your testing effort to the lead or the offer, the bigger levers in the impact order. Come back to the headline when the ads, the audience or the market's awareness change.

Common VSL headline mistakes

• Clever instead of clear. A pun that needs a second read loses the visitor who only gives it one.
• Mismatch with the ad. The headline should be the ad's promise in page form.
• Promising what the video doesn't deliver. It wins the unmute, then loses viewers in the first minute and invites refunds.
• Pushing the video below the fold. Measure on a phone, not on your laptop.
• Hard-to-read styling. Thin fonts, low contrast and busy backgrounds slow down the one line that has to be read instantly.
• Testing synonyms. On normal traffic, near-identical headlines produce differences you can't measure.
• Calling a winner on a few hundred visitors. Use the sample-size math above.
• Changing the headline and the poster together. You won't know which one moved the number. Test the VSL thumbnail in its own test.

What Caples-era headline testing still teaches

Headline testing is older than the web. John Caples wrote the 1926 mail-order ad "They Laughed When I Sat Down at the Piano" for the U.S. School of Music, and later collected his methods in Tested Advertising Methods, first published in 1932 (biography). Mail-order ads could be judged by the replies and orders they produced, and Caples is credited with popularizing testing as a core part of advertising.

The lesson carries over directly: the scoreboard decides, and your opinion of a headline is a hypothesis until the numbers come in. Your unmute rate and RPV are the coupon count. The Caples headlines guide covers his principles in more depth.

A 30-minute headline test checklist

Tick these off as you go; the ticks stay in this browser:

Seven steps in 30 minutes

• Open the ad your traffic comes from and write its promise in one sentence.
• Read your current headline next to it. If they don't match, that's your first test.
• Write two challengers from different formulas in the table above.
• Check each one on a phone: the video must stay on the first screen.
• Check every claim against what the video and the product deliver.
• Work out the sample size from your current unmute or play rate.
• Set up the split, confirm each version loads, and write the end date in your test log.

0 of 7 done