Code Interview Notes Without False Trends

Coding interview notes can make a large set of observations easier to review. It can also create false confidence. A label such as “setup difficulty” may look like a trend after it appears beside three excerpts, even when those excerpts describe different tasks, different moments, or different meanings.

The purpose of coding is not to make the notes look quantitative. It is to create a traceable map between a question, the permitted source material, and the interpretations a team may want to examine. A useful coding pass keeps the original context close, permits disagreement, and makes gaps visible.

Begin with the decision the coding should support

Write one narrow analysis question before creating any codes. For example:

Analysis question:
Which parts of the fictional community workshop registration process caused participants to pause, seek help, or change their intended action?

This question is specific enough to guide the work without assuming that every pause has the same cause. It also helps exclude interesting comments that do not answer the current question.

Build an excerpt table before a theme list

Use one row per meaningful excerpt or observation. Keep a stable source key, a short paraphrase, the relevant context, and the exact location of the source. The team members reviewing the synthesis should be able to return to the permitted original material.

Source key	Short excerpt	Context	Initial code	Evidence note
P01-L08	“I am not sure which session this form is for.”	Before choosing a date	session identity unclear	Direct statement
P02-L11	Asked whether the confirmation was the ticket	After submission	confirmation meaning unclear	Observer note; wording not verbatim
P03-L05	Returned to the event page to check the address	While completing the form	location information missing	Observed action; motive uncertain
P04-L09	“The first date works for me.”	Immediately selected a session	no observed selection difficulty	Counterexample to a broad selection problem
P05-L14	Paused beside the accessibility field	No follow-up question was asked	unexplained pause	Do not infer the reason
P06-L07	“Can I change this after I submit?”	Reviewing contact details	editability unclear	Direct question

This fictional set contains six different pieces of evidence. It does not yet contain a single trend.

Define each code narrowly

A code should say what evidence belongs under it and what does not. Without a definition, reviewers may apply the same label to unrelated excerpts.

Code: confirmation meaning unclear
Include: a direct question or observed misunderstanding about what the confirmation represents.
Exclude: a general request to edit submitted information.
Evidence needed: direct words, recorded action, or an observer note with context.
Boundary: does not prove the confirmation design caused the misunderstanding.

The boundary is as important as the label. It prevents “confirmation meaning unclear” from becoming “the confirmation page is bad” before the evidence supports that interpretation.

Allow overlapping codes when the excerpt supports them

One excerpt can answer more than one part of the research question. If a participant returns to the event page and asks whether the address changes by session, the same excerpt may support both “location information missing” and “session differences unclear.”

Do not force every row into one exclusive bucket merely to make counting easier. Record the overlap and explain it. Otherwise, the coding scheme may hide the relationship between two difficulties that occur together.

Do not code a motive that was never observed

P05-L14 records a pause beside an accessibility field. The notes do not record why the participant paused. Coding it as “privacy concern” or “accessibility wording unclear” would add a motive that is not in the source.

Use a code such as “unexplained pause,” then write the missing follow-up question:

Unknown: reason for the pause
Next check: ask what the participant expected at this field and what made them stop

Unknown is a valid evidence state. It is more useful than a confident label built from guesswork.

Keep counterexamples in the main table

P04-L09 is not noise. The participant chose a date immediately and stated that the first date worked. This excerpt limits a possible claim that “people cannot choose a session.”

The counterexample may point to a useful distinction: some participants know their preferred date before opening the form, while others need to compare session details. That distinction can guide a better follow-up question, but it remains an interpretation until checked.

Count coded excerpts only as a description of the material reviewed

A coding table can report that a code appeared in three of the six excerpts reviewed. That number describes the table. It does not establish that half of all participants experience the issue.

Mention counts can be distorted by interview prompts, note-taking detail, repeated comments from one person, overlapping codes, and differences in the task each participant completed. Use wording such as:

Three source-linked excerpts in this six-excerpt review were coded as session-information questions.

Avoid:

Session information is a major user trend affecting 50% of users.

The second statement reaches beyond the evidence.

Revise the codebook when a label is doing too much

Suppose the first pass uses one code, “missing information,” for the session name, address, confirmation state, and editability. The label groups the excerpts, but it does not help the team decide what to check next.

Split it into narrower codes only when the source material supports the distinction:

session identity unclear;
location information missing;
confirmation meaning unclear;
editability unclear.

Record the revision date and recode earlier rows. Otherwise, the same label may mean one thing in the first half of the table and another in the second.

Use a second pass to test consistency

After the initial coding, reopen each cited source and ask:

Does the excerpt actually meet the code definition?
Did the paraphrase preserve the source meaning?
Was relevant context removed?
Does an overlapping code belong here?
Is there a counterexample or unknown that changes the interpretation?
Would another authorized reviewer understand why this code was applied?

If two reviewers disagree, preserve the disagreement long enough to identify its source. The code may be vague, the excerpt may lack context, or the reviewers may be answering different questions.

Move from codes to a bounded finding

A finding should combine the evidence, its limits, and the next check. For the fictional dataset, a bounded finding might read:

Four different information questions appeared across the six excerpts reviewed, but each concerned a different stage of registration. One participant showed no observed difficulty selecting a session, and one pause remains unexplained. The next review should test the meaning of the confirmation and ask explicitly about the accessibility field rather than treating all pauses as one trend.

This finding is more useful than “registration is confusing.” It shows what was observed, what did not fit, and what evidence is still missing.

Connect the coding table to the next research artifact

When several well-supported codes point to a shared difficulty, use them to draft a narrow problem statement from interview notes. Keep the cited excerpts attached so the problem statement can be checked against its source.

If the evidence comes from observed behavior rather than interview answers, separate observation from interpretation using an observation brief. The two evidence types can inform each other, but they should not be merged without labels.

Copy this interview-note coding template

Analysis question:
Permitted source set:

Source key:
Excerpt or observation:
Context:
Code or codes:
Evidence state: direct words / observed action / paraphrase / unknown
Reason the code applies:
Counterexample or conflicting evidence:
Missing follow-up:

Code definition:
Include:
Exclude:
Boundary:
Revision date:

Bounded finding:
Evidence reviewed:
Limit:
Next check and owner:

Run the final evidence check

Before sharing the synthesis, reopen every cited excerpt. Confirm the source key, context, code definition, and evidence state. Check that an unknown has not become a motive, a mention count has not become prevalence, and a counterexample has not been removed because it complicated the theme.

Choose six excerpts from interview material you are authorized to analyze. Apply narrow, source-linked codes, allow overlap where the evidence supports it, preserve one counterexample and one unknown, then write a bounded finding that states what the notes can and cannot show.