The Recency Effect in UX: Why the Last Thing Wins, and How Long It Lasts

Give someone a list of twenty words, one per second, and ask them to write down everything they can. The answers are not evenly distributed. The first few words come back well, the last few come back better, and the middle is a wasteland. Bennet Murdock ran that experiment across list lengths and presentation rates in 1962 and got the same U every time — a curve steep at both ends and flat through the middle (Murdock, 1962).

The bump at the end is the recency effect. It is the most reliably reproduced result in memory research and the one most often misquoted in design decks, usually as a general-purpose claim that recent things feel more important. That version describes a different thing, recency bias, and the two get swapped constantly. The swap costs you the property that makes the recency effect operationally useful, which is that it expires after about thirty seconds.

The U has two different engines

Primacy and recency both lift the curve, and they are not two views of one mechanism. The opening items get rehearsed — you hear "table," repeat it, hear "window," repeat both, and by the fourth item you have already given the first one four passes. That rehearsal writes them into long-term storage. The closing items get no rehearsal at all. They are simply still sitting in short-term memory when the recall prompt arrives, and they get read straight off the top.

Glanzer and Cunitz separated the two engines cleanly in 1966. Insert a filled delay between the last word and the recall prompt — thirty seconds of counting backwards, enough to occupy the short-term store without touching what is already consolidated — and the tail of the curve collapses while the front of it stays exactly where it was (Glanzer & Cunitz, 1966). Primacy survives the delay because it was written into long-term storage on the way in; recency has nothing to survive on.

Spend the end on action and the start on memory

That gives a rule sharper than "put important things last." Put things at the end when the response you want happens inside the same session. The final screen of an onboarding flow and the summary row above the pay button both borrow real strength from the recency effect, because the action and the recall are seconds apart.

Put things at the start when the response you want happens later. Brand promises, pricing structure, the fact that the trial does not need a card — anything the user has to carry out of the session and into a decision next week belongs in the primacy zone, because primacy is the half that survives a night's sleep. Teams routinely get this backwards, saving their strongest claim for a big finish and then wondering why nobody who came back could repeat it.

The middle of a sequence gets none of this. Anything sitting there needs a heading of its own.

Where it lands in an interface

The clearest case is anywhere a person holds options in their head and then picks one. In a menu or a set of radio options read top to bottom, the last few entries are the ones still available when the choice is made. The pull is weaker in self-paced visual scanning than in the auditory lab task, because you can look back up and people do. It gets stronger as the list grows and as the options blur into each other, so a pricing page with a dozen near-identical plan names is letting its own row order carry some of the decision.

Voice and conversational interfaces are the strong version of the same thing. A user cannot glance back at the second-to-last option in a phone tree or a smart-speaker list; the interface is pure serial presentation with no visual backstop, exactly the condition Murdock was testing. Read out five choices and the fifth is the one that gets chosen, because it is the only one still in the buffer. Cap the list at three, or restate the two the person is likely to want.

Then there is the survey. Long grids of response options are answered by the tail in aural administration and by the head in visual administration — an order effect large enough that survey methodologists treat rotating option order as routine. If your in-product feedback widget shows every respondent the same five reasons in the same order, your top reason is partly an artifact of the list. Rotating the order between respondents costs nothing and is the standard fix, alongside care over how you write the response options themselves.

The version that damages your research

Six usability sessions in a day, debrief at the end of the day, and the sixth session is disproportionately what the group argues about. Nobody rates it as more important. It is simply the one everybody can still narrate in detail, so it supplies the examples, and the examples set the frame. Sessions two and three are a general impression by 5pm, sitting in the flat part of the curve.

The countermeasures are dull and they work. Write the findings for each session before the next one starts, not at the end of the block. Debrief from those notes. When you rank issues, rank them against tallied counts across sessions, not against the vividness of the recollection. And if you can, run the analysis a day later, once the recency advantage of the last session has decayed and every session is competing on the same terms. A night's delay is the same filled interval that flattens the tail in the lab, applied to your own session-by-session notes.

Not the same thing as recency bias

Three ideas get filed under the same word and they are worth pulling apart.

The recency effect is a memory result about serial position: the last items in a just-presented sequence are recalled best, and that advantage disappears after a filled delay. It is a narrow claim about one just-presented list, and it has a clock on it.

Recency bias is a judgment tendency: recent events are weighted more heavily in an assessment than older ones — last quarter's numbers over the three years before it, the most recent support ticket over the backlog. It has no thirty-second expiry, and it is one of the main reasons some things come to mind easily at all, which makes it one of the inputs feeding the availability heuristic.

The peak-end rule governs how an experience is remembered rather than what was in it: an episode is summarised by its most intense moment and its ending, with duration counting for far less than it should (Kahneman et al., 1993). This is the one people usually mean when they say "end on a high note," and it is why a slow flow that finishes cleanly can be recalled more warmly than a fast one that ends in an error.

Conflating them usually survives contact with reality. It falls over the first time someone asks how long the effect is supposed to last.

Lab lists are not interfaces

Serial position curves come from lab tasks with arbitrary word lists and a recall prompt seconds later. An interface is none of those things. Users skim, they can re-read, they arrive with goals, and the "list" competes with a page full of other structure.

That is a fair description of the limits, and it should make you suspicious of any redesign justified by the curve alone. But on a screen the user can re-read, the effect shrinks toward zero. It stays close to full strength in the conditions a lot of product surfaces impose: voice output, linear flows you cannot scroll back through, and the debrief at the end of the day. What it governs is the part of your interface that behaves like a list read out loud.

Frequently asked questions

What is the recency effect?

The tendency for the last items in a sequence to be recalled better than the items in the middle. It is one half of the serial position effect, the other half being primacy, and unlike primacy it depends on short-term memory: the advantage disappears after roughly thirty seconds of an intervening task.

What is an example of the recency effect in UX?

A voice assistant reads five options and the caller picks the fifth. Or a research team runs six sessions in a day, debriefs at 5pm, and builds most of its argument out of the sixth — the four in the middle having quietly become a general impression somewhere around lunch.

What is the difference between the primacy and recency effects?

Primacy comes from rehearsal, which writes early items into long-term memory, so it survives a delay. Recency comes from items still sitting in short-term memory at the moment of recall, so it does not. Practically: the start of a sequence is what a user remembers tomorrow, the end is what they act on now.

Is the recency effect the same as recency bias?

No. The recency effect is a memory result about position within a just-presented list, with an expiry measured in seconds. Recency bias is a durable tendency to overweight recent events when forming a judgment. Related, and often confused, but only one of them decays if you wait.

How do you reduce the recency effect in surveys?

Rotate the order of response options between respondents so no option holds the last position for everyone, keep option lists short enough to be held in mind, and be especially careful with aurally administered surveys, where the respondent cannot look back.

Does the recency effect apply to visual interfaces?

Weakly, and it depends on the layout. When a user can freely scan and re-read, position matters much less than hierarchy and grouping. When the presentation is serial and irreversible — a voice flow or a step-by-step wizard — the effect is closer to its laboratory strength.

How long does the recency effect last?

Seconds. Glanzer and Cunitz found that around thirty seconds of a filled distractor task removed the recency advantage entirely while leaving primacy untouched. Any design that depends on the tail of a sequence is depending on something with that shelf life.

Should the most important option go last in a list?

Only if the choice is made immediately and the list is short and serially presented. If the user will decide later, or can re-read the list, the first position and clear visual prominence are the better bets.

Take it further

The recency effect is a small, well-measured result that gets stretched into a general theory of endings and then quietly stops being checkable. The UX Clarity framework treats claims like this the way the original studies did, one condition at a time against one observable outcome, so a design argument can be wrong in a way you can check. A Full UX Audit applies it to the flows where serial order is actually forced on the user, which is usually not where a team expects. The rest of the biases that shape how a screen gets read and remembered are collected in UX psychology.

Sources: Murdock, 1962 — The Serial Position Effect of Free Recall · Glanzer & Cunitz, 1966 — Two Storage Mechanisms in Free Recall · Kahneman, Fredrickson, Schreiber & Redelmeier, 1993 — When More Pain Is Preferred to Less · NN/g — Serial Position Effect.

Not sure whether your flow is remembered for its first screen or its last? Apply for a Full UX Audit →