Satisficing: Users Pick the First Good-Enough Option, Not the Best One

Summary: Users don’t choose the best option; they choose the first satisfactory one, and then they stop looking. This behavior, called satisficing, decides which link gets clicked, which default survives, and which of your features nobody ever sees. Design so that the first plausible option is also the correct option, and the lazy click becomes the right click.

Somewhere in the forest, a grander fire is always burning. The satisficer stays with the one that heated his mug because warm coffee was the goal, and warmth has been achieved. Users camp at the first feature that works and never tour your showpieces. (Muse Image)

Nobody circles a parking lot 6 times hunting for the mathematically optimal space. You take the first spot that’s reasonably close and reasonably legal, because the 40 seconds a better spot might save aren’t worth 5 more minutes of searching. Congratulations: you’re a satisficer. So is every user of your product.

Definition: Satisficing is a decision strategy in which a person accepts the first option that meets his or her threshold of acceptability, rather than continuing the search for the best possible option.

Door 1 stands plain and open; door 12 is a goldsmith’s life work. The satisficer is already inside door 1, so all that gilding persuaded exactly nobody. Polish the option users reach first, not the options they never reach. (GPT Image 2)

Example: Alice needs a hotel in Copenhagen. She doesn’t score all 240 properties on 12 attributes; she opens the list, takes the first one with decent photos, an acceptable price, and free cancellation, and books it in under 4 minutes. The remaining 239 hotels got exactly as much consideration as your unvisited features do.

If Goldilocks had been satisficing, the story would have been much shorter: no tasting all 3 bowls of porridge, no trying out 3 different chairs, no sleeping in 3 beds. That’s how we know it was a fairy tale. (GPT Image 2)

A Nobel Prize for Admitting That Humans Don’t Optimize

Classical economics assumed a tireless optimizer who gathers all information, weighs every alternative, and selects the mathematically best one. Herbert Simon called that assumption unrealistic. Real decision-makers work with limited information, limited time, and limited computing power in the skull, a condition he named bounded rationality in his 1955 paper “A Behavioral Model of Rational Choice” (PDF; The Quarterly Journal of Economics, vol. 69, pp. 99–118). The rational strategy under those bounds is to set an aspiration level, search until an option clears it, and stop.

Bounded rationality is the result of the limited computing power in the skull. In this metaphorical illustration, the maze is the full problem, and the flashlight beam is the working memory available to solve it: it’s only practical to consider parts of the whole. The user takes the first exit the light reaches. Your design competes for that beam. (GPT Image 2)

The word itself (a blend of satisfy and suffice) debuted the following year in “Rational Choice and the Structure of the Environment” (PDF; Psychological Review, vol. 63, pp. 129–138), where Simon observed that organisms “adapt well enough to ‘satisfice’; they do not, in general, ‘optimize.’” The idea had been brewing since his 1947 book Administrative Behavior, and in 1978 it earned him the Nobel Memorial Prize in Economic Sciences. Economists needed decades to accept that Homo economicus, the tireless maximizer, doesn’t exist. UX professionals should need less convincing: we watch him not exist in every usability session. Simon pointed at actual humans: H. sapient, not H. economicus.

Why Evolution Built a Satisficer

Simon’s 1956 paper wasn’t about shoppers. It was about survival. He modeled a hungry organism crossing a landscape where food lay in scattered patches, and showed that a creature equipped with nothing but a threshold rule (start searching when hunger crosses a set point, eat the first food that clears the bar) survives indefinitely. No utility function required, no ranking of every berry in the valley. Simple perception plus a stopping rule was enough, and that was Simon’s point: the structure of the environment, not the sophistication of the brain, decides how much thinking a good decision needs.

Evolution shaped our decision machinery in a world that rewarded stopping early:

Searching cost time, and time was dangerous. An hour spent comparing patches was an hour exposed to predators, weather, and competitors who ate while you deliberated.

The two errors weren’t symmetric. Overlooking a slightly better tuber cost a few calories. Failing to eat cost everything. When one mistake is survivable and the other isn’t, you set the bar low and move.

Weighing every option isn’t worth the delay when the options are roughly similar, as they were throughout our evolutionary past. (GPT Image 2)

Options were few and similar. You compared the three patches within walking distance, not 240 hotels. When the alternatives cluster closely in value, exhaustive comparison gains you almost nothing and still costs all the time and calories.

Computation was rationed. The human brain is about 2% of body weight but burns a staggering 20% of the body’s energy at rest. Evolution favors the smallest decision rule that works, not the best rule imaginable.

Behavioral ecology later formalized the same logic. Eric Charnov’s 1976 marginal value theorem (PDF; Theoretical Population Biology, vol. 9, pp. 129–136) says a foraging animal should abandon a patch the moment its yield rate falls to the average rate for the habitat, rather than emptying it or surveying every patch first. Leaving food behind is optimal. The user who abandons your page at second 12 is running the same stopping rule that kept his or her ancestors alive.

Too much work to pick all the fruit on every tree. The hominin who moved to the next tree had a slightly better calories-in-vs.-out balance. Multiplied over a million years, that sliver of a probabilistic advantage decided who became our ancestor and who became a dead end. Thus, we carry genes that encourage us to leave a website before reading everything. (GPT Image 2)

What changed is the option set, not the rule. We evolved to choose among three berry patches and now face 40 subscription tiers. The aspiration level made the trip across the millennia unmodified, still calibrated for berries.

The genes that continue to drive our behavior were honed for survival in the ancestral environment, where eating good-enough berries would enable a hominin to survive and thus pass down his DNA to us, his descendants, a quarter of a million years later. Being able to choose the best subscription plan among 40 had no survival value, so we’re bad at that. Design for cavepeople, because they’re your customers at the genetic level. (GPT Image 2)

Web Users Are Champion Satisficers

Satisficing is rational whenever search is expensive, and in an interface, search costs attention: the scarcest resource in the digital economy. In 1997, John Morkes and I found that 79% of users scanned a new webpage instead of reading it word by word. Steve Krug devoted a chunk of Don’t Make Me Think (2000) to the same behavior: users take the first reasonable link rather than hunting for the best one, and then they muddle through. And Peter Pirolli and Stuart Card at Xerox PARC formalized the economics in their information-foraging research during the 1990s: like predators, users weigh expected nutrition against expected effort and pounce on whatever smells adequately of prey. That smell has a name in our field: information scent, the cues in a link or label that predict what lies on the other side of the click.

Is that laziness? No, it’s arithmetic. Reading your entire navigation to guarantee the optimal click costs more time than an occasional wrong click plus the Back button, so the fast strategy wins on expected value. Thus, your lovingly crafted option 7 might as well not exist if option 2 looks passable. You have 10–20 seconds to persuade an unconvinced visitor to stay (my long-standing estimate from page-visit statistics). Harsh? Watch a usability test.

The maximizer’s city glitters with endless doors and staircases promising a perfect solution just a little farther on. The satisficer takes the plain doorway marked “good enough” and gets on with his day. Users take the doorway. Design for the doorway. (GPT Image 2)

AI hasn’t repealed any of this. People accept a chatbot’s first plausible answer the way they accept a default setting, and they ship the AI’s first draft the way they click the first decent link. The pattern extends upstream, too: most people type the first prompt that comes to mind, accept the first answer that reads well, and never discover what a second attempt would have bought them. Prompt satisficing is why so many confident verdicts about AI quality are really verdicts about first drafts. Satisficing didn’t retire when AI arrived; it just got a broader remit.

So users travel a first-plausible path through your interface: the sequence of first-acceptable choices at every decision point. The design job is to make that path and the correct path one and the same. Users are doing what evolution built them to do, and a design that punishes normal foraging behavior is the defective party in the transaction.

Users take the first plausible path through a user interface that seems to meet their goal. Make sure it’s also the best path. (GPT Image 2)

The Satisficing Trap

Designers get satisficing wrong in two opposite ways.

The first way is to forget it: burying the money answer in paragraph 4, writing links that don’t diverge until word 3 (“Information about billing” vs. “Information about accounts”), or assuming people will survey the whole menu before choosing. They won’t. Whatever half-fits first gets the click, and when that click leads somewhere wrong, the user blames your product, not his or her own haste. Correctly so, in my view.

The second way is to remember it all too well. A satisficing trap is a default or first option engineered so that good-enough acceptance quietly hurts the user, and e-commerce has assembled a full catalog of such dark design: the pre-checked travel insurance that sneaks into baskets because satisficers don’t audit defaults, the decoy plan planted so the target plan clears the aspiration level faster, the confirmshaming button (“No thanks, I hate saving money”) that exploits the urge to grab the first plausible exit. Do these tricks convert? Of course they do. That’s why they spread. (Survival of the fittest works on design patterns, just as much as it did on our satisficing ancestors.) But they also generate refunds, chargebacks, one-star reviews, and regulators: the US Federal Trade Commission cataloged these designs in its 2022 staff report Bringing Dark Patterns to Light. Exploiting a cognitive strategy that users can’t switch off amounts to mugging people in slow motion. Short-term conversion, long-term litigation. Some trade.

Fortunately, the honest remedies are cheap. Make the default the option you’d choose for your own mother. Surface the consequences (price, commitment, cancellation terms) at the decision point, not 3 screens downstream. And test with real users: 5 participants will expose the places where your first plausible option points the wrong way.

Designers Satisfice Too

The bias doesn’t spare its students. The first design concept that clears the team’s aspiration level ships unexamined. That’s satisficing turned on your own work, where it hurts the most. The phenomenon has a name: design fixation. In 1991, David Jansson and Steven Smith at Texas A&M University asked engineers to design a spill-proof coffee cup; half of them saw an example first (Design Studies, vol. 12, pp. 3–11). The example group copied its features, including a leaky straw the brief had banned. Professional engineers fixated too.

You’ve watched milder versions all week: the first Figma template that roughly fits, the market leader’s checkout flow copied because “they surely must have tested it” [ha!], the 5 nearest colleagues drafted as a “representative sample.” Each cleared a bar; none faced an alternative. AI sweetens the temptation: a model coughs up a polished mockup in 40 seconds, the first render looks done, and “done” is a seductive aspiration level.

The measured antidote is parallel design: several independent alternatives before any critique. Steven Dow and colleagues at Stanford ran the decisive experiment in 2010 (PDF; ACM Transactions on Computer-Human Interaction, vol. 17, article 18): 33 novices designed web ads; those drafting several prototypes before any critique beat the one-at-a-time designers on click-through rates across 1.2 million real impressions. Nearly half the serial designers took critique of their lone concept personally; no parallel designer did. Jan Maurits Faber and I prescribed the same medicine in an earlier 1996 case study (IEEE Computer, vol. 29, pp. 29–35): 3–4 independent designs, then merge the best of each. With generative AI, the old cost objection to parallel design is dead: you can get 5 variants in a minute.

Then comes cheap iteration: test the design with 5 users, fix what they stumble over, and repeat. Iteration is how a team raises its own aspiration level. Raising it costs a week; shipping your first guess costs a redesign.

The quest specified a chalice; the knight’s aspiration level specified a container, so he satisficed. Note the map: the actual Grail sat 3 days’ ride away, or one more design iteration. The gasping court is his usability test, arriving after launch, as usual. (GPT Image 2)

Respect the Maximizer Minority

One warning against overcorrecting. A minority of users are genuine maximizers who want exhaustive comparisons before committing, and researchers have their number. Barry Schwartz of Swarthmore College and colleagues built a 13-item Maximization Scale (PDF; Journal of Personality and Social Psychology, vol. 83, pp. 1178–1197) in 2002. A sample item: channel-surfing through the other options even while watching a program you like. Across thousands of respondents, Schwartz reported that roughly 10% score as extreme maximizers and another 10% as extreme satisficers, with everybody else strung between the poles. So the 1-in-10 rule of thumb has an empirical anchor, as long as you remember what it counts: the extreme tail, not a population split.

Roughly 10% of users are extreme maximizers who’ll happily burn an hour on a decision you expected to take a minute. Enough to be worth including as a persona in your design project. Too few to be the main design target. (GPT Image 2)

Two things keep that 10% from being a tidy segment. Maximizing is a continuum rather than a species, so the headcount follows wherever you place the cutoff (Nathan Cheek and Schwartz counted 11 competing scales by 2016, several of which disagree about what the trait even is). And it’s situational: the same person compares 12 laptops on 9 attributes, then takes the first sandwich in the case. A 2022 study of domain specificity by Minfan Zhu, Jun Wang, and Xiaofei Xie at Peking University found that people maximize in the domains they personally care about and satisfice everywhere else.

Behavioral data tells the same story. Babur De los Santos of Clemson University analyzed a panel of online book purchases and found that in 76% of transactions (PDF; International Journal of Industrial Organization, 2018), the buyer had visited exactly one bookstore in the week before buying, averaging 1.29 stores in total. Count only the purchase day, and 90% visited a single store. Comparison shopping for a $15 book is a 1-in-4 behavior at most. Raise the stakes, and the share climbs: before buying a digital camera, shoppers in Bart Bronnenberg, Jun Kim, and Carl Mela’s 2016 tracking study (PDF; Marketing Science, vol. 35, pp. 693–712) averaged 14 searches across 2.8 brands and 6.4 models, though 39% of them still looked at a single brand and 21% at a single product.

The design conclusion survives all of it: expect about 1 in 10 users to be maximizers by disposition, expect that share to swell as price, risk, and irreversibility rise, and expect the same person to switch modes between one task and the next. Serve the maximizing mode with comparison tables and full specifications behind progressive disclosure. But never force the entire population through a comparison wizard to reach a decision that satisficers would happily have made at first glance.

And don’t confuse serving maximizers with multiplying options for everyone. The classic warning here is the jam study: Sheena Iyengar and Mark Lepper set up a tasting booth at Draeger’s Market, my favorite local grocery store in Silicon Valley, where 24 jams drew a bigger crowd than 6, yet only 3% of the 24-jam browsers bought anything, against 30% at the small table (Journal of Personality and Social Psychology, vol. 79, pp. 995–1006).

You’ve no doubt heard of the old jam-selling study because it’s a favorite among design speakers. Unfortunately, it failed replication in newer research. Choice overload is real enough, but the culprit is our habit of satisficing rather than the sheer number of items. (GPT Image 2)

That result makes for a funny cartoon, but unfortunately it hasn’t held up as stated. Benjamin Scheibehenne, Rainer Greifeneder, and Peter Todd pooled 50 experiments in 2010 (Journal of Consumer Research, vol. 37, pp. 409–425) and found a mean choice-overload effect of zero, including a direct replication of the jam study with 504 shoppers that produced nothing at all. Alexander Chernev’s team at Northwestern University then pooled 99 observations from 7,202 participants (Journal of Consumer Psychology, vol. 25, pp. 333–358) and reconciled the contradictions: overload is real, but it depends on 4 moderators, namely how complex the option set is, how hard the decision task is, how unsure users are of their own preferences, and whether they came to choose or merely to browse. Assortment size on its own predicts little.

Satisficing explains why. Users never compare the full set, so the count isn’t the burden; the cost of telling the options apart is. A catalog of a million items works fine when search, filters, sorting, reviews, and an honest recommendation do the winnowing before the user arrives. A single screen of 8 plans that differ on 12 attributes, with no dominant option and no guidance, defeats the stopping rule: nothing visibly clears the bar, so the user leaves instead of deciding. Cutting the option count is one fix, and usually the crudest one. Differentiate what you keep, order it well, and carry the comparison burden yourself.

The sign at the hut states the hiker’s aspiration level, and dinner has cleared it. The summit stays on the horizon for the 1 in 10 who insist, path fully visible. That’s progressive disclosure rendered as landscape: satisfy the majority at the hut, and keep the climb available for a few maximizers. (GPT Image 2)

10 Design Guidelines for Satisficing Users

Users will keep taking the first decent parking spot no matter how eloquently you advertise the far corner of the lot. Paint the lines where people already pull in:

1. Make the most prominent action the user’s most likely goal. The biggest button gets clicked on plausibility alone, so it had better be correct. Prime screen real estate is for the probable, not for the profitable.

2. Spend the first 11 characters of every link and heading on meaning. Users judge labels by roughly the first 2 words (per my eyetracking work) and click before they finish reading, so lead with the differentiating words.

3. Strengthen information scent. Link wording must predict the destination specifically enough that the first reasonable guess is also the right one; see my article Information Scent.

4. Set defaults you’d defend in public. Most users keep them, so a default is a decision you make on the user’s behalf, multiplied by millions. Make it in the user’s interest.

5. Put the best option first. Position is destiny in search results and option lists; satisficers rarely reach item 7.

   
   

   

   Users don’t read far down the list of options, especially not on a SERP (search results page). (GPT Image 2)
   

6. Mark a recommended choice and state the criterion. An honest “Most popular” or “Best for small teams” lets users stop searching with confidence.

7. Keep mistakes cheap. Satisficers navigate by trial and error, so give them a prominent Back button and a real undo. Cheap recovery is what makes fast choosing safe.

8. Winnow near-duplicate options. A few well-differentiated choices beat a panoply of lookalikes that nobody will compare anyway. The problem is rarely the size of the catalog; it’s options users can’t tell apart without work you should have done for them.

9. Put consequences before the click, and never booby-trap the fast path. Total cost, renewal terms, and data use belong at the decision point, in plain sight, in readable type. Preselected add-ons and confirmshaming monetize trust exactly once per customer.

10. Test your first design on 5 users before falling in love with it. Designers satisfice too, and iteration is how you raise your aspiration level, one round of user testing at a time.

Conclusion: Make the Lazy Click the Right Click

Good enough is what users choose: the parking spot this morning, your checkout flow this afternoon. No tooltip will talk anybody out of a stopping rule that’s a quarter of a million years old, and you shouldn’t want it to. Meet users where they stop, and make good enough genuinely good.