# Which caption styles actually work?

A study of 13,046 TikTok and Instagram talking-head videos from 1,132 creators across 10 niches. Published 2026-11-07 by Yap & Done. Data: Posts 7–60 days old on TikTok (7–90 on Instagram) as of November 6, 2026.

Canonical page: https://www.yapanddone.com/studies/caption-styles/ · Data: https://www.yapanddone.com/studies/caption-styles/data.json · https://www.yapanddone.com/studies/caption-styles/data.csv · License: CC BY 4.0

## Key findings

- **Words on screen get saved more.** When the same creator put captions and a headline on a video, it got 12% more saves per view (range +7% to +18%) than their videos without. Their videos with no text got 12% fewer saves. This held after accounting for the type of video, and the creator's previous post showed no such pattern.
- **No single caption look wins.** Font, letter case, colours, highlighting the spoken word, one word or three at a time: none changed views, likes, comments, shares, saves or breakouts when creators are compared with themselves. The comparisons were precise enough to rule out effects bigger than about 15% on likes, shares and saves, about 17% on comments, and about 25% on views.
- **But looking out of place in your niche costs views.** Every niche has its usual caption setups. When a creator used one that fewer than 5% of their niche uses, that video got 14% fewer views (range -21% to -6%) than their own videos with a usual setup. The creator's previous post showed no such dip, and it held on TikTok alone, without typed-on captions, and with any one niche left out. The out-of-place setups were mostly ordinary styles that just aren't used in that niche.
- **The first frame needs words.** 83% of creators already have text up from the very first frame. When the same creator did, the video got +14% views; when text came in even half a second later, -12%.
- **Don't type captions on word by word.** Typed-on captions went with 18% fewer views and 11% fewer shares per view for the same creator; captions that simply appear went with 20% more views. Strong evidence, but short of proof after correcting for many tests.
- **Big differences between creators mostly weren't the captions.** At first one-word captions looked like +46% saves. Comparing like with like, it shrank to +12%, and to +4% within the same creator: the kind of creator and video using it, not the caption.
- **No caption style is a shortcut to going viral.** No style made a video measurably more likely to break out (reach at least 3x the creator's usual views).
- **76% of creators put a hook headline on screen.** The single most common setup (23% of creators) is a short headline with no captions. Only 8% use no text. Most captioners (78%) don't highlight the spoken word.
- **One word at a time is rising; typed-on captions are falling.** Compared with posts from 6–12 months earlier, one-word captions rose 6.9 points (now 20% of captioning creators) and words typed on one by one fell 7.5 points (now 10%).

## In one line each

- **Works:** Put words on screen. Captions plus a headline got 12% more saves, comparing creators only with themselves.
- **Works:** Look like your niche. A caption setup that's rare in the creator's niche got 14% fewer views than their own videos with a usual setup.
- **Works:** Have text up from the first frame. It went with +14% views; text that arrived even half a second later went with -12%.
- **Probably works:** Let captions appear whole. Words typed on one by one went with 18% fewer views and 11% fewer shares.
- **Doesn't matter:** Any one look on its own: font, ALL CAPS, colours, word highlighting, one word or three. None changed views, likes, comments, shares, saves or breakouts once creators are compared with themselves.
- **Read next:** What the headline promises and how you open moved views, shares and saves far more than any caption look.

## Cause or coincidence?

Each result checked three ways: all creators (raw), same type of video (content held fixed), same creator (each creator against their own posts). Previous post = the same test on the creator's prior post, which a caption cannot affect.

| Style → metric | All creators | Same type of video | Same creator (95% CI) | Previous post | Verdict |
|---|---|---|---|---|---|
| Captions plus a headline → saves | +26% | +14% | +12% (+7% to +18%) | +1% (flat) | likely causal |
| No text on screen → saves | -39% | -16% | -12% (-19% to -5%) | +1% (flat) | likely causal |
| Typed-on captions → views | -35% | -19% | -18% (-30% to -4%) | -8% (flat) | probable |
| Typed-on captions → shares | +7% | +12% | -11% (-19% to -4%) | -4% (flat) | probable |
| Captions that appear whole → views | +54% | +25% | +20% (+3% to +39%) | +6% (flat) | probable |
| Headline only, no captions → views | +33% | +29% | +10% (+0% to +20%) | +3% (flat) | probable |
| One word at a time → saves | +46% | +12% | +4% (-3% to +11%) | +7% (flat) | no effect |
| Box behind the spoken word → comments | +47% | +23% | -1% (-10% to +8%) | +3% (flat) | no effect |
| Plain (unboxed) headline → likes | +18% | +18% | +7% (+4% to +10%) | +7% (moves) | confounded |

## Fitting your niche

A post's caption setup (headline, words on screen, word highlight) is "rare in its niche" when fewer than 5% of the other creators in that niche use it, and "common" when 15% or more do. The creator's own posts are left out of the count.

- **Rare in the niche, same creator:** -14% views (-21% to -6%, p = 0.001, 463 creators who used both). Previous post: -5% (flat).
- **Common in the niche, same creator:** +7% views (-1% to +15%).
- **Without typed-on captions:** -12% views (-20% to -4%, p = 0.005).
- **TikTok only:** -14% views (-22% to -6%, p = 0.001).
- **Any one niche left out:** -11% to -16% views across all 10 versions.
- Likes, comments, shares and saves per view did not move; breakout odds leaned the same way but were not significant.


## Timing, placement and pace

- Text on screen from the first frame: 83.1% of creators; +14% views for the same creator (likely causal). Text arriving within the first second instead: -12% views.
- Caption position: lower middle 52%, middle 23.1%, top third 15.6%, bottom strip (app UI) 9.3%. Under the app's buttons: 4%; across the face: 3.5%; headline under the top tabs: 2.6%.
- Captions in the top third: -11% shares (likely causal). Captions in the bottom strip: -2% views (no effect).
- Talking fast for the caption type: -10% views (likely causal). Median with one-word captions: 2.9 words a second.

## Tools and habits

- Captions look made with: CapCut auto-captions 73.3%, Custom design 18.6%, TikTok in-app text 7.5%, Submagic / "Hormozi" style 0.5%.
- Shadow 45.2%, outline 48.3%, one line 66.9% vs two lines 26.1%, emojis 2.8%, white text 93.4%; headline only during the hook 50%. None of these (tool, shadow, outline, lines, size, colour, how long the headline stays) changed results when creators are compared with themselves.

## Small vs big accounts (same creator)

| Result | Under 10k followers | 10k–100k | 100k–1M | 1M+ | Tiers differ (p) |
|---|---|---|---|---|---|
| Captions + headline → saves | +15% | +12% | +10% | +14% | 0.92 |
| No text on screen → saves | -27% | -10% | +4% | – | 0.02 |
| Setup rare in the niche → views | -25% | -11% | -10% | – | 0.30 |
| Typed-on captions → views | -9% | -13% | -33% | – | 0.39 |

## Keeping one look

- 760 creators: one look 33.9%, mostly one 41.1%, mixed 25%.
- One look vs mixed: views per follower +41% (p = 0.002); views now vs 2–8 months earlier -13% (p = 0.26). Between creators only; follower growth is measured at the next edition.

## Claims registered on 2026-11-07 for re-testing next quarter

Each claim is re-tested on posts collected at the next quarterly refresh, leaving out every post used here, with the same code. Same creator, same controls (type of video, hook, production, setting, post age, length). A claim replicates when the new estimate points the expected way with p < 0.05. Every claim is reported, whether it replicates or not. File: https://www.yapanddone.com/studies/preregistered.json (SHA-256 c4b3a1411f993f5d45693ceec15bd6b1c099a186a7adf837ea9602992a204d31).

- Text on screen from the first frame gets more views: now +14% (+5% to +25%)
- Captions plus a headline on screen get more saves per view: now +12% (+7% to +18%)
- Captions typed on word by word get fewer views: now -18% (-30% to -4%)
- A caption setup that is rare in the creator's niche gets fewer views: now -14% (-21% to -6%)

## The Caption Index, Q4 2026

Share of creators, all niches weighted equally. Change compared with posts from the same niches 6–12 months earlier. Next edition: Q1 2027. History: https://www.yapanddone.com/studies/caption-styles/index.csv

- Hook headline on screen: 75.9% (+7.4 pts)
- Headline only, no captions: 23.2% (+0.7 pts)
- No text on screen: 8.2% (-2.8 pts)
- One word at a time (of captioned): 20.4% (+6.9 pts)
- ALL CAPS captions: 35.5% (+0.4 pts)
- Spoken word highlighted (karaoke): 21.5% (-2.4 pts)
- Typed-on captions: 10% (-7.5 pts)
- Heavy or tall fonts: 43.1% (-2.3 pts)
- CapCut auto-caption look: 73.3% (-1.8 pts)
- Text on screen from the first frame: 83.1%

## Caption styles

| Style | Used by (% of creators) | Share of breakouts | Trend vs 6–12 months ago (pts) | Breakout odds (95% CI) | Verdict |
|---|---|---|---|---|---|
| Hook headline, no captions | 23.2% | 25.4% | +0.7 | 1.11 (0.99–1.25) | no clear effect |
| Full-line captions | 16.8% | 16.6% | -4.3 | 0.98 (0.86–1.11) | no clear effect |
| 2–3 words, no highlight | 7.9% | 8.2% | +1.0 | 1.01 (0.84–1.21) | no clear effect |
| One word at a time | 7% | 8.4% | +2.0 | 1.20 (1.00–1.45) | no clear effect |
| One word at a time, ALL CAPS | 5.2% | 5.3% | +1.5 | 1.04 (0.86–1.25) | no clear effect |
| Karaoke colour highlight, ALL CAPS | 5.8% | 5.9% | -0.7 | 0.93 (0.75–1.14) | no clear effect |
| Karaoke box highlight, ALL CAPS | 3.6% | 4.9% | -1.6 | 1.17 (0.94–1.47) | no clear effect |
| No text on screen | 8.2% | 6.5% | -2.8 | 0.81 (0.63–1.04) | no clear effect |
| Typed-on captions | 10% | 7.3% | -7.5 | 0.73 (0.57–0.93) | suggestive worse |
| Only key words coloured | 9.6% | 6.1% | +2.9 | 0.67 (0.48–0.94) | suggestive worse |
| Tall bold font with pink highlight | 3.7% | 5.3% | -0.8 | 1.36 (1.03–1.80) | suggestive |
| Blue highlight | 4.3% | 3.6% | +1.5 | 0.73 (0.56–0.97) | suggestive worse |
| Serif captions | 5.4% | 5.3% | +1.9 | 1.08 (0.81–1.44) | no clear effect |

Shares for archetypes are among all talking-head videos; single-field styles (typed-on, key words, font, colour) are among videos with captions.

## Text on screen (974 creators)

- Captions and a headline: 52.7% of creators (TikTok 51.3%, Instagram 61.7%); breakout odds 0.99 (0.89–1.10)
- Headline only: 23.2% of creators (TikTok 25.4%, Instagram 13.2%); breakout odds 1.11 (0.99–1.25)
- Captions only: 15.9% of creators (TikTok 15%, Instagram 20.3%); breakout odds 0.95 (0.81–1.11)
- No text: 8.2% of creators (TikTok 8.3%, Instagram 4.9%); breakout odds 0.81 (0.63–1.04)

## Words on screen at once (734 creators)

- 2–3 words: 40.9% of creators (TikTok 40.5%, Instagram 39.2%); breakout odds 1.05 (0.93–1.19)
- A full line: 38.7% of creators (TikTok 38.8%, Instagram 40.1%); breakout odds 0.89 (0.78–1.02)
- One word: 20.4% of creators (TikTok 20.7%, Instagram 20.7%); breakout odds 1.09 (0.94–1.27)

## How the spoken word is marked (734 creators)

- Not marked: 78.5% of creators (TikTok 77.8%, Instagram 83.4%); breakout odds 1.02 (0.88–1.17)
- Changes colour: 12.8% of creators (TikTok 13%, Instagram 10.5%); breakout odds 0.92 (0.76–1.12)
- Box behind it: 7.7% of creators (TikTok 8%, Instagram 6.1%); breakout odds 1.10 (0.89–1.36)
- Colour and bigger: 0.6% of creators (TikTok 0.8%, Instagram 0%); breakout odds not tested

## How captions appear (733 creators)

- Appear whole: 89.3% of creators (TikTok 88.7%, Instagram 92%); breakout odds 1.35 (1.07–1.71)
- Typed on word by word: 10% of creators (TikTok 10.6%, Instagram 8%); breakout odds 0.73 (0.57–0.93)
- Pop in: 0.6% of creators (TikTok 0.7%, Instagram 0%); breakout odds not tested

## Letter case (734 creators)

- Sentence case: 37.8% of creators (TikTok 36.1%, Instagram 44.2%); breakout odds 0.99 (0.86–1.13)
- ALL CAPS: 35.5% of creators (TikTok 37.2%, Instagram 27.9%); breakout odds 1.05 (0.93–1.17)
- all lower case: 26.7% of creators (TikTok 26.8%, Instagram 27.9%); breakout odds 0.96 (0.82–1.13)

## Font family (734 creators)

- Clean sans (Inter, Helvetica, TikTok): 45.1% of creators (TikTok 44.4%, Instagram 49.7%); breakout odds 1.00 (0.89–1.13)
- Heavy or tall sans (Montserrat, Anton): 43.1% of creators (TikTok 44.8%, Instagram 35.3%); breakout odds 1.02 (0.90–1.15)
- Decorative (serif, rounded, script): 11.7% of creators (TikTok 10.6%, Instagram 15%); breakout odds 0.96 (0.78–1.19)

## Highlight colour (734 creators)

- No highlight: 69.9% of creators (TikTok 69.7%, Instagram 74.4%); breakout odds 1.13 (0.98–1.29)
- Yellow: 8.6% of creators (TikTok 8.3%, Instagram 10.2%); breakout odds 0.94 (0.73–1.20)
- Pink: 5.7% of creators (TikTok 5.8%, Instagram 6.5%); breakout odds 1.23 (0.96–1.56)
- Red: 4.5% of creators (TikTok 5%, Instagram 2.5%); breakout odds 0.78 (0.55–1.11)
- Blue: 4.3% of creators (TikTok 4.5%, Instagram 2%); breakout odds 0.73 (0.56–0.97)
- Green: 3.4% of creators (TikTok 3.2%, Instagram 1.9%); breakout odds 0.68 (0.38–1.22)
- Other: 1.4% of creators (TikTok 1.3%, Instagram 2.2%); breakout odds not tested
- Purple: 1.3% of creators (TikTok 1.3%, Instagram 0%); breakout odds not tested
- Orange: 0.6% of creators (TikTok 0.7%, Instagram 0%); breakout odds not tested

## Caption position (734 creators)

- lower middle: 75.2% of creators (TikTok 76.1%, Instagram 77.5%); breakout odds 0.98 (0.83–1.16)
- Centre: 8.7% of creators (TikTok 8.4%, Instagram 6.4%); breakout odds 1.05 (0.82–1.33)
- bottom: 7.3% of creators (TikTok 6.9%, Instagram 6.7%); breakout odds 0.92 (0.69–1.22)
- upper middle: 6.6% of creators (TikTok 6.4%, Instagram 7.1%); breakout odds 1.08 (0.79–1.47)
- top: 2.2% of creators (TikTok 2.2%, Instagram 2.3%); breakout odds 1.10 (0.60–2.02)

## Box behind the caption (734 creators)

- No box: 92% of creators (TikTok 92.7%, Instagram 91.3%); breakout odds 1.04 (0.80–1.37)
- Box behind each line: 6.7% of creators (TikTok 5.8%, Instagram 8.7%); breakout odds 1.02 (0.76–1.36)
- Box behind each word: 1.3% of creators (TikTok 1.5%, Instagram 0%); breakout odds not tested

## Headline style (844 creators)

- Plain or outlined text: 52.1% of creators (TikTok 51.5%, Instagram 56.8%); breakout odds 0.96 (0.85–1.09)
- In a box (like TikTok's text sticker): 47.9% of creators (TikTok 48.5%, Instagram 43.2%); breakout odds 1.04 (0.92–1.18)

## Looks made with (734 creators)

- CapCut auto-captions: 73.3% of creators (TikTok 74.3%, Instagram 72.8%); breakout odds 1.01 (0.86–1.19)
- Custom design: 18.6% of creators (TikTok 17%, Instagram 21.3%); breakout odds 0.96 (0.78–1.18)
- TikTok in-app text: 7.5% of creators (TikTok 8%, Instagram 5.6%); breakout odds 1.00 (0.79–1.28)
- Submagic / "Hormozi" style: 0.5% of creators (TikTok 0.6%, Instagram 0%); breakout odds not tested

## By niche

- **Skincare and beauty** (140 creators): headline on screen 76.6%, headline only 39.2%. Most used: No text on screen 10.1%; One word at a time 6.1%; 2–3 words, no highlight 9.3%; Full-line captions 19.2%
- **Fitness and nutrition** (110 creators): headline on screen 67.7%, headline only 29.1%. Most used: No text on screen 11.7%; One word at a time 5.8%; 2–3 words, no highlight 4.6%; Full-line captions 15.9%
- **Career advice** (81 creators): headline on screen 85.4%, headline only 26.8%. Most used: No text on screen 2.9%; One word at a time 4.1%; 2–3 words, no highlight 7.8%; Full-line captions 13.8%
- **Dating and relationships** (111 creators): headline on screen 78.1%, headline only 25.4%. Most used: No text on screen 2%; One word at a time 9.2%; 2–3 words, no highlight 5.1%; Full-line captions 25.5%
- **Manifestation and spirituality** (113 creators): headline on screen 82.2%, headline only 24.8%. Most used: No text on screen 6.9%; One word at a time 6.9%; 2–3 words, no highlight 11.2%; Full-line captions 24%
- **Personal finance** (135 creators): headline on screen 81%, headline only 22.5%. Most used: No text on screen 6.2%; One word at a time 14.9%; 2–3 words, no highlight 11.4%; Full-line captions 10.7%
- **Real estate** (54 creators): headline on screen 70.1%, headline only 20.7%. Most used: No text on screen 17.9%; One word at a time 2.5%; 2–3 words, no highlight 5.6%; Full-line captions 10.8%
- **Nervous system regulation** (111 creators): headline on screen 80.5%, headline only 17.9%. Most used: No text on screen 3.9%; One word at a time 11%; 2–3 words, no highlight 5.3%; Full-line captions 25.2%
- **Business and marketing** (81 creators): headline on screen 72.6%, headline only 14.8%. Most used: No text on screen 5.8%; One word at a time 7.8%; 2–3 words, no highlight 13.4%; Full-line captions 13.6%
- **Video editing with AI** (38 creators): headline on screen 64.5%, headline only 11%. Most used: No text on screen 14.9%; One word at a time 1.3%; 2–3 words, no highlight 5.3%; Full-line captions 9.2%

## TikTok vs Instagram

Instagram creators more often pair captions with a headline (62% vs 52% on TikTok) and less often use a headline alone (13% vs 25%). Caption details (words on screen, highlight, font, case) did not differ significantly between the platforms. The Instagram sample is smaller (161 creators, about 120 breakouts), so its numbers are less certain.

## Method

- Creators: in each niche we found English-language creators through keyword and creator search, then had a language model screen every account for being genuinely in the niche (score 60 of 100 or more) with at least 2,500 followers. Instagram creators were found through search and by checking the same handles TikTok creators use.
- Videos: we collected each creator's latest posts and kept videos 8–180 seconds long, posted 7–60 days before collection (7 days so views had settled; 90 days on Instagram, where creators post less). Paid or boosted posts were excluded.
- Breakouts: a breakout is a post with at least 3x the views of the creator's typical post, measured against the median of their roughly 20 nearest posts in time. The comparison group is a random sample of the same creators' posts, not filtered on performance. At most three posts per creator in each group.
- Talking head: Apple Vision face detection kept videos where a face fills the frame in at least half the sampled frames; the labelling model then confirmed a person speaking to camera carries the video (including split-screen and picture-in-picture).
- Labelling: from each video we took the first 8 seconds at 6 frames per second plus one frame at 60% of its length, picked the 2-second stretch with the most caption activity, and showed Gemini 3.8 Flash three images: four moments across the video, a 12-frame burst stamped with times, and one full-resolution frame. It answered a fixed codebook of 24 questions.
- Words on screen: checked against the raters, the first pass called full-line captions "2–3 words" too often. Every "2–3 words" label was re-checked with one narrower question (how many words the caption shows at once, headline excluded); 745 of 4,200 moved to a full line. Agreement with the raters on captioned videos rose from 80% to 90%.
- Quality checks: 48 videos were labelled independently by separate raters (agreement shown below); 211 were labelled twice to check consistency; every label was cross-checked against on-device text recognition. Fields with less than 85% agreement are reported in coarser groups.
- Cause or coincidence: 11,525 random posts from 1,026 creators (about 11 each) were also labelled for what the video is: its type (teaching, list, story, opinion…), hook, production, setting, b-roll and framing. Each result was checked three ways: between creators with all of that held fixed; within each creator, comparing their posts with a style against their own posts without it; and on the creator's previous post, which a caption cannot affect (a pattern there means timing or circumstance, not the caption). We also report how strong a hidden factor would need to be to explain a result away (the E-value).
- Timing, placement and pace: from the on-device text recognition of the first 8 seconds (6 frames a second). Text that stays on screen 1.5 seconds either side is the headline; the rest is captions. App UI zones: the bottom 20% and the right-hand button column (x > 85%, 40–80% down) on TikTok and Instagram, and the top 9% for the tabs. Talking pace counts each caption word once, when it first appears, and is compared within each caption size.
- Consistency: each creator's share of random posts using their most common caption setup, compared between creators with niche, platform, follower count and posts sampled held fixed. Follower counts at collection are stored so follower growth can be measured at the next edition.
- Statistics: "used by" is the share of creators using a style in the random sample, each creator counted once and each niche weighted equally. Breakout odds come from a logistic regression of breakout vs comparison posts with controls for niche, platform, follower count, post age and video length, and standard errors clustered by creator. With about a hundred styles tested, we apply a Benjamini–Hochberg false discovery correction (10%): "proven" requires passing it; "suggestive" clears the 95% interval but not the correction. Trends compare the same niches against posts from 6–12 months earlier.

## Limitations

- This is observational data, not an experiment. Comparing creators with themselves and holding the type of video fixed removes most of the obvious explanations, but a factor we did not measure could still play a part.
- Only text burned into the video is measured. TikTok and Instagram can also show their own captions, which do not appear in the video file.
- Labels come from still frames, so fast animation details can be missed; font identification is the least reliable field (77% agreement in three groups).
- Talking pace is read from the captions, so it covers captioned videos only and is noisy; consistency is compared between creators, not within them, so it is a correlation.
- Instagram coverage is thinner than TikTok, and creators were found through search, so very small accounts are under-represented.
- Results describe English-language talking-head videos in these 10 niches in late 2026; styles change, and we plan to refresh the study.

Cite as: Yap & Done (2026). Which caption styles actually work? https://www.yapanddone.com/studies/caption-styles/