Content Analysis – Methods, Types and Examples

Content analysis is a systematic research method for examining recorded communication. Researchers define units, code relevant features, organize the codes into categories, and interpret patterns in text, images, audio, video, or digital content. It may be quantitative, qualitative, or combine numerical description with contextual interpretation.

Introduction

Researchers regularly face data that cannot be understood through numerical measurements alone. Interview transcripts, newspaper articles, policy documents, advertisements, photographs, videos, websites, discussion forums, and social-media posts all contain language, symbols, representations, and relationships that can be studied systematically.

Content analysis provides a structured way to convert this material into research evidence. Depending on the research objective, an analyst might count how often a concept occurs, compare representations across sources, examine relationships among concepts, or interpret the explicit and underlying meanings of communication.

This guide explains what content analysis is, how its major approaches relate to one another, how to develop a coding framework, how to assess reliability and trustworthiness, and how to report the method transparently. It also covers software, artificial intelligence, digital ethics, and common methodological mistakes.

Key takeaways

• Content analysis systematically examines recorded communication rather than relying on an informal reading.
• It can be qualitative, quantitative, or part of a mixed-methods design.
• Codes, categories, units, sampling rules, and interpretive procedures should be defined transparently.
• Frequency is useful only when the denominator, context, and coding rules are clear.
• Reliability or qualitative trustworthiness should be addressed in a way consistent with the study’s methodological tradition.
• Software and AI can assist analysis, but they do not replace research design, validation, interpretation, or ethical responsibility.

What Is Content Analysis?

Content analysis is a research method for making systematic and defensible inferences from recorded communication to the context in which that communication is produced, circulated, or interpreted.

The method typically involves five core activities:

1. Selecting a body of relevant material
2. Defining the units that will be examined
3. Developing codes or categories
4. Applying the coding rules consistently
5. Analysing and interpreting the coded evidence

The “content” may be written, spoken, visual, audiovisual, or digitally mediated. It can include individual words, statements, arguments, people, images, events, frames, emotions, relationships, or broader patterns of meaning.

Content analysis is more than counting words. A word-frequency table may be one component of an analysis, but a defensible study must explain why the counted terms represent the intended concept and how context, ambiguity, negation, and variation were handled.

Core Characteristics of Content Analysis

A well-designed content analysis is usually:

Systematic

The researcher follows explicit procedures for choosing material, defining units, assigning codes, and interpreting findings.

Rule guided

Coding decisions are based on operational definitions and decision rules rather than intuition alone.

Relevant to a research question

The categories and measurements should help answer a defined question or test a specified expectation.

Context sensitive

Meaning depends on who communicates, to whom, in what setting, through which medium, and for what purpose.

Transparent

Other researchers should be able to understand how the data were obtained, prepared, coded, analysed, and interpreted.

Inferential

Researchers generally want to learn something beyond the physical presence of words or images. They may infer communication priorities, representations, frames, attitudes, relationships, or changes over time.

Qualitative, quantitative, or integrated

The method can produce numerical summaries, interpretive categories, or both.

What Data Can Be Analysed?

Content analysis can be applied to any stable or retrievable form of communication relevant to the research question.

Data source	Possible research focus
Interview transcripts	Experiences, barriers, beliefs, explanations
Open-ended survey responses	Recurring concerns or categories of opinion
Newspapers	Issue attention, framing, sources, representation
Political speeches	Policy priorities, rhetoric, emotional appeals
Textbooks	Topic coverage, cultural representation, bias
Policy documents	Definitions, responsibilities, implementation priorities
Academic articles	Research trends, theories, methods, reporting practices
Advertisements	Persuasive strategies, gender roles, product claims
Photographs	People represented, setting, symbols, composition
Film and television	Character roles, themes, dialogue, visual stereotypes
Websites	Institutional priorities, accessibility, public messaging
Social-media posts	Topics, frames, sentiment, interaction, misinformation
Podcasts and videos	Arguments, speakers, themes, tone, visual or audio features
Organisational records	Decision patterns, complaints, risks, procedures

Before using digital content, researchers should determine whether the material can be collected consistently, whether deleted or edited items matter, and whether ethical or legal restrictions apply.

When Should Content Analysis Be Used?

Content analysis is appropriate when the research question concerns the characteristics, patterns, meanings, or relationships found in recorded communication.

It is particularly useful when the researcher wants to:

• Describe what is present in a body of communication
• Compare content across groups, sources, platforms, or periods
• Measure how frequently defined features occur
• Study representation, framing, tone, or emphasis
• Analyse existing records without influencing participants
• Combine structured counts with contextual interpretation
• Examine a large corpus using human or computational coding
• Reanalyse existing interviews, documents, or media archives

Content analysis may be unsuitable when the main aim is to reconstruct a person’s complete life story, examine conversational turn-taking, study language as a system of power in depth, or develop a new explanatory theory through constant comparison. Narrative analysis, conversation analysis, discourse analysis, or grounded theory may be more appropriate for those aims.

Is Content Analysis Qualitative or Quantitative?

Content analysis can be quantitative, qualitative, or integrated. The correct label depends on the research question, assumptions, coding process, form of evidence, and type of inference.

Quantitative content analysis

Quantitative content analysis converts defined characteristics of communication into numerical data.

Researchers may calculate:

• Frequencies
• Percentages
• Rates
• Differences between groups
• Changes over time
• Co-occurrences
• Associations between variables
• Statistical models using coded variables

For example, a study could code 1,000 news reports for whether they mention economic, health, security, or environmental consequences and compare the proportions across newspapers.

Qualitative content analysis

Qualitative content analysis systematically interprets meaning in context. It usually focuses on categories, patterns, similarities, differences, and levels of meaning rather than treating counts as the main result.

For example, a researcher could analyse interviews with doctoral students to identify categories describing how supervisory feedback is experienced and managed.

Integrated content analysis

A study may combine both forms. It could count the frequency of frames in news articles and then closely interpret representative examples of how each frame constructs responsibility.

The inclusion of counts does not automatically make a study quantitative, and the use of interpretation does not automatically make it qualitative. Researchers should explain which form of evidence has priority and how the quantitative and qualitative components are connected.

Types and Approaches to Content Analysis

The labels used in content analysis refer to different design dimensions. They should not be treated as a single list of mutually exclusive types.

Design question	Main options
What form will the evidence take?	Qualitative, quantitative, integrated
Where do the categories come from?	Inductive, deductive, abductive
What level of meaning is studied?	Manifest, latent
What is being examined?	Conceptual, relational
How is qualitative coding organized?	Conventional, directed, summative
What is the research purpose?	Descriptive, comparative, evaluative, explanatory, predictive

A study can occupy several positions at once. A researcher might conduct a qualitative, deductive, directed, latent, and relational analysis.

Manifest and Latent Content

Manifest content

Manifest content is the directly observable or explicitly expressed material.

Examples include:

• Whether a policy mentions climate change
• How many women appear in an advertisement
• Whether a news story quotes a government official
• The number of references to cost
• Whether an image contains a national flag

Manifest coding usually requires less interpretation, although decisions about definitions and boundaries remain important.

Latent content

Latent content concerns implied, underlying, symbolic, or contextual meaning.

Examples include:

• Whether an advertisement presents success as individual or collective
• How a policy constructs responsibility
• Whether a news report implies that a group is threatening
• What an interview account suggests about institutional trust
• How visual composition positions one person as more powerful than another

Latent analysis can generate richer interpretations, but it requires a clearly stated interpretive framework, careful attention to context, and stronger reflexive documentation.

Can a study use both?

Yes. Researchers may begin with manifest features and then interpret their latent significance. The report should distinguish observation from interpretation.

Inductive, Deductive, and Abductive Content Analysis

Inductive content analysis

In inductive analysis, codes and categories are developed primarily from repeated engagement with the data.

It is useful when:

• Existing theory is limited
• The phenomenon is poorly understood
• Researchers want categories that remain close to participants’ accounts
• Unexpected patterns are important

The researcher still brings prior knowledge to the analysis. “Inductive” does not mean that interpretation occurs without assumptions.

Deductive content analysis

Deductive analysis begins with concepts derived from a theory, model, previous study, policy framework, or research question.

It is useful when researchers want to:

• Apply an established framework
• Compare a new setting with previous research
• Test whether known categories appear
• Evaluate compliance with criteria
• Examine how a theoretical construct is represented

A deductive framework should allow researchers to record material that does not fit the initial categories. Otherwise, the analysis may force the evidence into the theory.

Abductive content analysis

Abductive analysis moves between data and theory. Researchers begin with sensitising concepts, examine surprising observations, revise categories, and seek the most plausible interpretation.

This is often a realistic description of applied research because coding frameworks are frequently refined through pilot analysis.

Conventional, Directed, and Summative Content Analysis

Hsieh and Shannon distinguish three influential qualitative approaches.

Conventional content analysis

Categories are developed mainly from the data. This approach is suitable when existing theory or research is limited.

Directed content analysis

Initial codes are derived from theory or previous findings. The purpose is often to extend, refine, or examine an existing framework.

Summative content analysis

The researcher begins by identifying and comparing selected words or content and then interprets their contextual use. It is not merely a word count; the analysis investigates how the terms function within communication.

Conceptual and Relational Content Analysis

Conceptual analysis

Conceptual analysis examines the presence, absence, frequency, or distribution of defined concepts.

A researcher may ask:

• How often is “equity” mentioned?
• Which types of risk appear?
• What proportion of documents includes a complaints procedure?
• Which speakers are quoted?

Relational analysis

Relational analysis investigates connections among concepts.

A researcher may ask:

• Which concepts appear together?
• Is immigration associated with security or employment?
• Which actors are linked with responsibility?
• How are causes connected with proposed solutions?
• Which emotions are associated with particular groups?

Relational analysis may use co-occurrence tables, semantic networks, proximity measures, or qualitative interpretation of relationships.

Units in Content Analysis

Clear unit definitions prevent inconsistent coding and misleading denominators.

Unit	Meaning	Example
Sampling unit	The entity selected into the study	A newspaper issue, interview, policy document, video, or social-media account
Recording or coding unit	The segment to which a code is assigned	A sentence, paragraph, speaking turn, image, scene, post, or entire document
Context unit	The surrounding material a coder may examine to interpret a coding unit	The full article surrounding a paragraph
Enumeration unit	The basis on which occurrence is counted	Presence per document, number of mentions, duration in seconds
Unit of analysis	The entity about which conclusions are ultimately drawn	Newspaper, organisation, speaker, platform, country, or time period

These units may differ. A study might sample complete news articles, code individual paragraphs, read the full article as context, count whether each frame appears at least once, and compare results at newspaper level.

How to Conduct Content Analysis: Ten Steps

Step 1: Define the Research Question

A good question specifies:

• The communication being studied
• The population, setting, or context
• The feature or meaning of interest
• Any comparison or time period

Too broad: How does the media discuss education?

More focused: How did three national newspapers frame university tuition increases during the six months surrounding the policy announcement?

The question determines the data, units, categories, and analytical method.

Step 2: Define the Corpus or Communication Universe

Specify the complete body of material to which the research question refers.

Document:

• Platforms or publications
• Geographic boundaries
• Languages
• Dates
• Content formats
• Search terms
• Inclusion criteria
• Exclusion criteria
• Duplicate-handling rules
• Procedures for unavailable or deleted material

A corpus assembled through keywords reflects the search strategy. It should not automatically be treated as a complete representation of all relevant communication.

Step 3: Choose a Census or Sampling Strategy

A census analyses every eligible item. This may be feasible for a small archive but impractical for large digital datasets.

Possible sampling strategies include:

Simple random sampling

Each eligible item has a known chance of selection.

Systematic sampling

Items are selected using a regular interval after a starting point has been chosen.

Stratified sampling

The corpus is divided into meaningful groups, such as year, newspaper, country, or platform, and items are sampled within each group.

Cluster sampling

Naturally occurring groups, such as publication dates, classrooms, channels, or organisations, are sampled before items within them are analysed.

Purposive or relevance sampling

Items are selected because they meet theoretically or practically relevant criteria.

The sample size should be justified by the intended form of inference. Quantitative studies may require statistical precision and adequate cases for subgroup comparisons. Qualitative studies require sufficient information diversity and analytical depth. There is no universal number of documents that is appropriate for every content analysis.

Step 4: Define the Units

State:

• What enters the sample
• What receives a code
• How much surrounding material coders may inspect
• What counts as one occurrence
• At what level comparisons will be made

Test the unit definitions on real data. Sentences may be easy to identify but can split a single idea. Paragraphs preserve more context but may contain several ideas. Meaning units can be analytically useful but require interpretive boundary decisions.

Step 5: Select the Analytical Approach

Decide whether the study is primarily:

• Qualitative, quantitative, or integrated
• Inductive, deductive, or abductive
• Manifest, latent, or both
• Conceptual, relational, or both

Explain why the selected approach answers the research question better than alternative methods.

Step 6: Develop the Coding Framework

A code is a label assigned to a relevant feature of the material.

A category groups related codes or observations at a more general level.

A theme usually expresses a broader pattern of meaning that connects multiple observations through a central idea. The term is used differently across qualitative traditions, so researchers should define it rather than assume universal agreement.

A strong coding framework should be:

• Relevant to the question
• Conceptually coherent
• Clear enough to apply
• Detailed enough to distinguish similar categories
• Flexible enough to capture important unanticipated material
• Consistent with the coding unit
• Appropriate to the required level of measurement

Content-Analysis Codebook Template

Field	What to record
Code or variable name	A short, consistent label
Conceptual definition	What the concept means theoretically
Operational definition	How the concept will be recognised in the data
Unit	The segment to which the code applies
Values	Binary, nominal, ordinal, count, or open text
Inclusion criteria	Features that qualify
Exclusion criteria	Similar features that do not qualify
Positive example	A clear example that receives the code
Negative example	A similar example that does not receive it
Borderline rule	How ambiguous cases are decided
Multiple coding rule	Whether more than one code may apply
Missing-value rule	How unavailable or unclear information is recorded
Source	Theory, previous study, pilot data, or research question
Revision date	When and why the rule changed

Example codebook entry

Code: Financial barrier
Definition: A statement indicating that money, price, fees, income, or affordability restricts access.
Include: Tuition costs, transport costs, inability to pay, loss of earnings.
Exclude: General inconvenience without a financial element.
Unit: Sentence or meaning unit.
Multiple coding: May overlap with transport barrier when cost of transport is explicitly mentioned.
Example: “I could not attend because the bus fare was too expensive.”

Step 7: Pilot the Codebook and Train Coders

Pilot coding should occur before the complete dataset is coded.

A pilot helps identify:

• Ambiguous definitions
• Missing categories
• Overlapping categories
• Units that are too large or too small
• Coding rules that require unavailable information
• Categories that almost never occur
• Differences in coder interpretation

A practical pilot process is:

1. Select a varied subset of the corpus.
2. Have coders apply the draft codebook independently where reliability is relevant.
3. Compare disagreements.
4. Diagnose whether disagreement comes from definitions, unit boundaries, coder training, or genuine ambiguity.
5. Revise the codebook.
6. Repeat the pilot when revisions are substantial.
7. Freeze or formally version the framework before final coding.

Do not improve agreement merely by discussing each final case until coders concur. Agreement calculated after joint resolution does not estimate independent reproducibility.

Step 8: Assess Reliability or Qualitative Trustworthiness

Percent agreement

Percent agreement is:

[
\text{Percent agreement} =
\frac{\text{Number of agreements}}
{\text{Total coding decisions}} \times 100
]

It is easy to understand but does not adjust for agreement that may occur by chance.

Cohen’s kappa

Cohen’s kappa is commonly used for two coders assigning categorical values under suitable conditions.

[
\kappa =
\frac{P_o-P_e}{1-P_e}
]

Where:

• (P_o) is observed agreement
• (P_e) is expected agreement under the coefficient’s chance model

Kappa can behave unexpectedly when categories are highly unbalanced, so the category distribution and confusion matrix should also be examined.

Krippendorff’s alpha

Krippendorff’s alpha has the general form:

[
\alpha = 1-\frac{D_o}{D_e}
]

Where:

• (D_o) is observed disagreement
• (D_e) is disagreement expected under the coefficient’s model

Alpha can be adapted to nominal, ordinal, interval, or ratio data and is often selected when there are multiple coders or some missing ratings.

A coefficient should not be chosen only because software provides it. Researchers should justify why it matches the number of coders, level of measurement, missing-data structure, and coding design.

Reliability is not only a coefficient

Researchers should also report:

• Number of coders
• Coder qualifications
• Training procedure
• Pilot size
• Whether units were independently identified
• Variables tested
• Coefficient and measurement level
• Confidence intervals where available
• Category distributions
• How disagreements were resolved
• Whether the final corpus was single- or double-coded

Qualitative trustworthiness

Some qualitative content-analysis traditions do not treat independent coding agreement as the primary quality criterion. Depending on the epistemological position, researchers may instead or additionally demonstrate:

• Credibility: Interpretations are well supported by data and context.
• Dependability: Analytical decisions and changes are documented.
• Confirmability: Conclusions can be traced to evidence rather than hidden assumptions.
• Transferability: Readers receive enough contextual detail to assess relevance elsewhere.
• Reflexivity: The researcher explains how their position and assumptions shaped the analysis.
• Negative-case analysis: Material that challenges the developing interpretation is examined.
• Peer discussion: Interpretations are critically reviewed rather than accepted automatically.

The quality strategy should fit the claimed form of knowledge.

Step 9: Code and Analyse the Full Dataset

Apply the finalized procedures consistently and preserve an audit trail.

For quantitative content analysis, common summaries include:

Frequency

[
f_c = \text{Number of times category }c\text{ is recorded}
]

Percentage of eligible units

[
\text{Percentage}_c =
\frac{\text{Units coded with }c}
{\text{Total eligible units}} \times 100
]

Rate per standard amount

[
\text{Rate}_c =
\frac{\text{Occurrences of }c}
{\text{Total words, minutes, posts, or other exposure}}
\times k
]

The constant (k) might be 100, 1,000, or another interpretable amount.

Rates can be more informative than raw counts when documents differ substantially in length.

Co-occurrence

Two codes co-occur when both appear within a defined window, such as the same sentence, paragraph, article, post, or scene.

The window must be defined before interpreting the relationship. Co-occurrence does not by itself establish causation or conceptual dependence.

Comparative analysis

Researchers may compare coded variables by:

• Source
• Time
• Country
• Platform
• Author or speaker
• Audience
• Document type
• Participant characteristics

Inferential statistics may be used when the sampling design and measurements support them.

For qualitative content analysis, the process may involve:

1. Reading coded extracts in context
2. Comparing similarities and differences
3. Condensing meaning units
4. Grouping related codes
5. Developing categories or themes
6. Examining contradictory cases
7. Moving between parts and the full dataset
8. Writing analytic memos
9. Selecting evidence that illustrates variation
10. Relating interpretations to the research question and literature

Software retrieves and organizes coded material; it does not decide what the material means.

Step 10: Interpret and Report the Findings

Interpretation should distinguish:

• What was directly observed
• What was calculated
• What was inferred
• Which alternative interpretations remain possible

Avoid treating the most frequent category as automatically the most important. A rare feature may be theoretically significant, and an absent feature may be meaningful when a strong expectation supports that interpretation.

Conclusions should remain within the limits of:

• The sample
• The selected platforms or sources
• The time period
• The coding framework
• The quality of the available data
• The reliability or trustworthiness evidence
• The assumptions connecting content to context

Worked Example of Content Analysis

The following is a hypothetical example and does not report real findings.

Research question

How do public universities describe acceptable and unacceptable student use of generative AI in publicly available academic-integrity guidance?

Corpus

The researcher identifies public universities in three countries and collects the central academic-integrity or generative-AI guidance page available on a specified date.

Inclusion criteria

• Official university webpage
• Publicly accessible
• Addresses student academic work
• Published or active during the study period
• Available in English or translated through a documented procedure

Sampling strategy

Universities are stratified by country and institution size. A random sample is selected within each stratum.

Sampling unit

One official guidance document or webpage.

Coding unit

A complete policy statement or bullet point expressing one rule, permission, responsibility, or consequence.

Context unit

The complete guidance page and any directly linked definitions page.

Deductive codes

Initial categories are developed from academic-integrity principles:

• Disclosure required
• Instructor permission required
• AI use prohibited
• AI use conditionally permitted
• Verification responsibility
• Citation or acknowledgement
• Privacy warning
• Fabrication warning
• Assessment-specific variation
• Penalty or misconduct consequence

Inductive provision

Coders may flag statements that do not fit the initial framework. These are reviewed during the pilot and may generate additional categories.

Pilot

Two coders independently code a varied subset. They discuss unclear distinctions, such as the difference between “disclosure” and formal “citation,” revise the definitions, and repeat the pilot.

Quantitative analysis

The researcher calculates the percentage of institutions whose guidance contains each category and compares countries.

Qualitative analysis

The researcher interprets how responsibility is framed. For example, policies may frame AI primarily as a misconduct risk, as a learning tool requiring transparency, or as a technology whose acceptability depends on instructor-defined context.

Reporting

The methods section states the search date, sampling frame, document-retrieval procedure, codebook development, pilot, reliability statistic, coding software, and comparison method. The results separate prevalence estimates from interpretive observations.

Content Analysis Versus Related Methods

Method	Main focus	Typical output	Best suited to
Content analysis	Systematic categorization of communication and, where relevant, counting	Categories, frequencies, comparisons, interpretations	Describing and comparing recorded content
Thematic analysis	Patterns of shared meaning across a qualitative dataset	Interpretive themes	Experiences, perceptions, and patterned meanings
Discourse analysis	How language constructs social reality, identities, and power	Interpretive account of discourse and its effects	Language, ideology, institutions, and power
Narrative analysis	How stories are structured, performed, and made meaningful	Narrative types or detailed case interpretations	Life stories, identity, sequence, and storytelling
Textual analysis	Close interpretation of the meanings and features of particular texts	Contextual critical interpretation	Cultural, literary, media, or visual texts
Framework analysis	Matrix-based comparison using an analytic framework	Case-by-theme matrix	Applied policy, health, and evaluation research

Content Analysis Versus Thematic Analysis

Content analysis usually provides a more structured categorization of communication and may quantify the occurrence of categories. Thematic analysis primarily develops patterns of shared meaning that answer a qualitative research question. However, qualitative content analysis and some forms of thematic analysis can overlap substantially.

The distinction should not be reduced to “content analysis counts words while thematic analysis interprets meaning.” Qualitative content analysis may examine latent meaning, and thematic analysis also involves systematic coding.

Choose content analysis when:

• The communication itself is the primary object of study
• Comparability across documents or cases is important
• Categories need clear operational definitions
• Presence, absence, or frequency matters
• A structured codebook is central
• Numerical and qualitative findings may be combined

Choose thematic analysis when:

• The aim is to interpret patterned meanings across accounts
• Participant perspectives are the main focus
• Themes need to express a central organizing concept
• Quantification is secondary or unnecessary
• A flexible qualitative approach best fits the question

Advantages of Content Analysis

It can examine naturally occurring material

Researchers may analyse communication that was produced independently of the study.

It supports historical and longitudinal analysis

Archived materials can be compared across periods when collection procedures and source availability are documented.

It handles multiple media

Text, image, audio, video, and digital interaction can be analysed individually or together.

It can combine breadth and depth

A study can summarize a large corpus and closely interpret selected material.

It supports comparison

A common coding framework can be applied across groups, institutions, countries, sources, or periods.

It can be unobtrusive

Researchers can study existing records without introducing interviewer or observer effects into their production.

It can be transparent and auditable

A documented corpus, codebook, coding records, and analysis script allow readers to inspect the research process.

Limitations of Content Analysis

It is vulnerable to decontextualization

Breaking communication into units can separate statements from their narrative, historical, interpersonal, or institutional context.

Coding is not automatically objective

Categories are designed and interpreted by researchers. Even apparently simple manifest codes depend on definitions.

It may confuse frequency with significance

Common content is not necessarily influential, sincere, accurate, or important.

Source material may be incomplete

Digital content may be deleted, edited, personalized, inaccessible, or unevenly archived.

Meaning may vary across groups

The same word, image, gesture, or symbol can have different meanings in different communities.

Latent coding can be difficult to reproduce

Interpretations of implication, ideology, emotion, or symbolism may vary among coders.

Large projects require substantial preparation

Corpus construction, cleaning, codebook development, training, pilot coding, adjudication, and documentation can require more time than expected.

Content cannot always reveal effects

A study of messages cannot by itself demonstrate how audiences understood them or whether the messages changed behaviour. Audience research or experimental evidence may be needed.

Common Mistakes in Content Analysis

1. Starting without a focused research question

Collecting thousands of documents before defining the analytical purpose creates an unmanageable dataset.

2. Calling an informal reading “content analysis”

A research method requires a documented selection and analysis procedure.

3. Treating keyword frequency as meaning

Keywords may be ambiguous, negated, ironic, quoted, or used in unrelated contexts.

4. Failing to define the corpus

Readers need to know what could have been selected as well as what was selected.

5. Confusing units

Sampling articles but reporting paragraph percentages without explaining the denominator makes results difficult to interpret.

6. Using overlapping codes without decision rules

Coders cannot work consistently when categories are conceptually indistinguishable.

7. Changing the codebook silently

Revisions should be dated, justified, and applied consistently to material coded under earlier versions.

8. Reporting only a reliability number

The coefficient is uninterpretable without the variables, coder number, pilot design, data level, and category distribution.

9. Assuming software performs the analysis

Software stores, searches, retrieves, counts, and visualizes coded material. Researchers remain responsible for concepts and inferences.

10. Generalising beyond the sample

A purposive sample of popular posts cannot support claims about all users, all communication, or the wider population without additional justification.

11. Ignoring absence and visibility mechanisms

What appears online may reflect algorithms, moderation, deletion, access restrictions, or platform design.

12. Using AI output without validation

A plausible explanation generated by a model is not evidence that its coding is accurate, stable, unbiased, or appropriate for the research context.

Content Analysis in Modern Research

Contemporary content analysis increasingly involves:

• Large digital corpora
• Multimodal data
• Social-media archives
• Automated transcription
• Natural language processing
• Dictionary-based coding
• Supervised machine learning
• Topic modelling
• Network analysis
• Image and video classification
• Human–AI collaborative coding
• Reproducible scripts and version-controlled codebooks

These methods expand analytical capacity but introduce new sources of error. Automated systems may reproduce biases in training data, misclassify unusual language, lose cultural context, or change when models and software versions are updated.

The basic methodological questions remain the same:

• What construct is being measured?
• What material represents the target communication?
• What is the unit?
• How was the coding rule validated?
• How consistent is the coding?
• What inference is justified?
• Can another researcher inspect or reproduce the process?

Software for Content Analysis

Tool type	Examples	Appropriate uses	Important limitation
Spreadsheet	Excel, Google Sheets, LibreOffice Calc	Small structured projects, codebook development, binary or categorical coding	Difficult to manage large, complex, or multimedia corpora
Qualitative analysis software	NVivo, MAXQDA, ATLAS.ti	Manual coding, memos, retrieval, multimedia, queries, team projects	Software does not establish methodological quality
Statistical software	R, Python, SPSS, Stata	Frequencies, cross-tabulations, reliability, regression, reproducible analysis	Requires suitable data structure and technical skill
Text-analysis software	quanteda and other NLP libraries	Tokenization, dictionaries, document-feature matrices, modelling	Automated features may not represent contextual meaning
AI-assisted tools	LLM-based coding and summarisation systems	Draft coding, classification, retrieval, comparison, exploratory support	Privacy, variability, hallucination, bias, and weak reproducibility require controls

Tool selection should follow the method, not determine it.

Can Artificial Intelligence Conduct Content Analysis?

AI can assist content analysis by proposing codes, classifying passages, summarising coded material, retrieving examples, or applying a deductive codebook. It should not be treated as an autonomous replacement for research design, validation, contextual interpretation, or accountability.

A defensible AI-assisted workflow should include:

1. Define the research question and codebook independently of model marketing claims.
2. Remove or protect confidential and identifying data.
3. Record the model, provider, version, date, settings, and interface.
4. Preserve the exact prompts and output format.
5. Give the model operational definitions, inclusion rules, exclusions, and examples.
6. Test the procedure on a human-coded validation set.
7. Evaluate results separately for each category.
8. Examine false positives, false negatives, and uncertain cases.
9. Repeat selected prompts to assess output variability where the system is nondeterministic.
10. Require human review for ambiguous, latent, culturally specific, or high-stakes decisions.
11. Freeze the tested workflow before analysing the final dataset.
12. Report which tasks were performed by humans, software, and AI.
13. Preserve an audit trail where licensing and privacy rules permit.
14. Avoid uploading sensitive material to a system unless institutional approval and data-processing terms allow it.

An overall agreement score may hide serious failure in rare but important categories. Category-level precision, recall, confusion patterns, and qualitative error analysis may therefore be necessary for automated classification.

Ethical Considerations

Public does not always mean ethically unrestricted

Publicly visible content may still involve vulnerable individuals, sensitive experiences, or expectations of limited visibility.

Protect identities

Direct quotations can sometimes be located through search engines. Paraphrasing, masking details, or obtaining permission may be necessary.

Review consent requirements

Interview transcripts, private groups, classroom data, patient records, and organisational documents may require consent, institutional approval, or legal authorization.

Follow platform and archive rules

Researchers should examine terms of service, access restrictions, copyright, and data-use conditions.

Minimise data collection

Collect only the information needed to answer the research question.

Protect data sent to AI systems

Confidential or identifiable data should not be entered into third-party models without an approved legal, ethical, and security basis.

Represent communities carefully

Categories can stigmatise people when they are developed without cultural knowledge or applied without context.

How to Write Content Analysis in a Methodology Section

A complete methods section should report:

1. Research design and rationale
2. Research question
3. Data source and communication universe
4. Search or retrieval strategy
5. Dates of collection
6. Inclusion and exclusion criteria
7. Sampling method and sample size
8. Sampling, coding, context, and analysis units
9. Qualitative, quantitative, or integrated approach
10. Inductive, deductive, or abductive category development
11. Codebook sources and revisions
12. Number and qualifications of coders
13. Training and pilot procedures
14. Reliability or trustworthiness strategy
15. Software and version
16. Statistical or interpretive analysis
17. Ethics and data protection
18. Procedures for quotations and anonymisation
19. AI involvement, prompts, model details, and validation where applicable

Example methodology paragraph

A quantitative content analysis was conducted to examine how official university guidance described student use of generative AI. The sampling frame consisted of publicly accessible academic-integrity guidance published by eligible institutions on the collection date. Universities were stratified by country and size, after which institutions were randomly selected within each stratum. The sampling unit was the official guidance document, while the coding unit was an individual policy statement. A deductive codebook was developed from academic-integrity principles and refined through independent pilot coding. Two trained coders coded the reliability subset, and agreement was assessed separately for each nominal variable using an appropriate chance-corrected coefficient. The final analysis compared category prevalence across institutional groups and interpreted selected policy statements in their full-document context.

Replace the generic wording with the actual procedures and results of the study.

How to Present Content-Analysis Results

Quantitative results may be presented through:

• Frequency tables
• Percentages with explicit denominators
• Cross-tabulations
• Time trends
• Bar charts
• Co-occurrence matrices
• Network diagrams
• Regression models
• Confidence intervals or uncertainty estimates

Qualitative results may be presented through:

• Categories and subcategories
• Analytic themes where methodologically appropriate
• Definitions of each category
• Representative extracts
• Contrasting or negative cases
• Interpretive commentary
• Relationships among categories
• A conceptual model grounded in the evidence

Do not use quotations as decoration. Explain how each extract supports, complicates, or limits the interpretation.

Content-Analysis Quality Checklist

Before submission, confirm that:

• The research question is specific.
• The corpus is clearly bounded.
• Retrieval procedures are reproducible.
• Inclusion and exclusion criteria are stated.
• The sampling strategy is justified.
• All units are defined.
• The analytical tradition is named.
• Codes have operational definitions.
• Inclusion and exclusion rules are available.
• Pilot coding is documented.
• Codebook changes are traceable.
• Reliability or trustworthiness is addressed.
• Counts have clear denominators.
• Interpretations are supported by contextual evidence.
• Contradictory cases are considered.
• Software and versions are reported.
• AI use is disclosed and validated.
• Ethical and privacy issues are addressed.
• Conclusions do not exceed the sample or data.
• Limitations are stated directly.

Conclusion

Content analysis is a flexible but rigorous method for studying recorded communication. Its quality depends less on the number of documents or sophistication of the software than on the alignment among the research question, corpus, units, coding framework, reliability or trustworthiness procedures, analysis, and inference.

Researchers should not reduce content analysis to counting words or sorting quotations. When sampling decisions, codes, context, uncertainty, ethics, and interpretation are documented transparently, the method can support credible qualitative, quantitative, and integrated research across many disciplines.