Research & Validation · Field Guide

Research Tactics

Not all evidence is created equal — some methods tell you what people say, a few catch what they actually do, and the best are hard to fake at any price. Here are two dozen ways to learn from the people you're designing for, seen through two lenses: the 2×2 sorts them by the kind of question each answers; the effort-vs-believability map shows what each one costs against what its evidence is worth. Flip between the graphs — your selection stays put, and the detail below reads it from both angles.

Sorted by the kind of question each method answers — qualitative vs. quantitative, and what people say vs. what they do.

Attitudinalwhat people say
Behavioralwhat people do
Qualitativewhy & how to fix
Quantitativehow many

Usability Test — Qualitative, Behavioral; effort High

Usability Test

QualitativeBehavioral
EffortHighBelievability
LowHigh
What it is
Watch representative users attempt real tasks to find where the product helps and where it gets in the way.
How it works
Set concrete tasks, observe, and note success, errors, and friction — moderated in a session or run unmoderated.
Best for
diagnosing task-level problems in a design.
How believable it is
Good. You watch real people attempt real tasks, so you see genuine behavior and friction — the workhorse of trustworthy design evidence, short of shipping.

Expert / Heuristic Reviews

QualitativeAttitudinal
EffortLowBelievability
LowHigh
What it is
A usability specialist evaluates an interface against established principles to flag likely problems — no users required.
How it works
Reviewers inspect the product screen by screen against a heuristics checklist, then rate each issue by severity.
Best for
fast, low-cost issue-spotting before you involve users.
How believable it is
Lowest believability — it's expert opinion, not user behavior. Fast and cheap, so it's a great first pass, but every finding is a hypothesis until a real user confirms it.

Concept Testing

QualitativeAttitudinal
EffortLowBelievability
LowHigh
What it is
Puts an early idea or value proposition in front of people to gauge appeal and comprehension before anything is built.
How it works
Present a concept — a sketch, description, or rough mock — and ask about relevance, clarity, and interest.
Best for
validating direction early, while change is still cheap.
How believable it is
Low. You learn whether an idea is understood and appealing in the abstract — useful for direction, but reactions to a concept rarely predict what people will actually do.

Paper Testing

QualitativeAttitudinal
EffortLowBelievability
LowHigh
What it is
You put a rough sketch, wireframe, or paper mockup in front of someone and watch them react and try to make sense of it.
How it works
Show low-fidelity screens on paper or in a clickable mock, hand over a task, and observe where they get confused or lean in. Cheap to make, cheap to throw away.
Best for
catching comprehension and layout problems before you write code.
How believable it is
Low-to-moderate. More believable than an interview because there's a concrete artifact to react to, so you catch comprehension and usability problems early. But it still measures an opinion in a test room, not demand under real stakes.

Desirability Studies

QuantitativeQualitativeAttitudinal
EffortLowBelievability
LowHigh
What it is
Measures the emotional response and appeal of a design, separate from whether it’s usable.
How it works
Users choose from a curated set of adjectives (or rate them) to describe a design; you tally the patterns.
Best for
understanding brand and aesthetic perception.
How believable it is
Low-to-moderate. Captures emotional response and aesthetic appeal, which is real signal for brand — but it measures feeling, not whether the design works or gets used.

Focus Groups

QualitativeAttitudinal
EffortMediumBelievability
LowHigh
What it is
A moderated group discussion that explores perceptions, reactions, and the language people use around a topic.
How it works
Guide five to eight participants through prompts, watching where they agree, disagree, and spark on each other.
Best for
surfacing early attitudes and the words people use.
How believable it is
Low, and easy to over-trust. The group surfaces language and attitudes but breeds groupthink; treat it as a source of hypotheses and quotes, not evidence of demand.

Surveys

QuantitativeAttitudinal
EffortMediumBelievability
LowHigh
What it is
Structured questions sent to many people to quantify attitudes, satisfaction, and self-reported behavior.
How it works
Design closed and open questions, sample your audience, and analyze the responses at scale.
Best for
measuring sentiment across a whole population.
How believable it is
Low-to-moderate. Quantitative reach is the draw, but answers are self-reported attitudes — great for measuring sentiment across many people, weak for predicting behavior.

Advertising

QuantitativeBehavioral
EffortLowBelievability
LowHigh
What it is
You run a small paid ad for the product or feature and measure whether anyone clicks — buying a tiny slice of real attention.
How it works
Write an ad for the value proposition, put a modest budget behind it on a channel your audience actually uses, and track click-through against impressions.
Best for
gauging real interest in a value prop, cheaply and fast.
How believable it is
Moderate. A click is a real action with a small cost of attention, so it outranks any stated intent. But a click is cheap and curiosity-driven — it signals interest, not commitment.

Card Sorting

QuantitativeQualitativeAttitudinal
EffortMediumBelievability
LowHigh
What it is
Participants group and label content to reveal how they expect information to be organized.
How it works
Give people a set of cards and have them sort into groups — open (they name the groups) or closed (you provide them).
Best for
shaping information architecture and navigation.
How believable it is
Moderate. A semi-behavioral task that reveals how people expect things to be organized — trustworthy for IA decisions, but it's about mental models, not real usage.

Interviews

QualitativeAttitudinalBehavioral
EffortMediumBelievability
LowHigh
What it is
One-on-one conversations that surface motivations, mental models, and the “why” behind what people do.
How it works
Follow a semi-structured guide, ask open questions, and probe on answers to go deeper than a survey can.
Best for
deep qualitative understanding of individual users.
How believable it is
Low-to-moderate as demand evidence. Unmatched for the “why” and people's mental models; discount stated intentions, because what people say they'll do rarely matches what they do.

Tree Testing

QuantitativeBehavioral
EffortMediumBelievability
LowHigh
What it is
Tests whether people can find things in your navigation structure, stripped of visual design — the inverse of card sorting.
How it works
Give users a text-only hierarchy and a task, then record the path they take and where they succeed or get lost.
Best for
validating an information architecture before visual design.
How believable it is
Moderate-to-good. Observed findability behavior stripped of visual design, so it cleanly isolates whether your structure works — an abstraction, but a behavioral one.

Click Testing

QuantitativeBehavioral
EffortLowBelievability
LowHigh
What it is
Measures where people click on a design to complete a task, testing whether the layout guides action.
How it works
Show a static image or prototype, set a task, and record first-click location and time to click.
Best for
validating findability and call-to-action prominence.
How believable it is
Moderate-to-good. Real clicks on a real task measure whether the layout guides action — behavioral and quantifiable, though limited to a single first move.

Button to Nowhere

QuantitativeBehavioral
EffortLowBelievability
LowHigh
What it is
You add a button or link for a feature that doesn't exist yet and count how many people click it — the “painted door” test.
How it works
Place the entry point inside the real product, then show clickers an honest “coming soon” message (ideally capturing an email). Measure clicks against traffic.
Best for
measuring demand for a feature before you build it.
How believable it is
Moderate-to-good. This is real demand from real users, inside the real product, at the exact moment of intent — far harder to fake than a survey. Just keep it honest so a dead end doesn't erode trust.

Landing Page

QuantitativeBehavioral
EffortMediumBelievability
LowHigh
What it is
You build a page that describes the product as if it exists and ask visitors to sign up, join a waitlist, or pre-register.
How it works
Write the pitch, add a clear call to action (email capture or “notify me”), drive traffic to it, and measure the conversion rate.
Best for
testing whether a pitch pulls sign-ups before launch.
How believable it is
Good. Handing over an email is a small but real act of commitment, and conversion rate is a clean, comparable demand signal. The classic “smoke test” — it tells you whether the pitch pulls before you build the thing.

Task Completion

QualitativeBehavioral
EffortMediumBelievability
LowHigh
What it is
You give someone a realistic prototype and a real goal, then watch whether they can actually get it done.
How it works
Set a concrete task (“book a room for two nights”), stay quiet, and record success, errors, and exactly where they get stuck.
Best for
confirming people can actually finish the core task.
How believable it is
Good for usability, limited for demand. It proves people can use it — essential — but a participant finishing a task you assigned isn't a customer choosing to. It answers “can they?”, not “will they, and will they pay?”

Eye Tracking

QualitativeBehavioral
EffortHighBelievability
LowHigh
What it is
Records where people look, revealing what draws attention and what gets missed on a screen.
How it works
An eye-tracking camera captures gaze paths and dwell time during tasks, rendered as scanpaths and heatmaps.
Best for
attention and visual-hierarchy questions.
How believable it is
Moderate. Observed behavior — where attention actually goes — so it's harder to fake than a stated preference. Narrow, though: it tells you what's seen, not whether it works.

Usability Test

QualitativeBehavioral
EffortHighBelievability
LowHigh
What it is
Watch representative users attempt real tasks to find where the product helps and where it gets in the way.
How it works
Set concrete tasks, observe, and note success, errors, and friction — moderated in a session or run unmoderated.
Best for
diagnosing task-level problems in a design.
How believable it is
Good. You watch real people attempt real tasks, so you see genuine behavior and friction — the workhorse of trustworthy design evidence, short of shipping.

Prototype

QualitativeBehavioral
EffortHighBelievability
LowHigh
What it is
You build a working, interactive prototype — realistic enough that people can genuinely use it, even if it's held together with tape behind the scenes.
How it works
Assemble a functional facade (no-code tools, hardcoded data, a clickable build) and let target users run through real scenarios end to end.
Best for
pressure-testing a real interaction end to end.
How believable it is
Good. Higher fidelity and real interaction surface problems a static mock can't, and reactions get more honest as the thing feels real. It's still a test environment, so demand and willingness-to-pay stay open questions.

Diary Studies

QualitativeBehavioral
EffortHighBelievability
LowHigh
What it is
Participants self-report their experiences over days or weeks, capturing behavior in real context over time.
How it works
Prompt users to log activities, feelings, or friction as they happen, then review the entries for patterns.
Best for
longitudinal, in-the-moment insight.
How believable it is
Moderate — and it costs a lot to get there. Self-reported, but captured in real context over time, which beats a one-shot lab session; the catch is weeks of participant wrangling and heavy dropout for a signal that's still self-report.

Analytics & Clickstream

QuantitativeBehavioral
EffortHighBelievability
LowHigh
What it is
Analyzes behavioral data from real usage to reveal what large numbers of people actually do.
How it works
Instrument the product and examine paths, funnels, and drop-off across many sessions.
Best for
spotting patterns and prioritizing where to dig deeper.
How believable it is
High. Real usage from real people, at full scale, with nothing staged — supremely believable for what happened, though correlational: it shows the what, rarely the why.

Pre-selling

QuantitativeBehavioral
EffortHighBelievability
LowHigh
What it is
You ask people to actually pay — a pre-order, deposit, or crowdfunding pledge — for a product you haven't built yet.
How it works
Put up a real checkout or pledge page at a real price. Take (or authorize) the money, and be ready to refund if you don't ship.
Best for
proving people will pay before you build.
How believable it is
Very high. Money is the most honest signal there is — a credit card is skin in the game that talk, clicks, and emails never match. If people won't pay, you've learned the most important thing before building anything.

Concierge

QualitativeBehavioral
EffortVery HighBelievability
LowHigh
What it is
You deliver the service manually, by hand, to real customers — no product yet, just you doing the work behind the scenes to prove the value is real.
How it works
Find a few real customers and fulfill their need personally, step by step. You learn the workflow intimately while confirming they actually want the outcome.
Best for
validating the value by delivering it by hand.
How believable it is
Very high on value. Because you serve real people solving a real problem (often paying), you validate the value proposition and learn precisely what to automate. What it doesn't prove is that it scales — that's the next question.

Wizard of Oz

QualitativeQuantitativeBehavioral
EffortVery HighBelievability
LowHigh
What it is
Customers use what looks like a finished, automated product — but humans are secretly powering it from behind the curtain.
How it works
Build the real-looking front end, then have your team manually handle the matching, responses, or “algorithm” the interface implies. Users never know.
Best for
testing a hard-to-build experience with humans behind the curtain.
How believable it is
Very high. You capture genuine usage behavior at near-real fidelity without building the hard back end — the closest thing to live-product evidence before the live product exists. Labor-intensive, and only sustainable at small scale.

Field Studies (Contextual Inquiry)

QualitativeBehavioral
EffortVery HighBelievability
LowHigh
What it is
Researchers observe people in their own environment, revealing behavior and context a lab setting can’t.
How it works
Visit users where they work or live, watch real tasks unfold, and ask questions in the moment.
Best for
discovery work and grounding decisions in real context.
How believable it is
Good-to-high. Behavior observed in its real context, where the messiness lives — expensive and hard to scale, but few methods are more grounded in reality.

A/B Testing

QuantitativeBehavioral
EffortHighBelievability
LowHigh
What it is
Compares two versions with live traffic to see which one drives a better outcome.
How it works
Split users randomly between A and B, hold everything else constant, and measure the metric that matters.
Best for
evidence-based optimization decisions.
How believable it is
Very high. Live traffic, randomized, so you get a causal read on what actually changes behavior — the closest thing to proof, provided the test is powered and clean.

Multivariate Testing

QuantitativeBehavioral
EffortVery HighBelievability
LowHigh
What it is
Tests several element variations at once to learn which combination performs best.
How it works
Serve combinations of changes — headline × image × button — to live traffic and compare the results.
Best for
optimizing high-traffic pages with many variables.
How believable it is
Very high. Like A/B, but isolates which combination of changes wins — real behavior at scale; it just needs a lot of traffic to reach significance.

Live Product & Business

QuantitativeBehavioral
EffortVery HighBelievability
LowHigh
What it is
The real thing, in the market: real users, real transactions, real money, at real scale — the fully operating product and business.
How it works
Ship it, run it, and instrument everything — retention, revenue, funnels, churn — reading behavior from people with nothing staged.
Best for
the real, at-scale read once you've shipped.
How believable it is
The highest believability there is: unfaked behavior and real dollars at scale. The catch is cost — it's the most expensive way to learn, which is why every cheaper method exists: to buy as much of this certainty as you can before paying full price.

Positions are an experienced rule of thumb, not exact measurements — effort and believability shift with how a study is run. Use it to reason about trade-offs, not to settle an argument.