← Research
Research

Running out of messages makes you worse

The obvious part is hitting the limit. The part that costs you happens before that.

Last reviewed: 6 September 2026

The four behaviours, the scarcity research that explains the mechanism and does not transfer cleanly, why the first thing dropped is the thing with the evidence behind it, and three fixes.

Question this page answersQuestion this page partly answersAll 811 questions this research covers

Every free tier has a limit. Hitting it is annoying and obvious. What is not obvious is what the limit does to you before you reach it: as messages start to feel scarce, you take the first answer, stop pushing back, ask rushed and vaguer questions, and save the good tool for something important while doing the important thing badly with a worse one. Scarcity does not only slow you down. It turns a capable user back into a beginner.

The answer, in one line

Four things happen before you reach it. You take the first answer, because saying that is not quite right starts to feel like waste. You stop being tested, because the tutor approach costs more messages than an answer does.

Share as a card

Definition#

The scarcity effect on AI use: the degradation in how somebody uses a model as its remaining quota falls, before the quota runs out. It shows up as accepting first answers, abandoning the back-and-forth, compressing several questions into one, and hoarding the better tool.

Share this definition as a card

The four things that happen#

You take the first answer. When messages feel scarce you stop saying "that is not quite right". You were going to. Then it felt like waste.

You stop being tested. The tutor approach, where the model questions you rather than answering you, costs more messages than an answer does. Under pressure people abandon the thing that was working, which here happens to be the thing with the evidence behind it.

You ask worse questions. Rushed, vague, all at once. Then you spend three messages fixing what one careful one would have done, which is the part that makes the scarcity real rather than imagined.

You start hoarding. Saving the good tool for something important, and doing the important thing badly with a worse one.

Scarcity taxes the thinking, and that is measured elsewhere#

The mechanism is not specific to AI, and the best evidence for it is about money.

Mani, Mullainathan, Shafir and Zhao put shoppers through cognitive tasks after experimentally inducing thoughts about finances. Performance fell among poorer participants and not among better-off ones. They then tested the same sugarcane farmers in Tamil Nadu before harvest, when poor, and after harvest, when comparatively rich: the same people performed worse when the scarcity was present, by a margin the authors compare to a night without sleep.

The finding is that scarcity itself occupies cognitive capacity, independently of who has it. A published Comment in Science disputes aspects of the analysis, which anyone citing it should say.

Applying that to a message quota is an analogy and is labelled as one here. Nothing in that paper is about AI, and a message limit is not poverty. What transfers is the shape: a resource that feels short changes how the person thinks before it runs out.

Why this costs more than it looks#

The behaviour scarcity removes first is the behaviour with the strongest evidence attached to it.

Across nearly a thousand school students given unrestricted access, a hints-only tutor, or nothing, the unrestricted group scored 17 per cent below students who never had the tool once it was withdrawn, while the tutor group kept most of its gain. The difference between those arms is the difference between asking for an answer and being made to work, and being made to work is what a quota makes feel expensive.

So the scarce user does not simply do less. They convert themselves from the arm that learned into the arm that did not, and they do it for a reason that feels like prudence.

Three things that actually work#

Think first, on paper, then open it. Write the question and your own position before you spend a message. Most wasted messages are thinking done out loud in the wrong place, and this is the cheapest available fix. It is also the first step of think, AI, think arriving for an entirely practical reason.

Spend the quota on being questioned, not on being answered. If you have twenty messages left, twenty minutes of a model interrogating you on the reading is worth more than twenty answers, because that use survives the tool being taken away.

Use the limits separately. Quotas reset independently across providers. Hitting the wall on one does not mean the day is over, and knowing that removes most of the felt scarcity, which is the part doing the damage.

What this has not been shown#

Nobody has measured this. There is no study of how usage quality changes as a rate limit approaches, no measurement of the four behaviours above, and no comparison between users on free and paid tiers. The four are drawn from teaching and from watching students, and they are offered as a description somebody may recognise rather than as a result.

The scarcity literature it leans on is about money and time in populations under genuine financial pressure. Borrowing it for a chatbot quota is a reasonable analogy and an untested one, and if it turns out the effect does not transfer, the practical advice above survives anyway, because thinking before you type is cheap under any conditions.

Key sources

On the sequence this protects, think, AI, think, and on the rungs a scarce user falls down, the four levels. For students, how to use AI at university and how to handle forty readings. On asking better, goal, context, friction, standard and what makes a good question.

About this research#

The four behaviours and the phrase describing them are used by Rahim Hirji in teaching and appear in the Mastering AI deck, most recently in September 2026. No claim of first use is made. The scarcity finding belongs to Mani, Mullainathan, Shafir and Zhao. The reading offered here, that a quota converts a capable user into a beginner by removing the behaviour with the most evidence behind it, is an interpretation by Rahim Hirji and is marked as an interpretation and not a finding.

Evidence review · SS-2026-195 · Graded against the published rubric

Cite this page

Hirji, R. (2026). Running out of messages makes you worse. The SuperSkills evidence base, SS-2026-195. https://thesuperskills.com/research/running-out-of-messages-makes-you-worse. Last reviewed 6 September 2026.

An evidence review by Rahim Hirji, not peer-reviewed research. For a material claim, cite the underlying study as well; every study here carries its own permanent link.

How citations and IDs work
Questions answered on this page

Why does hitting a message limit make you use AI badly?

Four things happen before you reach it. You take the first answer, because saying that is not quite right starts to feel like waste. You stop being tested, because the tutor approach costs more messages than an answer does. You ask worse questions, rushed and vague and all at once, then spend three messages fixing what one careful one would have done. And you hoard, saving the good tool for something important and doing the important thing badly with a worse one.

Is there evidence that scarcity affects how well people think?

Yes, though not about AI. Mani, Mullainathan, Shafir and Zhao induced thoughts about finances in shoppers before cognitive tasks and found performance fell among poorer participants and not better-off ones, then tested the same sugarcane farmers before and after harvest and found the same people performing worse when poor, by a margin they compare to a night without sleep. A published Comment in Science disputes aspects of the analysis. Applying it to a message quota is an analogy and is labelled as one.

What is the cost of using AI under a quota?

The behaviour scarcity removes first is the behaviour with the strongest evidence attached. Among nearly a thousand school students, an unrestricted group scored 17 per cent below students who never had the tool once it was withdrawn, while a hints-only tutor group kept most of its gain. Being made to work is what a quota makes feel expensive, so a scarce user converts themselves from the arm that learned into the arm that did not, for a reason that feels like prudence.

How do you use AI well on a free tier?

Three things. Write your question and your own position on paper before you spend a message, because most wasted messages are thinking done out loud in the wrong place. Spend the quota on being questioned rather than answered: twenty minutes of a model interrogating you on the reading is worth more than twenty answers. And remember that limits reset separately across providers, so hitting the wall on one does not end the day, which removes most of the felt scarcity that was doing the damage.

In this hub

For students

Ask the evidence
What does the evidence actually show?What should our board be asking about this?Where does Rahim disagree with the consensus?
Bring this into your organisation

If this describes something happening in your teams, say so.

Keynotes, board sessions and advisory work, drawing on research across more than 200 organisations in 30 countries. Tell me the room, the date and the shift you need. A reply within 24 hours.

Start a conversation

Topics and audiences  ·  All research

Box of Amazing

Rahim’s free weekly letter on AI and human capability

If this was useful, the weekly letter is where the thinking happens first. Most of what ends up on this site starts there. Weekly essays on AI, capability and the future of work. Read by 25,000 people, every week since 2017. Free, and one click to stop.

Opens Substack to confirm. No pitch in it, unsubscribe in one click, and nobody follows up because you read something.

Running an event, or responsible for how AI arrives in your organisation? Keynotes  ·  Advisory for CEOs and boards  ·  Enquire