Why researchers want a unit
To study how culture spreads, it helps to have something countable. Hashtags and memes are attractive for that reason: they are discrete, labelled or recognisable, and they leave a trail of timestamps. Researchers can ask when a tag first appeared, who used it, how quickly it grew and where it faded. The word “meme” itself was introduced in popular writing as a way of thinking about ideas that copy and spread somewhat like genes, and online usage later narrowed it to a recognisable image, phrase or format that people copy and vary. This article looks at what these units do and do not allow researchers to say.
Hashtags: easy to count, but only a slice
A hashtag is a label chosen by the person posting. That makes it comparatively simple to collect: search for the tag and gather the posts. Typical measures include volume over time, the number of distinct accounts using it, and the structure of who is repeating whom.
The limits are significant:
- Not everyone tags. Many posts about a topic carry no tag, so a tag count covers only people who chose to use it.
- Tags can be adopted deliberately. Campaigns, brands and organised groups can push a tag, so volume may reflect coordination more than organic interest. Some of the activity may come from automated or fake accounts.
- One topic, many tags. Spelling variants, translations and competing tags split a conversation across labels, while one tag can be used for unrelated purposes.
- Platform rules matter. Access to data, search tools and display choices determine what can be collected, a theme in platform data and its blind spots.
Tag counts therefore describe the use of a label, which may or may not match the spread of the underlying idea.
Memes: harder to define, harder to follow
Memes change as they spread. A format may be reused with new captions, edited images, re-recorded audio or translated text. Deciding whether two items are the same meme or different ones is a judgement call. If researchers define the unit too narrowly, they miss variants and undercount; too broadly, and they combine things that have little in common.
Methods that researchers use to follow memes include:
- Image matching. Software compares pictures to find near-duplicates, which can group copies and light edits. It tends to struggle when a template is heavily changed or when the text is the real content.
- Text and phrase tracking. Researchers follow repeated phrases through posts. This can pick up verbal formats but also catches coincidental or common wording.
- Audio and video fingerprints. For clips and sounds, matching techniques identify reuse of the same recording.
- Manual coding. People review samples and classify them, which is flexible but slow and subject to disagreement between coders.
- Network tracing. Where the data allows, researchers follow shares and reposts to reconstruct paths from one account to another.
Each method has false positives and false negatives, and the results are only as good as the sample being examined.
Reconstructing paths is often incomplete
A tidy chart of a meme “spreading” from one community to another can create the impression of a clear route. In practice, the first appearance in the data is not necessarily the first appearance in the world. Material may have circulated in private messages, in smaller or deleted posts, or on platforms the researchers could not access. Public share counts record visible activity, not the full transmission. This is the same blind spot discussed in the context of group chats, where much cultural exchange takes place out of sight.
Algorithmic distribution adds another layer. A meme seen by many people may have been pushed by a recommendation system rather than passed from person to person, so network paths mix human sharing with platform selection. The mechanics are covered in the explanation of recommendation algorithms.
Survivors and the bias of the record
Researchers frequently notice a meme only after it has become widespread. By then, thousands of short-lived formats that never took off are absent from the data, which makes successful memes look more typical than they are. This is a form of survivorship bias: studying what lasted gives a distorted picture of what is likely to last, and makes it tempting to read the winners’ features as the cause of their success.
It is also hard to say why one variant did well. Timing, audience, an influential early sharer and platform promotion all play a part. Linking features to success is another case where causal claims need caution.
Shapes of spread
Researchers often compare the pattern over time to models of diffusion. Some items show a slow start, then rapid growth, then decline; others spike suddenly and vanish, and others persist at a low level for years. Frameworks such as diffusion curves give vocabulary for these shapes, and notions like tipping points describe moments when growth accelerates. These are descriptive tools, and fitting a curve to data does not by itself explain why the pattern occurred. Many items also fade quickly, so what looks like a lasting cultural shift may be a brief burst, a distinction explored in work on trend lifecycles.
Connection to language and community
Memes and tags often carry or create new words and expressions. Studies of internet slang and of emoji use draw on similar methods to follow words as they move between groups. In-group references can also define communities, so a meme may matter greatly to one group and be invisible to others. A large count in one community may not translate to broad public awareness.
Reading a meme-spread claim
- How was the meme or tag defined, and were variants included?
- Which platforms and which time window were covered?
- Is the start point the real start, or only the earliest visible post?
- Could coordination, automation or promotion explain the pattern?
- Are failed or forgotten examples considered, or only the successful one?
Hashtags and memes make cultural spread observable, but only in part, and mostly in the parts that leave public, searchable traces.