User research basics

In one line: good research starts by naming the question — what people say vs what they do, and whether you're exploring a problem or evaluating a solution — because that pair picks the method for you.

What it is

User research is the work of replacing your assumptions about people with evidence. It isn't one technique; it's a small kit, and the skill is choosing the right tool for the question in front of you.

Two distinctions do most of that choosing. The first: are you after attitudes (what people say they think, want, or believe) or behaviors (what they actually do)? These diverge constantly — someone swears they read every email, while the logs say they open one in ten. The second: are you exploring an open problem you don't yet understand, or evaluating a specific design to see if it works? Exploring is generative — it widens your view and generates questions. Evaluating is, well, evaluative — it tests something concrete against reality.

Cross those two axes and you get four quadrants, each with a natural fit: interviews for attitudes you're exploring, surveys for attitudes you want to measure, diary studies and field observation for behaviors you're exploring, and usability tests and analytics for behaviors you're evaluating.

Why it matters

Most wasted research is a method-to-question mismatch. Run a survey to explore a problem and you'll get tidy numbers about questions you didn't yet understand well enough to ask. Ask people in an interview what they would do and you'll collect confident predictions that the analytics later contradict — because saying and doing are different data. Naming the quadrant first stops you from buying precision you can't use or asking people to forecast their own behavior.

It also right-sizes the effort. Beginners reach for "more participants" as a proxy for rigor, but each method has a point of diminishing returns: a handful of users surfaces most severe usability issues, while a survey needs hundreds before its percentages hold still. Knowing roughly how many is enough is what separates research that informs a decision this week from research that becomes a project of its own.

See it

How it works

The matrix is two yes/no questions you answer before picking up any tool:

1. Say or do? If you need beliefs, motivations, and language, you want attitudinal data — talk to people. If you need to know what actually happens, you want behavioral data — watch or measure. When the two disagree, behavior usually wins.
2. Explore or evaluate? Early, when the problem is fuzzy, you explore with open, generative methods that widen your understanding. Later, when you have a design, you evaluate with focused methods that can fail it.
3. Read the quadrant. Attitudes + explore is an interview; attitudes + evaluate is a survey; behaviors + explore is a diary study or field observation; behaviors + evaluate is a usability test or analytics. The demo highlights the fitting cell and names a rough sample size and the bias most likely to mislead you there.

Sample size follows the method, not the other way around: qualitative methods aim for saturation (you stop when new sessions stop teaching you anything — often 5–8 people), while quantitative methods aim for stability (enough responses that the numbers don't lurch when one more arrives).

Build it

Research only works if people can answer cleanly, and that starts with the instrument. Here's a small, accessible Likert-scale question — the kind a survey or post-task questionnaire is built from — with a real <fieldset>, a <legend> that asks one thing, and radio options a keyboard and screen reader can navigate as a group.

<!-- One question, one fieldset. The legend is the question; each option is a real radio. -->
<form class="likert" novalidate>
  <fieldset class="likert__group">
    <legend class="likert__legend">
      Setting up the account was easy.
    </legend>
 
    <div class="likert__scale">
      <label class="likert__option">
        <input type="radio" name="ease" value="1" />
        <span>Strongly disagree</span>
      </label>
      <label class="likert__option">
        <input type="radio" name="ease" value="2" />
        <span>Disagree</span>
      </label>
      <label class="likert__option">
        <input type="radio" name="ease" value="3" />
        <span>Neutral</span>
      </label>
      <label class="likert__option">
        <input type="radio" name="ease" value="4" />
        <span>Agree</span>
      </label>
      <label class="likert__option">
        <input type="radio" name="ease" value="5" />
        <span>Strongly agree</span>
      </label>
    </div>
  </fieldset>
 
  <button class="likert__submit" type="submit">Submit answer</button>
</form>

.likert {
  display: grid;
  gap: 1rem;
  max-width: 30rem;
}
.likert__group {
  border: 1px solid #c9ccd4;
  border-radius: 0.5rem;
  padding: 1rem;
}
.likert__legend {
  font-weight: 600;
  padding: 0 0.25rem;
}
.likert__scale {
  display: flex;
  flex-wrap: wrap;
  gap: 0.5rem;
  margin-top: 0.75rem;
}
.likert__option {
  display: flex;
  align-items: center;
  gap: 0.4rem;
  font-size: 0.9rem;
  /* A subtle press is nice, but never required to understand the control. */
  transition: background-color 120ms ease;
  padding: 0.35rem 0.5rem;
  border-radius: 0.375rem;
}
/* The focus ring travels with the checked option — keyboard users must see selection. */
.likert__option:focus-within {
  outline: 2px solid #5b6cff;
  outline-offset: 2px;
}
.likert__submit {
  justify-self: start;
  padding: 0.6rem 1rem;
  border: 0;
  border-radius: 0.5rem;
  background: #5b6cff;
  color: #fff;
  font-weight: 600;
  cursor: pointer;
}
@media (prefers-reduced-motion: reduce) {
  .likert__option {
    transition: none;
  }
}

Make it yours

Reproduce it with an LLM

Pitfalls & accessibility

Related

Further reading

• A standard text on usability testing, for how few participants you really need and why.
• A practitioner's guide to writing survey and interview questions that don't lead the witness.
• An overview of the attitudinal/behavioral and generative/evaluative map, to place any method on the grid.