Fyslide

Strategy · 7 min read

How to test TikTok hooks: a matched-pairs protocol and a results sheet

Change only the hook. Pair two similar topics, let a coin flip decide which gets the challenger, post them on different days and read the same number at the same age. Repeat for at least six pairs, count the wins and compare the count with the coin-flip table below. One month of posts rarely settles it, so treat a result as a lead.

Updated · Published by Fyslide

Can you A/B test hooks on TikTok and Instagram?

Not the way an ad platform splits an audience. As of November 2026 we found no TikTok or Instagram statement that describes a built-in way to show two versions of one photo post to equal audiences. So you test by posting two separate posts and comparing them. Audiences, timing and topics differ between posts, which means one pair proves nothing. Many pairs can show a pattern.

Two facts shape the design.

  • The post is the unit. Instagram's Help Center (opens in a new tab) says likes and comments apply to the whole post, not to one slide. We found no statement that either platform reports numbers for slide 1 alone. So the variable is slide 1, and the thing you measure is the whole post.
  • Measure what the platform names. TikTok's Help Center (opens in a new tab) lists likes, shares, comments, watching to the end or skipping, and accounts followed as interactions. It says nothing specific about photo posts, so use those as numbers to log, not as a promise about ranking.

What is the protocol, step by step?

  1. Write the question. "Does a number hook get more shares per 1,000 views than a question hook?" Name hook A, your usual one, and hook B, the challenger. The hook type is the only variable.
  2. Choose the measure before you post. Pick one number, for example shares divided by views, times 1,000, read at 48 hours. If you choose after you see the results, you will choose the one that flatters your favorite.
  3. Build matched pairs. Two topics from the same pillar, with the same format and slide count, and about the same appeal. "Why shots taste sour" and "why shots taste bitter" match. "Dosing" and "my studio tour" do not.
  4. Flip a coin. For each pair, the coin decides which topic gets hook B. If you decide, you will hand the better topic to the hook you like.
  5. Freeze everything else. Design, slide count, caption length and structure, hashtags, sound and time of day stay the same. Slides 2 onward follow the same structure in both posts.
  6. Post the two on different days. Leave at least two days between them, and alternate which one goes first from pair to pair.
  7. Read both at the same age. Score each pair as an A win, a B win, or a tie. Set the tie margin before you start, such as 0.5 shares per 1,000 views.
  8. Stop at the planned number of pairs. Decide it in advance, and do not stop early because B is ahead.

The table sorts what to vary from what to hold.

VaryHold constant
Hook type on slide 1, and the title or first caption line that repeats itDesign and slide count
The topic, within a matched pairPillar and format
Caption length, hashtags and sound
Posting window and reading age

TikTok's Creator Academy (opens in a new tab) says you can edit the image cover, so the cover is a choice you make. Keep the cover picture the same in both versions of a pair, or you have changed two things.

How many pairs are enough?

No source gives a sample size for this. Our rule of thumb is a minimum of six pairs before you call anything a lead. The reason is arithmetic. If the hook made no difference, each decisive pair would behave like a coin flip, and the table shows how often a run of B wins would then happen by luck. Ties are left out.

Decisive pairsB winsIf hooks made no difference, this many or more would happen
441 time in 16
551 in 32
546 in 32, about 1 in 5
661 in 64
657 in 64, about 1 in 9
879 in 256, about 1 in 28
8637 in 256, about 1 in 7
10856 in 1,024, about 1 in 18
107176 in 1,024, about 1 in 6

Read the table as a minimum bar, not a verdict. Real pairs are noisier than coin flips, because topics, timing and the platform all vary, so a lucky streak is more likely than the table says. It does have one good property: a single breakout post counts as one win however many views it got, so it cannot carry the whole test.

What does a finished test look like?

Fictional numbers for a home espresso account. Hook A is a question and hook B is a number. The measure is shares per 1,000 views at 48 hours, and the tie margin is 0.5.

PairTopic with hook ATopic with hook BRate ARate BWinner
1"Why does my espresso taste sour?""3 reasons your shots taste bitter"4.25.1B
2"Is your grinder the problem?""4 signs your grind is too coarse"3.83.5Tie
3"Are you tamping too hard?""5 dosing mistakes at home"2.94.4B
4"Does your milk jug matter?""4 milk steaming errors"5.06.2B
5"When did you last clean the group head?""6 parts to clean weekly"4.13.4A
6"Does storing beans change the taste?""5 bean storage mistakes"2.23.0B

Pair 2 is a tie, because 0.3 is under the margin. That leaves five decisive pairs, and B won four. The table says four or more of five happens about 1 time in 5 by luck alone. That is a lead, and not a result.

Look for a leak as well. Three of B's four wins used the words "mistakes" or "errors". That is a second variable hidden inside "number", and it could explain the wins. The next round should keep the wording of the promise the same and change only the number.

What results sheet should you use?

Copy it once per test. Fill the header before you post anything.

Hook test results sheet
Question: {one sentence}
Hook A (usual): {type} · Hook B (challenger): {type}
Measure: {number} read at {age}
Tie margin: {number}
Planned pairs: {number}

Pair | Topic A | Topic B | Coin gave B to | Posted A | Posted B | Rate A | Rate B | Winner | What else differed
{n} | {topic} | {topic} | {A or B} | {date} | {date} | {number} | {number} | {A / B / tie} | {note}

Tally: decisive pairs {number} · B wins {number} · row in the table {text}

Tracking slideshow performance shows a sheet for the per-post numbers that feed it. In Fyslide, you can link a post and type in its numbers (analytics guide), or connect TikTok so the app reads the numbers of your public TikTok posts. The app can make up to ten slideshows from one topic in one job, so you can pick the A and B hooks from a batch. The hook generator gives hook lines in eight styles without AI.

What can spoil a hook test?

  • A second change. A new cover picture, a longer caption or a different sound turns a hook test into a guess.
  • Unmatched topics. A broad, popular topic against a narrow one tests the topic.
  • Small view counts. At 500 views, one extra share moves the rate by 2 per 1,000. Pairs with tiny counts are mostly noise.
  • Timing. A holiday week, a trend or a change in your posting habits can lift one half of a pair.
  • Stopping when you are ahead. Choose the number of pairs first and finish it.
  • An overpromising hook. B can win swipes and still disappoint. Check that slide 2 delivers, as in honest hooks without clickbait.

What do you do with the result?

If B leads and the table says the run is rare, make B's hook type your default for that pillar and test a new challenger against it. Keep the type, and vary the wording, so you do not repeat yourself. If no hook leads, keep what you use and look elsewhere. A no-difference result suggests that the hook type is not what holds your posts back. The diagnostic for slideshows not getting views lists other causes. For hook types to test, see the hook formulas and curiosity hooks.

What should you check before the first post?

  • One question, one variable, written down.
  • The measure, the reading age and the tie margin are fixed.
  • Each pair is matched on pillar, format and slide count.
  • A coin decided which topic gets hook B.
  • Everything except the hook is the same in both posts.
  • The number of pairs is decided, and it is at least six.
Can I post the same topic twice with two hooks instead of pairing topics?

You can. In our view a repeated topic can read as a repeat to your followers, and the first post may change how the second one does. Matched topics avoid both problems, so we prefer them.

Can a small account run this test?

Yes, because it compares rates, not totals. Small counts are noisy, though, so expect to need more pairs.

Sources

  1. How TikTok recommends content, TikTok Help Center. Checked November 5, 2026.
  2. Share a post with multiple photos or videos on Instagram, Instagram Help Center. Checked November 5, 2026.
  3. Unleash Your Creativity: Tell Stories with Photos on TikTok, TikTok Creator Academy, 2026-07-03. Checked November 5, 2026.
  4. Fyslide FAQ, Fyslide. Checked November 13, 2026.