Social media slang research: what to collect and clear

Social media slang research means collecting other people's words, and the words live in places with names: Reddit threads, TikTok comments, public X posts, Discord servers, and the group chats that leave no record at all. That is the whole method, and it is also the whole ethical problem, because the people whose words you collect wrote them for a room and not for a study. Most guidance treats collection as a technical matter and permission as a formality. It is the other way round.

This page covers building a small collection you are actually allowed to keep: what to gather, what to strip, which documents decide whether you may take it, what to ask, and what the result can and cannot answer.

What to take away

What counts as an item

A social media slang collection is not a dictionary and not a word list. It is a set of short items, each one a word or phrase in use — "sus", "rizz", "based", "no cap" — recorded with the context it appeared in: the platform, the kind of room it appeared in, the rough date, and whether the speaker was explaining the word or simply using it. One item with its context is worth more than a long list without one, because the context is the only thing that tells you what the word was doing.

Three ways to gather, with different costs

The middle row holds the most useful material and needs the most care. Interview material can tell you what a word signals, which no corpus can — a Discord regular explaining a server's private vocabulary, a TikTok commenter glossing a phrase for outsiders — and the same speaker can also tell you things that would identify them.

Then you must decide what to do with that, and the safe default is to record the explanation, not the explainer.

The first row has the opposite profile: no interaction, and a badly skewed sample. What is publicly reachable on Reddit, TikTok or X systematically excludes private conversation, closed servers, and everything already deleted, which is where most vocabulary is born. What that does to any conclusion is worked through in our page on internet culture archives.

Reactivity, and why it is not avoidable

The moment people know their language is being studied, they attend to it, and attention changes it. Speakers standardize, explain, and perform. This is the observer effect in a setting where the observed party can also read what you write about them, which adds a second loop the physical sciences do not have.

You cannot remove it. You can account for it:

Stripping identity at capture

The rule that prevents the most harm is the simplest: do not record who said it unless you need it, and you almost never need it.

Stripping at capture means, concretely: no handles or display names; no profile links, avatars or bio text; no server, subreddit or group-chat name that identifies a small room; no timestamps precise enough to place a person at a moment; and no quoted phrase so distinctive that pasting it into a search box leads back to the account that wrote it. Keep the word and the context; drop the account. If anyone in the material is in the EU, this is a legal matter as well as an ethical one: under the GDPR (Regulation (EU) 2016/679), a username that singles a person out is personal data.

The reason for the last two is that a collection of quotes with sources is a re-identification tool whether you intended one or not. Our page on critiques of social media describes what happens to an ordinary person who becomes findable, and nothing about a research purpose changes the outcome for them.

Permission, in three separate questions

Does the service allow it? A terms-of-service question, and each platform answers it differently: Reddit's User Agreement and API terms, TikTok's Terms of Service, X's Terms of Service, and Discord's Terms of Service each set their own rules for what you may pull and keep. Whatever they permit, they say nothing about ethics. A service can allow you to take material that the people who wrote it never meant to hand over.

What a small collection can answer

More than people expect, provided the question is scoped.

What it cannot answer is how common anything is, in that community or anywhere. Frequency requires a defined population and a defined denominator, and neither exists here, for reasons set out under benefits of social media, which meets the same wall from the direction of counting attention.

Tools do not close the gap. Google Trends reports relative search interest, not a count of speakers, and an API returns only what the platform chose to keep. Neither gives you a denominator.

Common questions

Is public material free to quote?

Legally it varies, and it is two questions rather than one: copyright, where a short quotation may fall under the fair-use exception in US law (Section 107 of the Copyright Act), and consent, where the writer's wishes are backed by no clause you can point to in the terms of service. This page cannot advise you on either. Practically, a quotable line is a searchable line, and searching it leads to a person. Paraphrase unless the exact wording is the finding.

Is a small collection large enough to be worth building?

There is no floor for a descriptive claim about existence and use. There is no ceiling that makes a frequency claim safe. The size question only matters for the claims you cannot make anyway.

Should I tell the community what I found?

If you can do it without singling anyone out, yes. It is the one part of the exchange that gives something back, and their corrections are usually better than your reading.

What if a moderator says no?

Then do not collect there. The material is not scarce, and a study that overrides a refusal has already answered the ethical question badly.