The line was almost too neat, which should have been the first warning.

There was Jeff Bezos, on a stage in Paris, calmly suggesting that people might drink a little less so the machines could take a little more. “Biological limits are real, but digital potential is infinite,” he supposedly said, before explaining that cooling a data centre mattered more than keeping humans comfortable.

It had the exact blend of cold arithmetic and missing ear we have come to expect from a certain tier of billionaire. Within hours, it had travelled across Reddit, X and Facebook. An Indian news site, The Print, published it as fact.

The problem is, he never said it.

Snopes traced the words to a parody account trading as BPD News, set up in March 2026 to imitate the BBC and flagged (for anyone who paused to look) as satire.

The fact-checking site Lead Stories read the full transcript of his session, all forty-nine minutes of it, and found no such line. The Associated Press had livestreamed the whole appearance. The sentence was simply not there.

And still it was believed, on sight, by a great many people, because the feeling of it rang true even though the words were invented. The fabrication, persuasive precisely because the shoe seemed to fit, tells us almost nothing about the man, and a good deal about how we now read the race to build AI. We treat it as a hushed reordering of whose needs come first, with most of us somewhere near the back.

For all that, nobody checked. Then again, we rarely do.

Not a checking species

We are not, on the whole, given to verifying things. Exploding Topics, a trend-analysis firm, found as much in its AI Trust Gap Report: only eight per cent of people always click through to the sources sitting behind an AI-generated answer. The rest read the summary and carry on.

This holds well beyond viral memes. Researchers at the University of Melbourne and KPMG asked more than 48,000 people how they felt about AI. Most said they used it regularly. Fewer than half said they trusted it. The survey reached dozens of countries, and when set beside an earlier round from before ChatGPT arrived, the mood had soured rather than warmed. Our wariness has climbed without our care keeping pace. Suspicion and scrutiny, it seems, are different muscles.

The machine always sounds sure

The reflex has a name, but the behaviour matters more than the label. Automation bias is the standing preference for the system’s answer over our own hesitation, first documented decades ago in cockpits, where pilots would follow a faulty instrument past the evidence of their own eyes. The same deference now greets the chatbot.

Part of it is effort. Reading a summary is easy, verifying it is work, and we are stingy with the kind of attention that cross-checking demands.

The rest is less flattering. We are softest on the claims that agree with us and hardest on the ones that do not, so a plausible reply pointing in the right direction sails through untouched.

The interface does its bit here too. A large language model delivers a wrong answer in the same tone it uses for a right one: fluent, composed, evenly confident. No furrowed brow, no dip in certainty, no visual tell that the ground has turned soft underfoot. That evenness is a design choice, and it makes the output easier to trust than it should be.

No one is too clever for this

The obvious assumption is that this catches the careless and spares the expert. It does not.

In a 2023 study in the journal Radiology, researchers at University Hospital Cologne showed radiologists a deliberately wrong AI reading of a mammogram.

Among the most experienced of them, accuracy fell from around 82 per cent to 45. Knowing more offered little cover. It mostly gave them further to fall.

It reaches past the lab, too. Late in 2025, Deloitte agreed to refund part of a A$440,000 fee to the Australian government after a report it delivered was found to contain non-existent academic references and a quote pinned to a Federal Court judgment that no judge had written. A Sydney University researcher, Christopher Rudge, caught it.

The thread he pulled was a professor credited with a book she had never authored, on a topic outside her field. Deloitte later disclosed it had used a generative AI tool to help produce the document. This was one of the largest professional-services firms on earth, not a student cutting corners, and the errors came to light only because a single reader went to the footnotes.

The courts have it worst. Damien Charlotin, a researcher at HEC Paris, keeps a running database of legal filings undone by AI-invented citations.

It has passed 1,700 cases and grows faster than he can log them.

The fakes are convincing at a glance: plausible case names and real-sounding judges, attached to rulings that never happened. In one much-quoted instance, a lawyer grew uneasy about his citations and asked the chatbot whether the cases existed. It assured him they did. The tool that produced the error also vouched for it. Remember that the next time someone proposes we simply use AI to check the AI. When researchers tested that very idea in PNAS in 2024, the fact-checks sometimes left people trusting true stories less and dubious ones more.

What Bezos actually said

Which returns us to Paris, and to the better question the fake quote buried. The danger of the invented sentence was not that it embarrassed Bezos. It is that a week of righteous argument burned itself out on words nobody had spoken, while the actual debate carried on, with far fewer people watching.

Bezos did talk about AI at VivaTech, and the true version is milder but says more than the hoax ever did. He argued that advanced AI will bring a labour shortage rather than mass unemployment, on the reasoning that better tools let people find more problems worth solving. Turning to his new venture, Prometheus, he described it as building an “artificial general engineer”. He raised the water question too, glancingly.

The number behind it needs no villain’s monologue: Amazon has disclosed that its data centres got through 2.5 billion gallons of water last year.

Let them have data centres

Take the invented lines at their word for a moment, and audit the world they imagine. The satire proposed a swap: our water for their cooling, our power for their scale. What would it actually cost? Three questions, three numbers.

• How much water do they need, and where does it come from?

Amazon’s 2.5 billion gallons is one company in one year. A single average data centre requires something in the region of 300,000 gallons a day, a fair share of it drawn in places that were short of water to begin with.

• How much power is required, and who queues behind it?

In 2024 they used roughly 415 terawatt-hours of electricity, about 1.5 per cent of world supply. The International Energy Agency expects that to more than double by 2030, to something close to what all of Japan runs on in a year.

• Finally, who pays to build it?

Data centres sit among the most heavily subsidised businesses around. Virginia’s tax break alone costs the state some $1.6 billion a year. Whether those costs land on household bills is contested: consumer groups and several US senators say they do, while a study commissioned by Amazon argues the centres pay their own way and can even ease bills a little. Nobody yet knows for certain, and the argument is far from over.

The made-up sentence got its week of outrage. The planning hearings where these questions are actually settled tend to get a wet Tuesday and an empty public gallery.

What is AI really costing the planet?

The market has noticed

If the episode showed how little we check, the business world has drawn the obvious conclusion and started building on it. Analysts valued the market for AI-detection tools at somewhere around half a billion dollars in 2025, and expect it to multiply several times over by the early 2030s, with the sharpest growth in software that spots deepfakes and synthetic media.

The logic is neat: as machines flood the world with plausible text and images, other engineered solutions will sort the true from the invented. There is one snag.

These detectors tend to lose accuracy as the models they police improve, and they carry a stubborn rate of false alarms, so the thing being sold as a fix keeps needing fixing.

The people who did this work by hand, meanwhile, are being shown the door. Human fact-checking grew for a decade, from about 110 projects worldwide to more than 450, before the trend turned. Then, in early 2025, Meta ended its US fact-checking programme, which had funded a large share of them, and the field tipped into retreat.

For the first time, roughly three times as many projects closed as opened. The International Fact-Checking Network found three-quarters of those it surveyed financially vulnerable or in outright crisis. All the same, readers have not lost interest. Most of these organisations reached larger audiences than the year before, even as the money drained away.

The appetite for a settled answer keeps rising while the specialists trained to produce one lose their funding, and the budget flows instead to tools that struggle to keep pace with the problem they were built to solve.

The job that stays ours

The fake quote itself still did damage. It caught on because it told a broadly true story in false words, and a false story that feels true is a peculiarly modern trap. The real stuff, however, doesn’t spread like that. There is no shareable image for a terawatt-hour, a tax abatement or a 237-page report. That is homework, and it mostly goes undone.

The obvious candidate for that task is the same tool producing the claims, and yet it makes a poor judge of its own output. So, the checking stays with us. A better interface would help: one that wore its uncertainty as plainly as its confidence would be a harder thing to take on trust, and more often than not, that is exactly what we need.

If you enjoyed this, follow me on Medium for more on design, psychology and technology.

References

Amazon. (2025). Do data centers raise electricity bills? What an independent study found. https://www.aboutamazon.com/news/sustainability/data-centers-electricity-bills-grid-power-amazon

Amazon. (2026). Amazon’s data centers are 7x more water-efficient than the industry average. https://www.aboutamazon.com/news/sustainability/amazon-data-center-water-usage

Charlotin, D. (2026). AI Hallucination Cases [Database]. HEC Paris. https://www.damiencharlotin.com/hallucinations/

Consumer Reports. (2026). AI data centers: Big tech’s impact on electric bills, water, and more. https://www.consumerreports.org/data-centers/ai-data-centers-impact-on-electric-bills-water-and-more-a1040338678/

DeVerna, M. R., Yan, H. Y., Yang, K.-C., & Menczer, F. (2024). Fact-checking information from large language models can decrease headline discernment. Proceedings of the National Academy of Sciences, 121(50), e2322823121. https://doi.org/10.1073/pnas.2322823121

Dratsch, T., Chen, X., Rezazade Mehrizi, M., Lorenz, A., Kröger, J. R., Persigehl, T., Pinto dos Santos, D., & Baeßler, B. (2023). Automation bias in mammography: The impact of artificial intelligence BI-RADS suggestions on reader performance. Radiology, 307(4), e222176. https://doi.org/10.1148/radiol.222176

Duke Reporters’ Lab. (2026). 2026 census: Fact-checking losses continue amid funding pressure, but most projects persist. https://reporterslab.org/2026/06/12/2026-census-fact-checking-losses-continue-amid-funding-pressure-but-most-projects-persist/

Exploding Topics. (2025). AI Trust Gap Report. https://explodingtopics.com

Gillespie, N., Lockey, S., Ward, T., Macdade, A., & Hassed, G. (2025). Trust, attitudes and use of artificial intelligence: A global study 2025. The University of Melbourne and KPMG. https://doi.org/10.26188/28822919

Good Jobs First. (2026). Even cloudier with a greater loss of spending control: How data center tax abatements undermine public budgets. https://goodjobsfirst.org/even-cloudier-with-a-greater-loss-of-spending-control-how-data-center-tax-abatements-undermine-public-budgets/

Grand View Research. (2026). AI detector market size, share & trends report, 2026–2033. https://www.grandviewresearch.com/industry-analysis/ai-detector-market-report

International Energy Agency. (2025). Energy and AI. https://www.iea.org/reports/energy-and-ai/executive-summary

International Fact-Checking Network. (2026). State of the fact-checkers report 2025. Poynter Institute. https://www.poynter.org/wp-content/uploads/2026/03/2026-State-of-Fact-Checkers-4.pdf

Lead Stories. (2026). Fact check: Jeff Bezos did NOT say water should be used to fuel AI rather than for human consumption. https://leadstories.com/hoax-alert/2026/06/fact-check-jeff-bezos-did-not-say-water-should-be-used-to-fuel-ai-rather-than-for-human-consumption-in-public-remarks-at-2026-vivatech-conference-in-paris.html

The Register. (2025). Deloitte refunds Australian government over AI in report. https://www.theregister.com/2025/10/06/deloitte_ai_report_australia/

Snopes. (2026). Did Jeff Bezos say human water consumption is limiting AI’s potential? https://www.snopes.com/fact-check/bezos-water-consumption-ai-quote/

Stillman, J. (2025, July 18). Are you too trusting of AI answers? 92 percent of people don’t check it for accuracy. Inc. https://www.inc.com/jessica-stillman/are-you-too-trusting-of-ai-answers-92-percent-of-people-dont-check-it-for-accuracy/91209990

Waziri, I. J. (2026, June 19). Amazon founder Jeff Bezos says human water consumption is limiting AI’s potential. ThePrint. https://theprint.in/feature/jeff-bezos-water-consumption-amazon-ai-potential/2964266/

World Economic Forum. (2025). How data centres can avoid doubling their energy use by 2030. https://www.weforum.org/stories/2025/12/data-centres-and-energy-demand/

Did you verify that — or was it just too easy to believe? was originally published in UX Collective on Medium, where people are continuing the conversation by highlighting and responding to this story.

Watch the podium closely at almost any Olympic Games and you will catch something that does not instantly add up. The silver medallist, second-best on the planet at their event that day, often looks faintly gutted. The bronze medallist, standing one step lower, tends to be beaming. By the result, silver did better. So why does bronze so often look like the happier person?

It is not just the cameras catching a bad moment. In 1995, Victoria Medvec, Scott Madey and Thomas Gilovich went through footage from the earlier Barcelona Summer Olympics and rated the faces of medallists, both at the moment they finished and later on the stand. Writing in the Journal of Personality and Social Psychology, they reported that bronze medallists looked reliably happier than the silver ones who had beaten them. The people who came closer to the top felt worse about it.

The reason is the comparison each athlete cannot help but reach for. Silver looks up at the gold that slipped away and thinks about how nearly it was theirs. Bronze looks down at fourth place, the near miss with nothing to show for it, and feels fortunate to be holding anything at all. Same podium, a very different film running in each head.

Counterfactual thinking is the label for it, the mind’s habit of playing the “what if” replay. And it has held up since: a more recent study ran facial-expression software over medal-stand photographs from several Olympic Games and found the same lopsided result.

Which sets up an awkward question for anyone who has ever been told to aim high. If getting close to the summit can leave you flatter than finishing further from it, should you really be aiming to be the best at all?

Two things wearing the same coat

The answer starts with a distinction we tend to skip. “Not being the best” is really two things bundled into one. One of them is the plain fact of it, a reasonably accurate read of where you stand. The other is the feeling that comes with it, the sting of measuring yourself against someone better. The first can be useful. The second is where things get slippery. For the silver medallist, the trouble was never what they knew, but the yardstick they chose.

The two can come apart. You can know perfectly well that you are middling at something and feel fine about it, the way most of us regard our own singing. You can also be objectively excellent and feel wretched. This is why so much of the usual advice misses its mark. “Be realistic” is aimed at the facts; “stop comparing yourself” at the sting. They are different problems, and they take different cures. The rest is learning to tell which one you are dealing with, and when to listen to it.

Reading yourself

Take the case in favour first. Knowing, accurately, that you are not the best is what lets you improve. After all, you cannot close a gap you cannot see.

This is the grain of truth buried inside the most over-quoted finding in popular psychology. In 1999, Justin Kruger and David Dunning reported that people who were worst at a task tended to wildly overrate themselves, while the truly skilled were more measured. The internet boiled this down to a meme: the incompetent are too clueless to spot it. What lies underneath is messier than that, and a good deal less settled. Several researchers now suspect much of the effect is a statistical illusion, a trick of the numbers rather than a real feature of the mind. One analysis even showed you can reproduce the famous chart from random data, which is hard to square with it saying anything about real people.

Even so, a smaller point stands. As you get better at something, you also get better at judging how good you are. Reading your own weaknesses is a skill in its own right, and it happens to be the one that makes progress possible. A designer who can see precisely where their work falls short has something to fix. One who thinks it is all flawless has nowhere left to go.

Faking it, making it

Feeling like a fraud seems a pure liability. Its story begins in a university counselling room in 1978, where the psychologists Pauline Rose Clance and Suzanne Imes kept meeting the same thing in high-achieving women: accomplished people privately certain they had fooled everyone, and that exposure was only a matter of time. One woman, told by her chairman that her paper was among the best he had ever read, assumed she had simply got away with something. The researchers called it the impostor phenomenon. It has since walked out of the clinic and taken up residence just about everywhere, from boardrooms to group chats.

For a long time the assumption was that these thoughts do nothing but damage. Then Basima Tewfik, a professor at MIT Sloan, looked more closely. In a 2022 paper in the Academy of Management Journal, she pried apart the thought itself (the belief that others rate you more highly than you rate yourself) from the distress usually assumed to come stapled to it. Across four studies involving more than 3,600 people, including surveys at an investment advisory firm and work with physicians in training, she found that those who had these impostor thoughts more often were rated by others as more interpersonally effective, with no dip in actual competence.

Her explanation is a kind of instinctive compensation. If you suspect you are not quite as good as people assume, you tend to turn outward, listen harder and read the room more carefully, and that attentiveness comes across as skill. In any work that leans on critique and collaboration, the person who feels like the weakest voice on the team may be the most valuable one on it.

Tewfik was careful not to overstate this, and neither should we. Those thoughts still chip away at self-esteem, and she was clear that they are not something to cultivate on purpose. But the neat story that believing yourself a fraud is simply and only bad for you does not survive her data.

Prove them wrong, but choose your doubter

Being underestimated carries its own strange charge. Samir Nurmohamed, of the Wharton School, studies what he calls underdog expectations, the sense that other people expect you to fail. Intuition says being written off is demoralising, and often it is. Yet he found that low expectations from others could sharpen performance rather than blunt it, powered by a very human engine, the urge to prove them wrong.

The crucial part comes with a condition: the effect flips depending on who is doing the doubting. When the doubter carries little credibility, being underestimated lights a fire. When it is someone whose judgement you respect, their low expectations tend to seep in and become self-fulfilling, dragging performance down with them. That darker version has a name of its own, the Golem effect, where being expected to fail helps make the failure happen. The doubt is identical; only the outcome flips, and that hinges on whether you buy the source.

The way you narrate your own position matters too. In 2021, Nurmohamed worked with 330 unemployed job seekers recruited from two re-employment centres in Pennsylvania, many of whom had faced discrimination. Those coached to cast themselves as underdogs, recalling times they had defied low expectations, came away more confident about landing work, and went on to fare better, than those who told “favourite” stories in which success had always been expected of them. The account you give of not being the obvious pick can, it seems, become fuel.

Small fish, big pond

Now the opposite case, where knowing starts to hurt. Put an able person in a room full of stronger people and, on average, something in them deflates.

Educational psychologist Herbert Marsh has spent much of his career on this, under a memorable name, the big-fish-little-pond effect. The evidence is consistent and a touch bleak.

Take two students of identical ability. Place one in a high-flying selective school and the other in an ordinary one, and the student surrounded by standout peers will tend to rate their own ability lower. He confirmed it across 26 countries using international student data, drawing on more than 100,000 fifteen-year-olds. Then a 2018 meta-analysis, gathering up decades of studies, showed the pattern held broadly across ages and settings.

That lowered opinion of yourself does not stay put. How good you believe you are feeds directly into how well you go on to do, and how high you dare to aim next. This is the gifted-programme paradox: the very move meant to stretch a talented child, dropping them among equally talented peers, can shrink their belief in themselves. The pond grew, and everyone in it felt smaller. It is the silver medallist’s problem again, only stretched across years rather than one afternoon on a podium.

What tips the balance

Line these findings up and a shape appears. Knowing you are not the best is neither medicine nor poison by itself. A few things decide which it becomes.

The direction you look does the most. The silver and bronze stood inches apart and felt a gulf between them, because one looked up at what was missed and the other looked down at what was escaped. Looking up is not automatically corrosive, though. See someone ahead of you and think “I could get there,” and it pulls you forward. Let the same sight tip into “I never will,” and it drags you under. One person to admire, two ways to take it.

The rest is a matter of framing, of who is doing the judging, and of where you happen to be. Treated as a final verdict, “I am not the best” is a wall; treated as a starting line, it is a heading. Doubt from someone you respect sticks, while doubt from someone you dismiss slides off, which was Nurmohamed’s point. Beyond that, your surroundings can matter as much as any of it, which is where this stops being purely personal and becomes something other people build.

If you build the pond

Everything so far has been about the individual. But Marsh’s finding is really about the setting, not whoever is dropped into it. The same person thrives or wilts depending on where they land.

Think about an ordinary design career. Portfolio galleries, endless feeds of other people’s most polished work, the nagging suspicion that everyone else is shipping something lovelier than you are. That is upward comparison of precisely the punishing sort, all distance and no path to close it. A junior designer scrolling that stream is a small fish being shown an ocean of big ones, all day, with no visible route from here to there.

Teams carry the same hazard. Stack a group entirely with stars and the research suggests some of them will privately mark themselves down, even while their output stays strong. The answer is not to lower the bar, but to make the climb the point, not the ranking, so that “I am not the best here yet” reads as a spot on a journey instead of a ruling on your worth.

So far, so internal. But most of what designers make ends up in front of strangers, and a fair amount of it deliberately manufactures the feeling of not being the best in them. Leaderboards. Streaks. Follower counts. “You are in the 62nd percentile.” Social proof that parades everyone ahead of the user. They are, quite literally, choosing the comparison direction on people’s behalf, and the evidence says that choice is rarely neutral.

Strava is the clearest case. Its crowns, King and Queen of the Mountain, go to whoever has posted the fastest time a segment has ever seen, so almost everyone who rides that hill is a permanent runner-up, chasing a record they will never beat. In 2020 the company added Local Legends, a crown that ignores speed altogether. It goes to whoever rides a segment most often over a rolling 90 days. Ability drops out of the picture, dedication is the only currency, and it changes hands constantly. What shifted is what riders are judged on, and that was a design choice.

Duolingo keeps the competition but shrinks it. Its weekly leagues sort you into a pool of only about thirty learners, matched to your own pace, so someone doing a few lessons a week is not thrown in with the ones grinding a hundred. Almost anyone can finish near the front just by turning up, and that is no accident.

The gentler move is to drop the crowd altogether. Apple’s activity rings close against a goal you set yourself, so nobody is ahead of you at all, because the only benchmark is your past self. It all points backward, at progress made, rather than forward at the gap to the top.

Peloton lets the rider choose outright. Its leaderboard can show the whole class, or only your age group, or only people you follow, or, with a “Just Me” filter, nobody at all, so you race your own earlier rides. One control, and today becomes a contest with the whole field or with last week’s self, whichever the rider prefers.

Point someone towards an unreachable top with no way to climb, and you hand them the silver medallist’s flatness. Offer a downward glance at how far they have come, or a rung within reach, and you hand them the bronze medallist’s lift. Same data, different film, and the business decides which one plays.

Kinder by design

Clear feedback still helps. Seeing where you come up short is how anyone improves, whether you are the user or the maker. The skill is in what you weigh it against, and a few guidelines can start paving the way:

• Default to the user’s own past. It works for almost everyone; a public ranking works only for the few at the top.
• If you must rank, make first place reachable and mark the step just above where someone stands. A ladder people can climb beats a peak only the elite will ever hold.
• Reward progress, not just position. Someone shown the ground they have already covered gets the bronze medallist’s view, the one that lifts.
• Let people pick their crowd. A filter that shrinks the field, or clears it, returns that choice to the person it affects.

So, should you?

Back to the question in the title. Should you aim to be the best?

Aiming high is fine. The trap is measuring yourself against the summit alone, so that anything short of it registers as failure. The silver medallist and the small fish are never let down by the facts. What does the damage is the yardstick, and it can be changed, chosen, and, for those who design things, handed to other people with a bit of care.

Knowing you are not the best can be the most useful thing you know about yourself. It shows you where to go. It turns you outward. It can even be the spark that makes you prove a doubter wrong. Or it can be the slow weight that talks a capable person out of reaching at all. The knowledge is identical in every case; whether it lifts you or flattens you turns entirely on what you set beside it.

If you enjoyed this, follow me on Medium for more on design, psychology and technology.

References

Clance, P. R., & Imes, S. A. (1978). The impostor phenomenon in high achieving women: Dynamics and therapeutic intervention. Psychotherapy: Theory, Research & Practice, 15(3), 241–247. https://paulineroseclance.com/pdf/ip_high_achieving_women.pdf

Fang, J., Huang, X., Zhang, M., Huang, F., Li, Z., & Yuan, Q. (2018). The big-fish-little-pond effect on academic self-concept: A meta-analysis. Frontiers in Psychology, 9, 1569. https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2018.01569/full

Fix, B. (2022). The Dunning-Kruger effect is autocorrelation. Economics from the Top Down. https://economicsfromthetopdown.com/2022/04/08/the-dunning-kruger-effect-is-autocorrelation/

Gignac, G. E., & Zajenkowski, M. (2020). The Dunning-Kruger effect is (mostly) a statistical artefact: Valid approaches to testing the hypothesis with individual differences data. Intelligence, 80, 101449. https://doi.org/10.1016/j.intell.2020.101449

Hedgcock, W. M., Luangrath, A. W., & Webster, R. (2021). Counterfactual thinking and facial expressions among Olympic medalists: A conceptual replication of Medvec, Madey, and Gilovich’s (1995) findings. Journal of Experimental Psychology: General. https://pubmed.ncbi.nlm.nih.gov/33166162/

Kruger, J., & Dunning, D. (1999). Unskilled and unaware of it: How difficulties in recognizing one’s own incompetence lead to inflated self-assessments. Journal of Personality and Social Psychology, 77(6), 1121–1134. https://doi.org/10.1037/0022-3514.77.6.1121

Marsh, H. W., & Hau, K.-T. (2003). Big-fish-little-pond effect on academic self-concept: A cross-cultural (26-country) test of the negative effects of academically selective schools. American Psychologist, 58(5), 364–376. https://pubmed.ncbi.nlm.nih.gov/12971085/

Medvec, V. H., Madey, S. F., & Gilovich, T. (1995). When less is more: Counterfactual thinking and satisfaction among Olympic medalists. Journal of Personality and Social Psychology, 69(4), 603–610. https://doi.org/10.1037/0022-3514.69.4.603

Nurmohamed, S. (2020). The underdog effect: When low expectations increase performance. Academy of Management Journal, 63(4), 1106–1133. https://doi.org/10.5465/amj.2017.0181

Nurmohamed, S., Kundro, T. G., & Myers, C. G. (2021). Against the odds: Developing underdog versus favorite narratives to offset prior experiences of discrimination. Organizational Behavior and Human Decision Processes, 167, 206–221. https://www.sciencedirect.com/science/article/pii/S0749597821000479

Tewfik, B. A. (2022). The impostor phenomenon revisited: Examining the relationship between workplace impostor thoughts and interpersonal effectiveness at work. Academy of Management Journal, 65(3), 988–1018. https://journals.aom.org/doi/abs/10.5465/amj.2020.1627

Should you really aim to be the best? was originally published in UX Collective on Medium, where people are continuing the conversation by highlighting and responding to this story.

A few years ago, buying an office chair meant suffering for it. You opened a dozen tabs, read strangers bickering about lumbar support, squinted at star ratings of dubious origin, and eventually committed to something while quietly suspecting you had chosen wrong. The work was tedious. It was also, in its small way, yours. You did the comparing.

Now you type a sentence into a chatbot and a chair surfaces at the top of the reply, already picked, with three tidy reasons attached. No tabs. No bickering strangers. No low hum of doubt that a better option is buried two pages down. The weighing-up has happened somewhere unseen, performed by something you cannot easily question, and returned to you as a single confident answer.

This is not a niche habit. In a January 2026 survey of US shoppers by Clutch, seven in ten said they now use AI somewhere in the buying process, and 65% turn to it to research a product before they spend a thing. Most still want the final tap to be their own: only 4% said they would let the AI go ahead and buy the item for them. We have handed over the digging, then, without quite handing over the wallet.

That difference is the interesting part. We are not delegating the purchase. We are delegating everything that comes before it, the part where we used to sift and second-guess and form a view. The more that slips offstage, the more it matters who is waiting in the wings.

Because somebody is. The same shift has already begun reshaping the work of the people who make the content. They now have to write and build as much for the crawler as for any human landing on the page. Those at the other end of it, the ones still doing the buying, are changed too, just less visibly.

A whole industry has grown up around shaping what the machine suggests. The reasons we trust a neat conclusion more readily than a messy list of links are oddly specific to how our minds work. The line between a helpful recommendation and a paid placement, meanwhile, grows fainter by the month. Worth knowing, surely, what we signed up for when we stopped opening tabs.

Where the legwork went

For three decades, searching meant being handed a list. You typed a few words, the engine served up its ten blue links, and the sorting was left to you. The ranking was a suggestion rather than a ruling. You could distrust it, scroll past it, open the fourth link because the first three smelled of advertorial.

A generated answer removes that step. Ask which cordless vacuum to buy, and you get a recommendation, not a directory. The shortlist you would once have assembled yourself arrives ready-made, and often it holds just one name. The same has happened to reviews. Wading through them to reach your own opinion used to be part of the ritual; now the machine boils them down for you.

The behaviour has moved faster than almost anyone predicted. Klaviyo’s survey of nearly 8,000 consumers across eight countries late last year found that 41% had bought something an AI recommended in the previous six months, with a further 27% nudged towards a product they then went away to research. A separate study by the fraud-prevention firm Riskified, covering more than 5,000 shoppers worldwide, put the share now using AI somewhere in their shopping at close to three-quarters. The tool has stopped being a curiosity at the edge of the process. It is becoming the front door.

Retailers have followed the traffic. Amazon’s assistant, used by more than three hundred million people last year, no longer merely answers a question about a product but runs searches, balances your budget against your constraints, and returns a favourite. A growing list of other brands now lets customers browse and order from inside a chatbot rather than on their own sites. The shopfront is migrating into the chat window, which often replies with a verdict.

The comfort of being told

A single answer lands easily for reasons that have little to do with whether it is any good.

Comparison is work. Holding several options in your head, scoring each against the rest, accepting that you might still be wrong: all of it costs effort, and we are built to spend as little of that as we can. A confident recommendation settles it for us. It folds a sprawling problem into one line, and the relief is immediate, whatever the merits of the pick.

There is a long line of research suggesting that abundance does not just tire us, it can keep us from choosing at all. In an experiment that has become a touchstone, two researchers set up a tasting table in an upmarket California grocery and stocked it on some days with six jams and on others with twenty-four. The larger spread drew the bigger crowd: 60% of passers-by stopped, against 40% for the smaller one.

Yet of those who paused at the big display, only 3% bought a jar, while almost a third did at the small one. Those faced with fewer were happier afterwards with what they took home.

Later work has shown the effect is not universal and depends on the setting, but the core finding has held up well enough to change how shops lay out their shelves. Seen in that light, an algorithm that returns one option does more than spare you the bother. It lifts the paralysis that an excess of choice can bring, and sets the decision in front of you already made.

Facing fewer options is one half of the comfort. The other is how much we defer to the source. Psychologists have a blunt name for that pull: automation bias, the habit of rating a system’s output above our own because a machine produced it. The snag is that the certainty a clean answer brings is no guarantee the response is correct. It is proof the question has stopped feeling difficult.

Julie Geller, a research director at Info-Tech, put the brakes on adoption neatly when she told CX Dive the real barrier was “psychological, not technical.” We hold back not because the tool cannot choose for us, but because some part of us is not ready to stop choosing.

The snow shovel and the scarf

When we do extend that trust, we are surprisingly particular about it.

The clearest map of this comes from Chiara Longoni and Luca Cian, whose work in the Journal of Marketing named what they called the “word-of-machine effect.” People lean towards an algorithm’s pick when a purchase is practical and measurable, and pull away when the choice is sensory or personal. We will take the computer’s word on a snow shovel, to borrow the usual illustration, while bristling at its view on a fashionable scarf. The belief underneath is that AI is sharp on function and deaf to feeling.

That instinct runs alongside a broad, level-headed scepticism. None of this means we have handed over our discernment wholesale. Confidence in these systems is patchy, and on the evidence fairly clear-eyed. A YouGov survey last summer found two in five Americans place no faith at all in AI shopping assistants.

Gartner, in research that September, put it more bluntly for AI search in particular: about half do not trust the answers and summaries on offer. The Clutch figure is the sharpest of the three, with just 17% of shoppers willing to act on a recommendation without checking it for themselves. This caution is widely held. Yet it has not dented the habit, because the people who would balk at a machine spending their money are quite happy to let it draw up the shortlist.

The contradiction gets stranger still. We are growing warier of words we suspect a chatbot wrote, while becoming more relaxed about products it tells us to buy. One person can hold both reflexes at once. The doubt that clings to AI’s prose seems to spare its shopping advice, perhaps because it reads as a service rather than a voice, and we do not scan it for sincerity the way we scan a sentence.

A thumb on the scale

If the answer carries this much weight, who gets to shape it? No longer the search engine alone.

A new craft has formed around being the name the machine mentions, traded under acronyms as ungainly as generative engine optimisation (GEO) and answer engine optimisation (AEO). The mechanics belong to another discussion, but the gist is that the prize has changed. Ranking first matters less than being the source a generated reply draws from. A 2025 study of AI search found these systems lean heavily on third-party, authoritative coverage rather than a brand’s own marketing, which shifts the work of persuasion from selling yourself to being vouched for by others.

For most of its short life, the chatbot reply has carried no advertising, which is a large reason it felt trustworthy. That is ending. In early 2026, OpenAI began showing labelled sponsored messages beneath ChatGPT’s responses, with rival assistants doing likewise, while insisting the paid content leaves the actual answer untouched. Perhaps it does. The deeper problem is that a sponsored suggestion and an honest one sound exactly alike. Both arrive in the calm, authoritative tone we have been trained to trust.

We spent two decades learning to discount the search ad, mostly because it announced itself as one. A recommendation that reads like considered advice offers no such tell.

When it buys for you

All of this still assumes one step belongs to you. You read what it offers, decide whether to trust it, and act on it or not. That step is the one the industry is now trying to absorb.

OpenAI tried the boldest version first. Late in 2025 it built a checkout into ChatGPT, so a recommended item could be bought on the spot. It did not take: few merchants signed up, shoppers barely used it, and within months the company pulled back to focus on discovery instead.

But the direction is not really in doubt. In January, at the National Retail Federation’s annual show, Google launched a commerce standard of its own. Behind it are retailers like Walmart and Target, payment firms like Stripe and Visa, and a couple of dozen others besides. The card networks have gone further still, building the rails so that an agent can settle up on your behalf, your authorisation granted in advance. Amazon’s assistant, the same one already narrowing your options, now buys as well: set a price, and it will purchase the moment the item falls to it. McKinsey reckons this kind of trade could be worth three to five trillion dollars a year by 2030.

Earlier, there was a hint that the final tap would stay ours. That expectation is being designed out. The infrastructure now assumes the agent will act, and asks only that you set the terms once, ahead of time. The question is shifting from whether you would let it buy to whether you would notice that it had.

Here the thumb on the scale becomes hardest to spot. When the pick and the purchase are a single motion, there is no list to set against it, no alternatives to see, no pause in which you might have looked twice. You cannot compare what you were never shown, and before long you may not be the one pressing buy.

The muscle we stop using

The commercial worry is not the only one.

Each time the shortlist arrives finished, a small piece of deliberation goes unrehearsed. The chance discovery of the fourth result, the thing you find while looking for something else, the slow education of sorting options yourself. None of these is a feature the machine is trying to strip out, yet all of them thin when the middle of the job is done for you. A 2025 systematic review in the journal Electronic Commerce Research noted that the speed AI brings to buying can carry a felt loss of control, a sense of the decision slipping slightly beyond reach.

There is a market-level echo too. When millions of shoppers are funnelled towards whatever the model names, spending pools around a shrinking set of winners. The brand that gets cited gets bought, and the one left unmentioned drops from view, not because it lost a fair contest but because no contest took place. Variety tends to dwindle once the choosing is centralised.

This is not a plea to return to a dozen open tabs and a war of conflicting reviews. It is a note about what a convenience costs once we forget we lean on it.

Back to the chair

The office chair, in fairness, is precisely the sort of purchase the machine handles well. It is practical, comparable and faintly dull, the snow-shovel end of the spectrum, and outsourcing it frees attention for choices that deserve more of it. Used that way, an AI tip is a gift, not a threat. The argument was never that we should do everything the slow way again.

The point is finer than that. A recommendation feels like advice, yet it is more and more often a position someone has paid for, or at least one moulded by an industry straining to be the answer. Holding that in mind, and keeping our own judgement in working order for the purchases that earn the effort, is the whole of the task. A shortlist of one is a fine thing to be handed. It is a poorer thing to mistake for the only list there ever was.

If you enjoyed this, follow me on Medium for more on design, psychology and technology.

References & Credits

Amazon. (2025). Amazon’s next-gen AI assistant for shopping is now even smarter, more capable, and more helpful. https://www.aboutamazon.com/news/retail/amazon-rufus-ai-assistant-personalized-shopping-features

Amazon. (2026). Meet Alexa for Shopping, your personalized, agentic AI assistant on Amazon. https://www.aboutamazon.com/news/retail/alexa-for-shopping-ai-assistant

Brady, M. (2025). AI search and summaries distrusted by about half of consumers: Gartner. CX Dive. https://www.customerexperiencedive.com/news/ai-search-summaries-distrusted-consumers-gartner/759373/

Chen, M., Wang, X., Chen, K., & Koudas, N. (2025). Generative Engine Optimization: How to Dominate AI Search. arXiv. https://arxiv.org/abs/2509.08919

Clutch. (2026). 65% of Consumers Use AI to Research Products Before Making a Purchase. https://www.businesswire.com/news/home/20260122526477/en/Clutch-Report-65-of-Consumers-Use-AI-to-Research-Products-Before-Making-a-Purchase

CNBC. (2026). OpenAI revamps shopping experience in ChatGPT after struggling with Instant Checkout offering. https://www.cnbc.com/2026/03/24/openai-revamps-shopping-experience-in-chatgpt-after-instant-checkout.html

Electronic Commerce Research. (2025). The impact of artificial intelligence on consumer behavior towards brands: a systematic review. Springer. https://link.springer.com/article/10.1007/s10660-025-10063-7

Goldman, S. (2026). ChatGPT tests ads as a new era of AI begins. Fortune. https://fortune.com/2026/01/16/openai-to-begin-putting-ads-in-chatgpt/

Google. (2026). New tech and tools for retailers to succeed in an agentic shopping era. https://blog.google/products/ads-commerce/agentic-commerce-ai-tools-protocol-retailers-platforms/

Iyengar, S. S., & Lepper, M. R. (2000). When Choice is Demotivating: Can One Desire Too Much of a Good Thing? Journal of Personality and Social Psychology, 79(6), 995–1006. https://business.columbia.edu/faculty/research/when-choice-demotivating-can-one-desire-too-much-good-thing

Klaviyo. (2026). Consumers are turning to AI to discover and buy products. https://www.klaviyo.com/newsroom/ai-persona-research

Longoni, C., & Cian, L. (2022). Artificial Intelligence in Utilitarian vs. Hedonic Contexts: The “Word-of-Machine” Effect. Journal of Marketing, 86(1), 91–108. https://journals.sagepub.com/doi/abs/10.1177/0022242920957347

Mastercard. (2025). Mastercard unveils Agent Pay, pioneering agentic payments technology to power commerce in the age of AI. https://www.mastercard.com/news/press/2025/april/mastercard-unveils-agent-pay-pioneering-agentic-payments-technology-to-power-commerce-in-the-age-of-ai

McKinsey & Company. (2025). The agentic commerce opportunity: How AI agents are ushering in a new era for consumers and merchants. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-agentic-commerce-opportunity-how-ai-agents-are-ushering-in-a-new-era-for-consumers-and-merchants

Mosier, K. L., Skitka, L. J., Burdick, M. D., & Heers, S. T. (1996). Automation Bias, Accountability, and Verification Behaviors. Proceedings of the Human Factors and Ergonomics Society Annual Meeting. https://journals.sagepub.com/doi/10.1177/154193129604000413

OpenAI. (2025). Buy it in ChatGPT: Instant Checkout and the Agentic Commerce Protocol. https://openai.com/index/buy-it-in-chatgpt/

Perez, S. (2026). ChatGPT rolls out ads. TechCrunch. https://techcrunch.com/2026/02/09/chatgpt-rolls-out-ads/

Walk-Morris, T. (2025). AI shopping assistants have a trust problem. CX Dive. https://www.customerexperiencedive.com/news/ai-retail-shopping-assistants-consumer-trust/753705/

Wassel, B. (2025). Consumers embrace AI for research, but few have allowed it to make purchases. CX Dive. https://www.customerexperiencedive.com/news/consumers-ai-research-few-purchase/803156/

A shortlist of one: how AI became our shopping adviser was originally published in UX Collective on Medium, where people are continuing the conversation by highlighting and responding to this story.