Listen where they talk
Your audience doesn’t talk in one place, and it rarely talks where your brand posts. Fans talk under trailers on YouTube, in Discord servers, in Twitch and Kick chat, in Reddit threads, in TikTok comments, in Facebook groups, and in reviews on places like Goodreads, Letterboxd and Steam. Search data shows what they look for. The full list of where we read is on our listening page.
| Where | What you learn there |
|---|---|
| Comments under trailers and posts | What fans reward: the callback, the joke, the verdict |
| Reddit and forums | What people ask each other, and what they never raise |
| TikTok formats they run themselves | The bits and the topics they own |
| Discord, Twitch and Kick chat | Live reaction, in their own words |
| Reviews | What the product is judged on |
| Search | What they look for, and when |
The rooms matter more than the mentions. In a real Arcus audience report for a streaming service, men 18 to 24 almost never asked each other what to watch. Of 4,283 questions on one large men’s forum, 1.9% were about hobbies, games and entertainment. Entertainment came up as a question in 47 titles, with a median of one upvote. The talk happened under the shows themselves: 78,682 comments under four of the service’s trailers on YouTube. A tool that tracked brand mentions would have found a quiet audience. Reading the right rooms found a loud one. See the audience report.
Count what they do
A summary says “young men use a lot of slang.” A count says one word in every 62 is slang. The first leads a brand to write more slang. The second leads it to write less.
Counting also lets you compare. If you count your audience and your brand with the same script, the gap between them is a finding. If you count the same rooms every quarter, a change is a finding. A summary can’t do either, because two summaries written by different people can’t be lined up.
Here’s what counting looked like in that report:
- 293 TikToks from eight formats the audience runs themselves, each sorted by hand.
- 2,965 comments measured for length, first person, punctuation and hashtags.
- 480 comments labeled by hand, one kind each: a joke, a verdict, a question, a sincere line.
- 6,622 comments counted for the words they use.
The counts changed the voice rules. An acted-out sketch got 42 times the views of a sincere post. The most searched slang of 2025, “6-7” at 141,000 searches, appeared in one comment of the 6,622. A sentiment score wouldn’t have shown either.
A worked example: the slang question
The brand’s question was simple: should our captions sound like our audience? The listening answered it in three counts.
| What was counted | The audience | The brand’s captions |
|---|---|---|
| Words in a typical line | 5 | 8 |
| Lines ending in a full stop | 4.9% | 35% |
| Lines with a hashtag | 0.2% | 92.5% |
The words they did use were verdicts on the work, like “peak,” “real” and “cinema.” “Rizz” appeared in none. The answer: the brand’s captions already sounded like a brand, and adding slang would make it worse. Write shorter, drop the hashtags, and let fans supply the words. Our guide to how young audiences write has the full count.
Notice the shape of the answer. It isn’t “young men are sarcastic” or “use more humor.” It’s three measurable differences between the audience and the brand, each one something a social team can change in its next caption. That’s the test for any listening finding: could someone act on it this week?
A listening brief you can use tomorrow
- Write the decision the listening should inform. “Should we change our caption voice?” beats “what are people saying?”
- Name the rooms. Where does this audience talk when you’re not posting?
- Name what you’ll count. Views by format, likes on the top comment, words per 1,000 comments, topics per 100 questions.
- Count the brand the same way. Same script, same measures, your own posts.
- Keep real quotes with their like counts, so every finding has an example.
- Line it up with your own numbers. Do the topics that spike in conversation show up in sales, sign-ups or viewing?
- End with keep, test, stop.
Questions worth asking your team: Which rooms do we read beyond brand mentions? What do we count, and can we see the counts? Do we measure our own posts with the same rules? Can we show three real quotes, with like counts, for each finding? What did we change because of the last listening report?
Common mistakes
- Tracking mentions only. Your audience may never say your name. They talk about the show, the drink, the moment.
- Trusting sentiment scores. A top reply under a brand post called the trailer bad and promised to watch anyway. It’s sarcasm, and it means “I’m in.” A sentiment tool would score it negative.
- Reading only the top posts. The comments are often where the meaning is.
- Stopping at a word cloud. If a listening report ends with one, it isn’t done.
- Counting the brand differently. If the audience is counted by script and the brand is judged by eye, the comparison doesn’t hold.
- Quoting without counts. One vivid comment can mislead. Pair every quote with how often its kind shows up.
What it doesn’t prove
Listening tells you what people say and reward in public. It doesn’t tell you who buys. Public rooms skew: the forum in the example skews older than 24, so the report showed the 57 posters who stated an age of 18 to 24 separately. Some platforms hide counts: Instagram returns no like counts on comments. Lining it up with your own numbers can be as simple as a weekly table: the count of a topic in conversation next to sign-ups, sales or viewing by title for the same weeks. When the two move together more than once, you have something worth testing. Listening gets useful when you line it up with your first-party data, and it becomes proof only with a test.
Arcus plugs in the listening tool you already have, reads the rooms your audience uses, and lines what people say up with your own numbers. To see it on your audience, try it free with a question like “Should our captions sound like our fans?”