Skip to content

Semantic vs latent themes in thematic analysis

By Khyati Ojha · · 6 min read

Semantic vs latent themes, from Braun and Clarke's paper "Using thematic analysis in psychology", shown on real-life quotes, with when to use each.

Semantic vs latent themes in thematic analysis

The calls are done at the dental clinic group where you work. Your team rang patients who stopped booking check-ups and asked why. Almost all of them gave the same answer: they're too busy. So is that the real reason, or is something under it?

Thematic analysis gives you two ways to read those answers. A semantic theme stays with what people said, like "no time." A latent theme names the idea under the words, like dread of the dentist's chair. Start with semantic themes. Look for latent ones when the stated reason doesn't explain what people do.

This choice comes from a famous paper, "Using thematic analysis in psychology." And it changes what your team decides to fix.

The short version

  • "No time" is a semantic theme: it repeats what patients said.
  • "Dread of the chair" is a latent theme: it names a feeling under the words.
  • The two readings point to different fixes, like evening hours or a kinder welcome back.
  • Code every interview at the semantic level first.
  • Go latent only when the stated reason doesn't match what people do.
  • In the report, keep the quotes and your reading apart.

What are semantic and latent themes?

Semantic and latent are two levels you can code at in thematic analysis. Thematic analysis means reading interviews, tagging what people say, and grouping those tags into themes. A semantic theme stays with the surface meaning of the words. A latent theme names the ideas, beliefs or feelings that sit under the words.

The two terms come from Virginia Braun and Victoria Clarke's 2006 paper, "Using thematic analysis in psychology." You'll see it cited almost everywhere. A 2025 review of thematic analysis studies in Frontiers found the paper is cited by over a third of all the articles it looked at. And in a 2025 count by the journal Nature, it ranked third among the most cited papers of this century.

The paper sets out six phases for the whole method, from getting to know the data to writing the report. This post sticks to one choice inside it: how deep you read.

That choice sits beside another one. You can let themes come from the data, called inductive coding. Or you can start from a theory or a list, called deductive coding. Braun and Clarke call this the inductive vs theoretical choice. Either way, you still decide whether to read at the semantic or the latent level.

One patient's answer, "I just haven't had the time," read at two levels: the semantic theme "No time" stays with the words; the latent theme "Putting off something unpleasant" names the idea under them and needs quotes from several people.

Semantic and latent theme examples

The easiest way to see the difference is to code the same quotes both ways. Say three patients from the calls said these things. The semantic code repeats the point in plain words. The latent reading asks what idea could explain it.

What the patient said Semantic code Latent reading
"I just haven't had the time." No time Putting off something unpleasant
"I'll book once things calm down." Plans to come back Waiting feels safer than going
"My teeth are probably a mess by now." Worried about their teeth Fear of being judged by the dentist

Read the semantic column, and the fix looks clear. Open the clinics in the evening, and send more reminders.

Read the latent column, and you get a different fix. People may stay away because the visit feels like a telling-off. Then the clinic should change how it welcomes people back, so a long gap doesn't feel like a failure.

Both readings could be true. But they lead to different decisions. And the latent reading needs more proof. One quote can't carry it. You need the same idea, in different words, from several patients.

Should you use semantic or latent themes?

Use semantic themes when your team needs to decide what to change, and use latent themes when the stated reasons don't explain what people do. Most studies at work start semantic. They go latent only on the few themes where the "why behind the why" changes the decision.

Semantic themes fit best in these cases:

  • The decision is about something concrete, like opening hours, prices or a booking step.
  • Several people or several clinics must agree on the same codes.
  • The people deciding will act on the theme without reading the transcripts.
  • You spoke to only a few people, so a deeper reading would rest on too few voices.

Latent themes are worth the extra work in these cases:

  • The stated reason doesn't match the behavior. Say patients blame time, but the evening slots stay empty.
  • The topic is fear, trust, habit or identity, which people rarely name out loud.
  • Every semantic theme is true, and the team still doesn't know what to do.

Latent reading depends more on the person doing it, because your own views shape what you see under the words. Braun and Clarke later named their approach reflexive thematic analysis. In reflexive thematic analysis, the researcher's view is part of the method, so you write that view down before you start. You can read more about reflexivity in the glossary.

Decision flow: code everything at the semantic level. If the stated reasons explain what people do, report semantic themes. If not, and several people say the same idea in different words, report a latent theme with quotes and reading apart; otherwise talk to more people.

How to report latent themes

To report a latent theme, split it into two parts: what people said, with quotes, and what your team thinks it means. Then say how many people the reading rests on. That way, anyone reading your report can check the evidence and judge the reading for themselves.

1. Put the quotes first

Show quotes from several patients, in their own words. Pick quotes that say the same idea in different ways. In the dental example, that might be "too busy," "once things calm down" and "probably a mess by now."

2. Write your reading as a reading

Use plain words that show it's your view. "We think patients avoid the visit because it feels like a telling-off." Don't write it as a fact the patients said.

3. Say how many people it rests on

Write "eight of the patients we called," not a percentage. A percentage suggests the reading holds for all your patients. You haven't tested that yet.

4. Note the readings you rejected

Say what else you considered and why you dropped it. Maybe you wondered about cost, but no one mentioned money. This shows you looked for other answers.

A reviewer should be able to read each quote in its full context. The post on building a codebook from transcripts shows how to keep codes and quotes linked as you go.

Where Pulse fits

Latent themes need rich answers, and a first reply like "too busy" is rarely enough. Pulse runs AI-moderated voice interviews. The AI moderator builds its follow-up questions on what each person said earlier. When an answer is vague, it asks a second question. That gives you more to read under the first reason.

Pulse then groups what people said into themes, each with the quotes behind it. Every quote links to the transcript it came from. So your team can check a latent reading against the full conversation.

The researcher still decides what a theme means, most of all at the latent level. Pulse runs the interviews and the first grouping, so the researcher's time goes to that reading. To see it on your own study, book a demo, start on the $0 Explorer plan, or ask for a free study.

Questions readers ask

1. What are the themes in thematic analysis?

A theme is a pattern you see across several interviews, built around one central idea. It's more than a topic people mentioned. "Booking" is a topic. "Patients put off booking because a long gap feels embarrassing" is a theme. A good theme tells your team something it can act on.

2. Can one study use both semantic and latent themes?

Yes. Many studies code everything at the semantic level first. Then they go latent on the two or three themes that matter most for the decision. Just label each theme clearly, so readers know which ones report what people said and which ones are your reading.

3. Can ChatGPT do a thematic analysis?

AI tools can suggest codes and themes quickly. A latent reading needs more care, because it goes beyond the words. So a person should check every suggested theme against the transcripts. The post on AI in qualitative analysis covers where it helps.

4. What's the difference between thematic analysis and grounded theory?

Thematic analysis finds patterns in what people said. Grounded theory goes further and builds a new theory from the data, step by step. Most teams at work need patterns they can act on, so thematic analysis usually fits better.

Book a demo

Book a Demo

Your details

By clicking “Book Now”, you agree Pulse may use your information as described in our Privacy Notice, including to contact you about our products and services.