This Paper Shows That ___

When a good becomes less scarce, the value is reallocated to something that remains scarce. We can see this in the “computer income” share of GDP over time (h/t Kevin Frazier via Alex Imas:

Alex Imas@alexolegimas

This is just the economics of scarcity. Here is agriculture—same graph. Once something becomes plentiful (eg through automation), value reallocated to something that is scarce. We don’t eat less than before, if anything we eat way more. We don’t use computers less, we use them

Kevin Frazier @KevinTFrazier

.@ChadJonesEcon shares a major narrative violation. Check out the “computer income” share of GDP over time.

1:24 PM · May 25, 2026 · 80K Views

16 Replies · 80 Reposts · 586 Likes

Tools like Claude Code mean that coding is “mostly solved”, and recent events in mathematics show that AI tools can solve highly complex problems. So what is left for us poor humans trying to do research? What should we focus on to stand out in today’s new world? What is going to be scarce?

For research, a hopeful take you sometimes hear is that ideas will become more prominent. While the idea is clearly important, I think we spend too little time on the message: what should the reader take away and still remember a year later? A sharp message is about to get more valuable, as AI makes technical skills less scarce.

Try an experiment. Take the paper you’re working on right now and finish this sentence out loud, in front of someone: “This paper shows that ___.” If you can’t do it cleanly, it usually means that you haven’t figured out what you want your paper to actually say.

That sentence is the message that you want the reader to still remember a year after they’ve forgotten everything else about your paper. I think we are doing a disservice, especially to young researchers, if we focus too much on coming up with “the idea”. We’re trained to obsess over the idea and to treat the message as a finishing task, something you tidy up in the last revision. I think that’s missing at least half the story.

We spend a lot of time in academia talking about ideas. PhD students in particular are often hunting for ideas, meaning a question nobody has answered, a clever source of variation, or a dataset nobody else has touched. We talk endlessly about ideas and almost never about messages. But often the difference between an okay paper and a great paper is precisely that the great paper has a very clear message that the audience remembers.

We can maybe quibble about what is the message and what is the idea. But let me try to separate the two. The idea is what you’re studying and why it’s identified. The idea is to study a reform or a specific dataset. The idea is to model a particular behavior. The idea is to experimentally vary something to identify the effect on something else. The idea includes the question, the design, the source of variation. This gets fixed early, the moment you commit to a question and a research design.

The message is what you want the reader to remember when they put the paper down. It’s the sentence they’ll repeat to a colleague in the hallway a year after they’ve forgotten your entire paper. And unlike the idea, the message isn’t fixed at all. It’s a choice you keep making for the entire life of the project. It’s a thing that you iterate over and over again when you edit your introduction. The message is why you edit your paper, although it is not always expressed quite in that way. A sharp message also tells you what to cut. Once you know the one thing the reader should remember, you suddenly see that 60% of what you could say belongs in an appendix, or in a different paper entirely.

This matters because two papers can use the same data and a similar design, but can end up saying very different things depending on which margin the author chooses to put at the center. We recently published a paper in the Review of Finance, which got rejected at several different journals before. What mostly made the difference between rejection and acceptance in our case was that we updated the first paragraph, not that we added new results. Our message became clearer, and suddenly the same results were much more interesting.

And that is the thing about the papers you can actually recall: it’s almost always the message that you are remembering. One of my favorite recent papers is called “What We Teach About Race and Gender: Representation in Images and Text of Children’s Books”. The first sentence of the abstract is “Books shape how children learn about society and norms, in part through representation of different characters.” Having read a lot of kids books lately, I buy that message. The rest of the paper just follows from there. It’s a beautiful paper because the message is so clear.

Another example from finance that comes to mind is Grossman-Stiglitz, where the message is expressed in the first paragraph:

“If competitive equilibrium is defined as a situation in which prices are such that all arbitrage profits are eliminated, is it possible that a competitive economy always be in equilibrium? Clearly not, for then those who arbitrage make no (private) return from their (privately) costly activity. Hence the assumptions that all markets, including that for information, are always in equilibrium and always perfectly arbitraged are inconsistent when arbitrage is costly.”

Put into different words, the paper says that markets cannot be informationally efficient, because then no one would pay to buy the information. I remember the message of this paper years after having been forced to digest it in my Masters. I don’t really remember anything else about the paper.

We like to pretend the idea is separate from writing the message of the paper, and that you can have a great idea with clumsy messaging, and then fix that with some cosmetic changes in the last revision. I don’t think that’s true, and I think this is why so many good projects quietly underperform. I also think this is the highest value-added thing you can think about when sitting down and writing your paper.

Start with the floor. Below some quality bar, no message saves you. If the identification is weak or the question is small, a sharper message won’t get the paper into a top journal. Economics referees are trained to attack the design, so a slick message sitting on a shaky idea tends to invite more skepticism, not less. C’est la vie.

But above the floor, delivery does most of the work. The reason is that you cannot actually write the message cleanly unless the idea underneath it is good. Try it. The moment you force yourself to finish “this paper shows that ___” in one sentence, you find out whether you have a result or a pile of regressions. Working backwards from the message is a great way to discover that an extension is a distraction or that the identification doesn’t quite support the claim you wanted to make.

The trap, especially for the empirically careful people who tend to make good economists, is to treat the message as the last task on the list. Run the regressions, polish the tables, then write a nice intro and submit. By then it’s too late: the regressions you ran were shaped by a message you hadn’t articulated yet. Good papers are the ones where the author knew the message early enough that it decided which regressions got run and which got dropped, and could articulate the point of the paper early.

This is well-known from many different fields, which is worth saying out loud. Journalists, rhetoricians, marketers, philosophers, lawyers, and historians have always known that the message is key for actually getting your point across. Think about some famous economists: Kahneman, Becker, Fama, Lucas, Akerlof, Coase. For the economist reader, each of these names is distinctly associated with a key message or concept. They were extraordinarily good at compressing a whole worldview into a portable concept that you could take with you.

I have seen points raised on social media that AI raises the relative value of ideas. The argument is clean enough: much of the skill in research is being made less scarce, as AI tools can help with coding, data cleaning, robustness checks, even first drafts. All the executional labor of a paper is collapsing in cost, the argument goes, so the binding constraint moves to other aspects of research: the idea.

There’s something to that. If three weeks’ worth of work now fits into a workday, then yes, a researcher’s comparative advantage moves upstream, toward judgment and taste and question selection. That part is not wrong, but I think focusing on the quality of the idea misses at least half the story. The papers that get across in this crazy new world will be the ones with the cleanest message.

This feels especially important these days. Readers are busy, referees are tired, and editors are suddenly faced with hundreds of submissions. If you are concerned that researchers with AI will mass-produce hundreds of papers, you have to make darn sure that you are able to communicate your message clearly. You have maybe 30 seconds to convince someone the paper is worth their attention, and if you spend it clearing your throat, you’ve lost them. The people who communicate well front-load the punchline and repeat the core message until it sticks. Of course it then really helps if you are working continuously on the message of the paper (maybe together with your favorite AI tool?).

So you should probably not sit down with an AI to brainstorm ideas now, and you should not try to furiously come up with new ideas. You should be trying your best to sharpen the message and your communication in your paper. After clearing the bar on the idea, the difference-maker that emerges on the other side is the actual message of your paper. So if you’re a student casting around for an idea, try writing the message first. What do you believe in, and can you find a way to show it? What are you trying to say?

Image courtesy of ChatGPT, who in this case did a much better job than Claude. Also thank you to Olga Balakina for conversations and feedback, and for listening to me talk.