The Paradox of Choice in the Age of AI: Why More Options Aren't Always Bad

Signup Write New Story The Paradox of Choice in the Age of AI: Why More Options Aren't Always Bad by Prateeka Kamath by Prateeka Kamath | @ prateekakamath Technology, human behaviour and the mechanics of business growth.

Technology, human behaviour and the mechanics of business growth.

I was on Netflix, scrolled through the recommendations, checked a few trailers, read some reviews, and eventually found myself wondering whether I should just rewatch something I’d already seen.

The irony wasn’t lost on me. I had access to thousands of films, a recommendation algorithm supposedly tailored to my taste, and more information than I could reasonably process. Yet choosing something to watch had somehow become a task in itself.

I mean, we’ve all been there, haven’t we?

The same thing happens everywhere… shopping online, choosing a restaurant, planning a holiday or trying to decide which of the hundreds of products claiming to solve our problems is actually worth buying.

We can ask an AI assistant to compare our options, narrow down a list, explain the differences and recommend the best one. We can even tell it what we want in plain English instead of spending half an hour configuring filters.

So why does choosing still feel like work?

I started looking into the psychology of choice, and what I found was more interesting than the familiar advice to simply give people fewer options.

The research suggests that the problem isn’t necessarily how much choice we have. It is how difficult that choice is to make. And AI might be changing that equation in ways we’re only beginning to understand.

Let’s start with something that might surprise you.

The popular version of the paradox of choice suggests that more options lead to worse decisions. Give people too much choice, and they become overwhelmed, struggle to decide and feel less satisfied with what they eventually pick.

It’s a compelling idea. It’s also an incomplete one.

In 2010, researchers Benjamin Scheibehenne, Rainer Greifeneder and Peter M. Todd published a meta-analysis in the Journal of Consumer Research titled Can There Ever Be Too Many Options? A Meta-Analytic Review of Choice Overload.

Rather than examining just one experiment, they analysed 63 experimental conditions from 50 published and unpublished experiments involving 5,036 participants.

Their finding? The average effect of choice overload was virtually zero.

That doesn’t mean choice overload is imaginary. It means that across the experiments they examined, there was no consistent overall effect of simply increasing the number of options. The results varied considerably between studies.

I find this distinction important because it challenges an assumption we’ve accepted almost without question: that fewer options automatically make us happier.

If you love coffee and know exactly what you like, a menu with 30 coffees might be exciting. You can explore different origins, roasts and brewing methods.

But if you’re buying your first coffee machine and don’t know the difference between half the features being advertised, even five options might feel like too many.

The number of options hasn’t changed in any meaningful psychological sense. Your familiarity with the decision has.

In 2015, researchers Alexander Chernev, Ulf Böckenholt and Joseph Goodman published Choice Overload: A Conceptual Review and Meta-Analysis in the Journal of Consumer Psychology.

Their analysis covered 99 observations involving 7,202 participants.

They identified four factors that help explain when larger assortments are more likely to create problems:

This gives us a much more useful way to think about choice.

Sometimes we don’t need fewer options. We need to understand the options better.

Sometimes we don’t need another recommendation. We need to figure out what matters to us in the first place.

And sometimes, yes, we really do need someone to narrow things down.

The trick is knowing which problem we’re trying to solve.

Consider what happens when you want to buy something today.

You can search Google, browse marketplaces, watch YouTube reviews, read Reddit discussions, ask friends, compare prices and check what people on Instagram recommend.

You can spend an entire evening researching a product that you could probably have purchased in five minutes.

Each source offers something useful. One has expert knowledge, another has real-world experiences, and another gives you a comparison you hadn’t considered.

But each source can also introduce new alternatives and new reasons to reconsider your decision.

You start by wanting a pair of headphones. Before you know it, you’re comparing codecs, noise cancellation, battery life, comfort, microphone quality and the relative merits of five brands you had never heard of that morning.

At some point, you have to wonder whether all this research is actually improving your decision.

Or whether you’ve simply made the decision more complicated.

This is where I think the conversation about choice becomes particularly interesting.

The internet didn’t just give us access to more products. It gave us more ways to evaluate them, more people to consult and more opportunities to second-guess ourselves.

And now AI is promising to make sense of all of it.

Here’s where the research gets really interesting.

In 2023, Jungkeun Kim, Jeong Hyun Kim, Changju Kim and Jooyoung Park published a paper titled Decisions with ChatGPT: Reexamining Choice Overload in ChatGPT Recommendations in the Journal of Retailing and Consumer Services.

The researchers conducted five studies examining how people responded to recommendations generated by ChatGPT.

In one experiment, 265 US adults were randomly assigned to receive recommendations containing 3, 10, 20 or 30 songs.

In another, 315 US adults received recommendations containing 3, 30 or 60 songs.

The researchers examined participants’ responses, including their satisfaction with the recommendations, their perceptions of accuracy and their intentions to purchase a recommended song.

The results challenged the conventional expectation that larger recommendation sets necessarily create a negative experience.

Participants responded positively to large sets of ChatGPT-generated recommendations, including sets containing 60 options.

The researchers also found that the source of the recommendations mattered. In their experiments, people responded differently to large recommendation sets from ChatGPT than to similarly large sets from human recommenders or online travel agencies.

Sixty options from a conventional recommendation source can feel like a lot of work. Sixty options from ChatGPT may be received differently.

But let’s be precise about what this means.

These were specific experiments, conducted with particular tasks and participants. They don’t establish that people will always prefer 60 AI recommendations, or that AI removes the difficulties associated with choosing.

What they do suggest is that the number of options alone doesn’t tell us enough about the experience.

The way those options are generated and presented may matter too.

I keep coming back to this distinction because it has implications far beyond online shopping.

When a system presents us with a list, we have to decide how to use it. When a conversational assistant gives us recommendations, we can ask it to explain, reorganise or refine them.

We aren’t just receiving options. We’re interacting with something that can help us interpret them.

Perhaps that’s part of what makes the experience different.

The study itself points to perceived information accuracy as a possible explanation for the positive responses to large AI-generated recommendation sets. Whether the same effect holds across different categories, higher-stakes decisions and newer AI systems is another question entirely.

Maybe the future of recommendations isn’t about having fewer options. Maybe it’s about making a large set of options feel more manageable.

You tell an AI assistant that you want somewhere warm, relatively quiet, within your budget, with good food and easy access to nature.

You could spend the next few hours comparing them. Or you could ask the assistant to narrow the list to five.

Then you could ask it to compare those five based on your priorities.

Finally, you could ask it which one it would recommend and why.

The assistant isn’t simply giving you information. You’re gradually delegating parts of the decision-making process.

First, you delegate the search. Then the filtering. Then the comparison. Potentially, you delegate the final recommendation too.

Each step reduces some of the work you would otherwise have to do.

But it also changes your relationship with the decision.

If the assistant recommends a destination you love, excellent. If it recommends somewhere that looks perfect on paper but feels completely wrong for you, you might wonder whether it really understood what you wanted.

And if it confidently recommends a destination based on outdated information, the consequences could be more than mildly annoying.

This is where another piece of research becomes relevant.

In 2024, Lior Fink, Leorre Newman and Uriel Haran published Let Me Decide: Increasing User Autonomy Increases Recommendation Acceptance in Computers in Human Behavior.

Across three online experiments, the researchers investigated how different forms of autonomy affected people’s willingness to accept algorithmic recommendations.

They examined two mechanisms: giving people a choice between multiple recommendations and allowing them to control the number of recommendations they received.

Their findings showed that providing multiple recommendations increased acceptance, while control over the number of recommendations also influenced the relationship.

In other words, the way control is distributed between a person and a recommendation system can affect whether its suggestions are accepted.

This is a useful reminder that helping someone choose doesn’t necessarily mean making the choice on their behalf.

Sometimes, a person might want one strong recommendation. At other times, they might want three alternatives and the freedom to compare them.

The ability to choose how much help we receive may itself be part of a good recommendation experience.

And that brings me to a question I find particularly interesting: what if the best AI assistant isn’t the one that makes decisions for us, but the one that understands how much help we actually want?

There’s something else worth considering.

We tend to associate convenience with a better experience. If an AI assistant can give us an answer in seconds, that sounds better than spending an hour researching the same question.

But speed isn’t the same as quality. And confidence isn’t the same as correctness.

Imagine you’re choosing a financial adviser.

You could ask an AI assistant to compare advisers based on their credentials, services, fees and public information. That might be an excellent starting point.

But you would still want to verify the information, understand how each adviser is compensated and consider whether their approach suits your circumstances.

A neat recommendation cannot replace every part of that evaluation.

The same principle applies, to varying degrees, across many decisions. The more consequential the choice, the more important it becomes to understand the reasoning behind a recommendation and the limitations of the information being used.

This isn’t an argument against using AI. It’s an argument for designing and using it thoughtfully.

A good recommendation system should help us identify what matters, understand our options and make informed decisions. It shouldn’t simply make us feel that the decision has been taken care of.

There is also an important difference between a recommendation that we accept because we understand it and one that we accept because we don’t want to spend any more time thinking about the problem.

From the outside, both might look like successful recommendations. From the user’s perspective, they’re quite different experiences.

That distinction deserves more attention as AI becomes a more common part of everyday decision-making.

So far, we’ve looked at experiments involving choice overload, ChatGPT recommendations and user autonomy.

But what happens when recommendation systems operate in actual shopping environments?

In 2024, Nurit Nobel published Recommender Systems: Friend (of Choice) or Foe? A Large-Scale Field Experiment in Online Shopping Platforms in the journal Decision.

The study involved 23,165 consumers across two online retailers.

Nobel tested a modified recommendation algorithm that presented the most appealing product results first, followed by results that were progressively less appealing.

The concern was that recommendation algorithms could unintentionally make choice harder by presenting consumers with an unusually attractive set of options, potentially making comparisons more difficult.

The field experiment found no indication that the altered recommendation algorithm harmed the measured choice process or outcomes.

This is particularly useful because it takes the question out of a hypothetical laboratory setting and examines consumer behaviour in real shopping environments.

It doesn’t prove that every recommendation algorithm works well, or that any particular ranking strategy is ideal. It does suggest that the relationship between recommendation design and decision quality is more nuanced than the assumption that algorithms inevitably make choosing harder.

Taken together, these studies point towards a more interesting question than whether we should have more or fewer choices.

How should choices be presented so that people can make decisions that work for them?

I don’t think the answer is to give everyone fewer options.

Sometimes, we genuinely want variety. We might be exploring a new interest, looking for inspiration or trying to discover something we didn’t know existed.

Other times, we want someone to do the filtering for us.

And sometimes, we don’t even know what we’re looking for until someone helps us articulate it.

The challenge is to recognise the difference.

Imagine an AI assistant that begins by asking whether you want a quick recommendation, a shortlist to explore or a comprehensive comparison.

If you want a quick answer, it gives you one and explains why.

If you want to explore, it gives you a broader set of options, organised in a way that makes the differences clear.

If you’re making an important decision, it helps you understand the trade-offs and encourages you to verify information that matters.

The point isn’t to make every interaction more complicated. It’s to make the experience fit the decision.

There is no reason to assume that the ideal number of recommendations is always one, three, five or ten. What works depends on the task, the person’s preferences and the amount of effort they’re willing to invest.

The research on choice overload has been telling us this for years. Research on AI recommendations is beginning to explore what it means when the options themselves are generated and presented by conversational systems.

We still have much to learn about how these experiences affect confidence, satisfaction, trust and the quality of our decisions.

But one thing seems clear: simply making more information available isn’t enough.

We also need to make that information useful.

We began with a simple question: if technology keeps making it easier to find what we want, why does choosing still feel like work?

The research doesn’t offer a single answer.

Having more options doesn’t automatically create choice overload. Large sets of AI-generated recommendations can be received positively in certain situations. And the amount of control people have over the recommendation process can influence whether they accept the suggestions they receive.

These findings complicate the familiar idea that we should simply reduce the number of choices available to us.

Maybe, the real challenge is understanding what makes a decision difficult in the first place.

Sometimes we need fewer options. Sometimes we need better information. Sometimes we need an explanation, a comparison or the freedom to explore alternatives.

And sometimes, we just need to trust ourselves enough to make a decision and move on.

As AI becomes more involved in helping us navigate the world, I think the most interesting question is not whether we should delegate our decisions.

It’s how we can use these systems to make better decisions without losing sight of what matters to us.

Because the purpose of a recommendation was never just to tell us what to choose.

And those are two very different things.

If you’d like to explore the evidence behind this article, these are the original papers.