Introduction
Content analysis is the method for turning open-ended feedback (survey verbatims, support tickets, app reviews, interview transcripts, video answers) into categories that can be counted and compared: how many respondents mentioned pricing, which complaint grew this quarter, what share of reviews raise reliability. It is where qualitative material meets quantitative discipline, and it is increasingly machine-assisted, which raises the bar for validation rather than lowering it. This article covers how content analysis works on product feedback, the manifest-versus-latent distinction, the codebook and reliability machinery that make the counts trustworthy, and how it pairs with thematic analysis.
What is Content Analysis?
Content analysis is a method for systematically categorizing text and other communication using explicit coding rules, so that the material's features can be quantified: counted, compared across groups and time, and tracked. In product research its raw material is the open-ended feedback every team accumulates (the comment box after a rating, the cancellation reason, the ticket, the review, the transcribed video answer from a study), and its output is a coded dataset in which each response carries one or more category labels from a defined codebook. Its lineage runs through mid-century communication research, where it was formalized for studying newspapers and propaganda, and its defining trade is stated up front: reduce rich material to categories, and in exchange gain counts that support comparison at scales close reading can't reach. The trade only pays if the categories are defined precisely and applied consistently, which is what separates content analysis from reading feedback with a highlighter.
Manifest and Latent Content
The method's central distinction. Manifest content is what's literally present and countable with minimal judgment: the word appears, the feature is named, the ticket mentions a price. Latent content is the meaning beneath the surface (frustration, sarcasm, an implied comparison), requiring interpretation that coders can reasonably disagree about. Manifest coding is more reliable and shallower; latent coding is richer and riskier; rigorous studies state which they're doing, and build the reliability checks to match, because "systematic" is precisely what separates content analysis from reading with a highlighter.
The Procedure
1. Define the question, the universe, and the unit.
What's being asked, of which body of material (all Q3 tickets? a defined sample of reviews?), coded at what unit (word, sentence, ticket, review)? Sampling from large corpora follows the same logic as sampling people.
2. Build the codebook.
Categories with names, definitions, inclusion and exclusion rules, and example cases: the instrument itself. Codes can come from theory and prior research (deductive) or from a first inductive pass over a subsample, then get frozen for the systematic run.
3. Pilot, train, and measure agreement.
Independent coders on a shared subsample, disagreements resolved into codebook refinements, and inter-coder reliability reported (Cohen's kappa and relatives): the step that converts opinion into measurement, and the first thing to check in anyone else's content analysis.
4. Code the corpus, then count and compare.
Frequencies, proportions, co-occurrences, and trends over time or across segments, analyzed with the standard categorical machinery where inference is claimed.
Content Analysis in Product Research
The product-side corpora are everywhere: open-ended survey answers coded into issue categories and tracked wave over wave; support tickets and app-store reviews quantified into a pain-point league table; competitor sites and onboarding flows coded for claims and features; and open-text or transcribed video answers from studies (the qualitative half of a mixed Ballpark study) converted into countable categories alongside their task metrics. Automation extends reach: keyword rules and machine-assisted classification (including sentiment analysis, essentially automated latent coding of tone) can process volumes no human team can, with the standing requirement that automated codes be validated against human judgment on samples before anyone trusts the dashboard.
Against its neighbors: thematic analysis pursues interpreted meaning and pattern, comfortable with themes that resist counting; content analysis pursues systematic enumeration, comfortable with the reduction that counting requires. Mature teams use both, often on the same material: content analysis for the "how much", thematic work for the "what's really going on".
What to Remember
Content analysis is measurement applied to words: a defined corpus, a disciplined codebook, agreement you can quantify, and counts you can compare. Declare the manifest-latent line, earn the reliability statistic, validate any automation, and let the enumerable questions be answered at scale, while remembering what was traded for the counts: everything about the material that a category can't hold.
Further reading
For the method's procedure and standards:
Articles:
1. Content Analysis: Guide, Methods and Examples - Scribbr
The full procedure (units, codebooks, reliability, analysis) with worked examples across material types.
2. How to Analyze Qualitative Data: Thematic Analysis - Nielsen Norman Group
The interpretive neighbor, for choosing between counting and meaning-making, or sequencing both.