The STAR method, with examples that are not made of cardboard

Situation, Task, Action, Result. Almost every page will define those four words for you. Far fewer will tell you that the Result is where most answers quietly fail, or that the whole thing is a 1974 training framework rather than a scientific law.

Updated 22 August 2026

The short answer

Situation: one sentence of context, no more. Task: what specifically was on you. Action: what you did, in the first person, in the most detail of the four. Result: what changed, with a number or a concrete consequence. Roughly 10 / 10 / 50 / 30 by weight. Most people invert that and spend half the answer setting the scene.

Where STAR actually comes from

It is worth knowing, because it changes how much authority to give it. STAR was introduced by the consultancy Development Dimensions International in 1974 as part of a hiring-training product. It is a practitioner framework, not a research finding, and no study says answers in this shape score better. What research does support is the thing underneath it: interviews that ask about real past behaviour predict job performance better than ones asking what you would hypothetically do. STAR is simply a reliable way of making sure a real example arrives with its evidence attached.

Why the R is where it falls apart

Situation and Task are easy, so people over-invest there. Action feels like the substance, so it gets told at length. Then the answer reaches the end and stops, because the result was never quantified, or it was genuinely mixed, or nobody measured it. An interviewer hearing no result will usually probe for one, and an answer that cannot produce it on the second ask lands worse than one that never claimed it.

Three worked answers

A result with a number

“Our onboarding drop-off was about forty percent at the second step and nobody owned it (situation). I picked it up because my team took the support load from it (task). I watched a dozen session recordings, found that the step asked for a company VAT number most self-serve users did not have, and made it optional with a follow-up email instead. I shipped it in a week rather than waiting for the redesign (action). Drop-off at that step went to about twelve percent and support tickets on it stopped almost entirely (result).”

A result with no number

“Two of our engineers had stopped speaking to each other, and code review had become where they fought (situation). I was the lead, so it was mine to fix, and I had been avoiding it for about a month (task). I spoke to each of them separately, and it turned out to be a disagreement about ownership of a service rather than anything personal. I wrote down who owned what, got them both to agree it in the same room, and moved one of them onto a different area for a quarter (action). They are not friends, but review went back to being about the code, and neither of them left (result).”

A result that was bad

“We shipped a pricing change I had argued for, and it cut conversion by about a fifth in the first month (situation and task). I had run the analysis on existing customers and never tested it against new ones, which is where the damage was (action). We rolled it back inside three weeks. What I took from it is that I had evidence for the population I could see and treated it as evidence for everyone, and I now say out loud which group a number actually covers (result).”

Using the structure without narrating it

Never say “so, the situation was”. The structure is a checklist for you before the interview, not a script during it. If your answer arrives with context, your own role, what you did and what changed, you have used STAR whether or not anyone can hear the seams.
One practical test: read your answer back and count the sentences before the first thing you personally did. If it is more than two, cut the front.

What to do when the result is not a number

Plenty of real work does not produce a metric. A decision that got made, a person who stayed, a process nobody complained about again, a launch that happened on the date it was supposed to: all of these are results.
What does not work is an adjective. “It was really successful” asks the interviewer to take your word for it, and the entire point of the format is that they do not have to.

When the result was bad

Say so, then say what you took from it. Failure questions exist precisely to find out whether you can do this, and a genuinely bad outcome described plainly scores better than a mediocre one dressed up. The tell that an answer has been sanitised is usually that the candidate is the only person in it who behaved well.

Now say it out loud

Reading a good answer is not the same as giving one.

Studies of interview coaching find a dose-response: candidates who get information do better than those who get none, and candidates who actually rehearse the skill out loud do better again. You can have the structure right and still lose the room when Daniel asks the follow-up. First session free, no card.

See where you stand

Common questions

Is STAR still what interviewers expect in 2026?

In most large organisations, yes. Behavioural interviewing is standard at that scale and interviewer training is often built around this exact framework, so an answer with the pieces present is easy to score. Smaller companies are less formal, but a specific example with an outcome still beats a vague one everywhere.

What about CAR, SOAR and the other acronyms?

Same idea with different letters. They all exist to make sure you supply context, your own action and an outcome. Learning several is wasted effort; knowing what evidence an interviewer needs is the transferable part.

How long should a STAR answer be?

About two minutes, and shorter if the result arrives early. The frequent failure is not length but proportion: a ninety-second setup followed by a rushed fifteen-second result is a worse answer than the reverse.

On the research: STAR was introduced by Development Dimensions International in 1974 as part of its Targeted Selection training; the underlying finding that past-behaviour questions outpredict hypothetical ones is from Taylor and Small (2002).