How they write
For a streaming service, Arcus measured 2,965 comments in the TikTok formats men 18 to 24 run themselves, then measured the brand’s own TikTok captions from the same month with the same script. Same rules, both sides.
| Measure | Their comments | The brand’s captions |
|---|---|---|
| Words in a typical line | 5 | 8 |
| Ends in a full stop | 4.9% | 35% |
| Says I, me or we | 37% | 5% |
| Carries a hashtag | 0.2% | 92.5% |
| All lower case | 27.6% | Not measured |
| Swears | 3.4% | Not measured |
One of the service’s competitors already captions at the audience’s rate. The brand’s typical caption was eight words, a full stop and a show hashtag. The audience noticed. Under one of the brand’s posts, a reply asked, less politely, who wrote the caption. They can hear a script in a few words. See the audience report.
What they reward
Arcus labeled the three most-liked comments under every post in ten TikTok pulls, 360 comments in all, with one kind each.
| Kind of comment | Share of the 360 | A real example, with likes |
|---|---|---|
| Plays along with the bit | 25% | “could be a amazing movie if it was a real movie” (464,928) |
| A flat verdict | 13.9% | “Tuff and cringe at the same time” (86,630) |
| Roast or banter | 10.8% | “What he be sounding like” (53,948) |
| Joke at own expense | 10.6% | Under a friends’ trip post (21,418) |
| Sincere | 10.3% | “That ain’t a lazy sleep that’s an exhausted sleep” (143,571) |
| Hype, superlative praise | 8.1% | “you got them mad, i like you” (227,672) |
| A question | 3.9% | |
| Plain reaction | 3.9% | |
| Sarcasm, deadpan | 2.8% | A reply calling it bad and promising to watch anyway |
| A demand on the poster | 2.5% |
Jokes of some kind make up 49%. Sincere lines are about one in ten of the top comments, though they’re 14% of a random sample: they write sincerely more often than the room rewards it. Sarcasm is one comment in 36, but when it lands it’s the second most-liked kind, with a median of 9,796 likes. It’s aimed at the post, never at other commenters. Questions are rare. They don’t ask the room.
What they talk about matters as much as how. Across 4,283 questions on one large men’s forum, 36.7% were about dating and relationships. Among the 57 posters who stated an age of 18 to 24, it was 36.8%. Hobbies, games and entertainment made up 1.9%. They don’t ask each other what to watch. They talk under the show itself: 78,682 comments under four of the service’s trailers on YouTube. A brand that wants to join the conversation should go to that thread, not start a new one.
Their words, counted
Across 6,622 comments from their rooms, Arcus counted each term once per comment.
| The words they use | Per 1,000 comments | The glossary words | Per 1,000 comments |
|---|---|---|---|
| 😭 (crying-laughing) | 63 | unc | 0.8 |
| bro | 40 | sigma | 0.6 |
| peak | 8.3 | mog | 0.6 |
| real | 7.9 | skibidi | 0.2 |
| aura | 5.7 | 67 | 0.2 |
| lock in | 5.7 | rizz | 0 |
| fr | 4.1 | huzz | 0 |
| cinema | 1.7 | gyat | 0 |
The most searched slang of 2025 barely shows up. “6-7,” at 141,000 searches, appears in one comment. “Huzz,” at 61,000, appears in none. Meanwhile a 450-term slang glossary was being sold on X with the pitch “You don’t know the language? You’re invisible.” (1,378 likes).
The words they do use are verdicts on the work. They apply them to a scene, a fight or an actor: “Deadass that was goated acting,” 41,254 likes. Coming from a brand, the same words read as decoration. The report also cites a 2025 Journal of Marketing Research study in which slang in brands’ posts cost them likes and sales, in every age group.
The voice rules that follow
- Play along. Reply to their bit. Don’t start one. The top comment under a bit gets a median of 242 likes; under a sincere post, 3.
- Tell, don’t ask. A statement with a time in it. Questions are 3.9% of what the room rewards.
- Hand over the mic. Let a fan’s line be the caption.
- No glossary. No slang in the caption, and no hashtag stack. Write like the brand’s best account already does.
The 42x view gap behind the first rule is covered in this guide.
One model line from their own rooms shows the second rule: “It should not even be a question. We tee off at X o’clock.” A statement, a time, and the assumption that you’re coming. Under one trailer, “same time next year?” took 4,938 likes: the fans set the cadence themselves. The brand’s job is to say when, and then show up.
A caption check you can use tomorrow
- Five to eight words. Cut the rest.
- No hashtag, unless the platform needs one to find the post.
- No word from a slang list. If you looked it up, cut it.
- No question to the audience. Say when, and what.
- Could a fan have written it? If not, find a fan’s line and use that.
- Read it aloud to someone in the audience. If they wince, start over.
Questions worth asking your team: How long are our captions compared with our audience’s comments? How many carry a hashtag? Which of our posts got a reply like “who wrote this”? Who on the team writes our best-performing lines?
Common mistakes, and what this doesn’t prove
- Trusting a glossary. The most searched words were nearly absent from the comments.
- Reading sarcasm as negative. “This looks bad. I’ll be watching.” means “I’m in.”
- Measuring the audience and not the brand. The gap only shows when both are counted the same way.
- Asking the room a question. Questions were 3.9% of the top comments. A caption that asks “who’s watching tonight?” asks for something the audience rarely gives.
The counts describe one audience, men 18 to 24, on public posts in September 2026. One reader labeled the 480 hand-sorted comments, and every label is filed. Instagram returns no like counts on comments, so Instagram examples carry none. The counts show how the audience writes and what it rewards. They don’t prove a caption change will move viewing. Lining your posts up with your own numbers comes next. To see how your own audience writes, try it free.