Progress indicators: why they work and when they do not

“Your profile is 70% complete”

Many platforms show a bar with a percentage and a note on what is still missing while you set up a profile. Most people who see it want to fill it. In marketing this urge is often explained with the Zeigarnik effect.

That is the popular version. The research paints a more precise picture: two other effects carry the impact of progress indicators far better, and the Zeigarnik effect in the narrow sense does not hold up in a recent meta-analysis.

Three effects, cleanly separated

Goal gradient: closer to the goal, more effort. The idea is old, and it comes from rats in a maze. Only the paper endlessly cited for it never ran a single rat.

Hull (1932) is a theoretical paper with no experiment of its own. Hull formulated the hypothesis that drive grows with proximity to the goal and derived its shape from someone else’s data: by his account, Yoshioka (1929) ran almost 200 rats on mazes with two paths of different length, and equal length ratios were equally hard to discriminate — regardless of the absolute difference. From this Hull inferred a logarithmically rising curve.

That animals run faster the closer the goal comes is, in this paper, a derived prediction, not an observation. Hull merely proposes there the apparatus with which to test it. Two years later he built it: by the account in Kivetz et al. (2006), Hull (1934) used a straight runway with electrical contacts to measure the time per section and found the prediction confirmed.

Anyone invoking “rats ran faster” therefore means Hull (1934), not Hull (1932).

Kivetz, Urminsky and Zheng (2006) transferred the idea to consumers. Their basis is 949 redeemed ten-stamp cards from a university café covering some 10,000 coffee purchases: the interval between purchases fell by 20 percent, or 0.7 days, from the first stamp to the last. In effect, members bought about two more coffees per month than they would have without the programme — to get one free.

Two controls make the finding robust.

Both rule out habituation and general time trends.

A merely perceived closeness to the goal also worked: in a field experiment with 108 customers, the group given a twelve-stamp card with two free stamps needed 12.7 instead of 15.6 days for the same ten purchases. The authors specifically tested whether sunk cost explains this and rejected it: people who had earned the two stamps themselves felt worse about losing the card than the gift group, whose feelings were identical to the control group’s.

Endowed progress: a head start motivates. Nunes and Drèze (2006) gave car-wash customers stamp cards. One group had a card with eight spaces, the other a card with ten spaces on which two stamps were already placed. In both cases eight purchases were required. A higher share of the group with the head start redeemed the reward (34 versus 19 percent). Three hundred cards were handed out and 80 redeemed. Along the way, the interval between visits was on average 2.9 days shorter with the head start; the interplay with progress, however, was not meaningful, so the acceleration towards the goal is no steeper than without it.

Whether the head start needs a reason depends on how it is counted. In a survey of 240 customers outside a liquor store — a hypothetical programme, no purchases — it worked without a reason only when counted in points (4.79 versus 4.11 with no head start). Counted in purchases it did nothing without a reason (4.24 versus 4.06, p = .25). The reason did not have to be sound: “because you are here today” worked as well as “as someone with your purchase record”.

Zeigarnik and Ovsiankina: remembering or carrying on? Two things get confused here that the research keeps apart.

Zeigarnik (1927) gave participants 18 to 22 small tasks each, interrupted half of them and then asked which ones they remembered. In the main experiment with 32 people the interrupted ones were named 90 percent more often. The well-known waiter story comes from the circle of her supervisor Kurt Lewin and is an anecdote, not part of the experiment.

Ghibellini and Meier (2025) summarized the research on this — and the two findings stand in sharp contrast.

A ratio of 1.0 would mean no difference between interrupted and completed tasks. So 0.99 is exactly nothing.

In practice this means: the urge to carry on is supported, better remembering of unfinished things is not. Anyone citing “Zeigarnik” as the reason for a progress bar should mean the Ovsiankina effect.

The original work on that is Ovsiankina (1928), by a colleague at the same Berlin institute. She interrupted 124 people at small craft and puzzle tasks. Where the experimenter did the interrupting, resumption ran at 74 percent; where the interruption looked accidental, at 100 percent.

Two of her controls matter more for progress displays than the rate itself.

So the urge attaches to one’s own started work, not to the sight of something unfinished — and it needs a visible end. Both argue for a display that shows real progress of one’s own and names a clear end, and against bars that depict the system’s work.

Where progress bars do not help

Villar, Callegaro and Yang (2013) evaluated 32 randomized experiments from 19 studies on progress bars in web surveys. The result depends on how the bar advances — and only part of it holds up.

• Constant bar (18 experiments): no meaningful effect. The estimate even leans slightly towards more abandonment (LOR = 0.07, p = 0.37) — LOR is a way of comparing frequencies in which 0 means “no difference”.
• Slow start (7 experiments): markedly more abandonment; the odds of dropping out were 1.56 times those without a bar (p = 0.001). This finding survives the removal of an outlier.
• Fast start (7 experiments): across all experiments not significant. Only after removing one outlier does abandonment fall significantly (factor 0.80, p = 0.02). The short version “a fast start helps” therefore rests on one leg.

A secondary finding bears directly on loyalty programmes: where a small reward was promised for completion, a constant bar increased abandonment by a factor of 1.17 (after removing two outliers, p = 0.03). Where an incentive is already pulling, the progress display can get in the way rather than help.

These are surveys, not checkouts — and the gap is wider than it sounds. The surveys studied took 18 to 22 minutes on average, none under five. The authors state explicitly that they cannot say anything about shorter processes. A checkout sits exactly there.

Practical cases

What we deliberately do not recommend

• Fake progress: A bar such as “Analysis 85% complete” with no real analysis behind it is a dark pattern. It exploits the urge to resume and deceives users about their status.
• Accelerated bars: A bar that starts fast and slows towards the end is the only pattern with any sign of reducing abandonment — but only because it distorts real progress. Villar et al. (2013) consider precisely that ethically questionable and advise against it. On their reading, an uneven bar is acceptable only when it mirrors the actual effort, for instance because the first steps really are shorter.
• Zeigarnik as the reason for email subject lines: For “Your setup is still incomplete” as an open-rate lever, this guide has no verified source. If you use it, do it as an A/B test.

Ethics check: nudge or sludge?

Limits and criticism