The case for researchers publishing less in the age of AI

Academics are publishing more papers than ever, and the usual complaint is that there are simply too many. But the volume is a symptom, not the cause. The real problem is an academic reward system that leans on publication records, the number of papers and the prestige of the venues they appear in, as a proxy for a researcher’s worth.

That proxy was always imperfect, but it was at least anchored by effort: papers, and especially good ones, were costly to produce, so the count could not climb without real work behind it. Generative AI has cut that cost, at every tier.

Once much of the labour of producing a paper can be automated, as it increasingly can, a long publication list stops being evidence of anything except a willingness to keep submitting. The reward system, still built around counting, has not caught up.

This is not a complaint about AI-assisted writing, which I use and which has real benefits. It is a point about what a number means once it becomes cheap and easy to produce. For as long as research has been published, producing the work was the slow, limiting step.

That is no longer true. What stays scarce, and grows more valuable as production cheapens, is judgment: deciding what is worth pursuing, what is worth publishing, and what is worth a researcher’s name.

Here is a proposal almost no one wants to hear: each researcher should appear as an author on no more than a set number of papers a year, calibrated by field, and the rest of the system, from hiring committees to funders to journals, should be built to support that limit rather than reward people for exceeding it.

There is no shortage of proposed responses to the academic paper glut, but nearly all of them aim at evaluating research more wisely. Far rarer is any willingness to limit how much gets produced in the first place. That is the case this article makes.

How a cap would work

The proposal, in essence, is a yearly limit on the total number of papers a researcher can publish, counted across all journals and conferences and calibrated by field.

This is not the per-venue submission cap that many conferences and journals already use; it is a cap on the total an author can publish anywhere in a year. Set it low enough, and every researcher is forced to be selective about which work is worth their name.

The exact form is a detail to be worked out, and a sensible version would not be a rigid count in a single year. It might be an average over several years, with room for an unusually productive stretch, so that one strong year does not put a researcher over the line. These finer details are not the point. The point is the principle underneath them: that there should be some limit on how many papers a researcher can author, rather than no limit at all.

Why would a limit help?

Because it puts a price back on authorship. When you can publish without limit, adding your name to one more paper costs nothing, so the incentive always tilts towards adding it. Under a cap, each paper uses up one of a small number of slots, so it carries a real cost: the more careful, better-developed paper you might have spent that slot on instead.

A cap does not necessarily guarantee quality. What it does is make authorship more selective: a name on a paper becomes a signal that the author considered the work worth one of their limited slots.

What changes under a cap

Three things would improve, and none requires the cap to be perfect.

The first is the one academics feel most directly: relief from the pressure on time. Most researchers do not run the publish-or-perish race because they want to; they run it because universities, funders, rankings, and hiring norms reward output above almost everything else, and opting out alone means falling behind.

When publishing less no longer means falling behind, there is room for what the race squeezes out: teaching with more care, mentoring closely, the service roles that hold a department or a field together, a life outside work, and the chance to hold an idea back and refine it rather than rushing it out to lift the count.

The second is that some of the worst distortions of authorship lose their fuel. When a name on a paper consumes one of a finite number of slots, the practices that depend on names being free, gift authorship, sprawling hyper-authorship, reciprocal citation arrangements, stop paying off.

There is little point accepting a place on a paper you did little to shape when it uses up one of your own slots, and little point offering such a place to a colleague when it would cost them one of theirs. Today it is not unusual for a single researcher to appear on dozens of papers in a year, far more than anyone can contribute to meaningfully; a cap makes that impossible by design.

The third is that evaluation becomes fairer. Today a researcher who publishes three careful, important papers in a year is hard to compare against one who publishes 10 competent but forgettable ones in good venues, and the count quietly favours the second researcher.

Under a cap, everyone is presenting roughly the same small number of papers, the work they chose to put their name to, so they are judged on their best rather than their most.

But would it work?

A proposal like this draws objections quickly, and the serious ones deserve answers.

• Disciplines differ. A historian and a particle physicist do not, and should not, publish at the same rate, which is why a single global number would be absurd.

But that is an argument for calibrating the cap field by field, not against having one: each discipline’s limit can be set relative to its own norms, so the disparity between fields is built into the cap rather than ignored by it. Within a field, the limit should be low enough that it binds for most researchers, not just a few prolific outliers, so that everyone has to choose carefully which work is worth their name.

• It will penalise those who lead large groups. A senior researcher who supervises many students, sets directions, and secures funding contributes a great deal, and under a cap could not appear on every resulting paper. The contribution is real, but the answer is not to let it inflate the author list.

Authorship is supposed to mean substantial intellectual contribution to the specific paper, and no one can contribute at that level to more than a handful of papers a year. Supervisory and enabling roles should instead be credited honestly through other channels, detailed acknowledgments, contribution statements, or a distinct tier of recognition that names those roles for what they are, rather than disguising them as line-by-line authorship.

• A cap is too hard to enforce. It does not need universal enforcement to work. A cap can enter through any one of four independent doors.

A journal or conference accepts submissions only from authors who do not exceed their annual cap. A funding agency makes cap compliance a condition of applying for a grant. A coalition of universities, along the lines of the signatories to the San Francisco Declaration on Research Assessment (DORA), requires its own staff and students not to exceed it. And individual researchers commit to a limit and declare it publicly.

Each works on its own; none waits for the others. Funding agencies may be the strongest lever, because a researcher cannot easily ignore a condition attached to the money that sustains their lab.

But all four share one dependency: the cap only works if evaluators read a constrained record differently. If hiring committees, funding panels and tenure cases keep rewarding raw volume, a researcher who publishes less to comply simply looks worse on the metrics everyone still uses, and restraint becomes self-punishment.

The reform is therefore two moves, not one: limit the volume, and agree to read the limited record as selectivity rather than deficit.

• Why not just evaluate better? This objection deserves the most weight, because the obvious alternative is to keep all the papers and simply judge them more wisely.

We have been trying that for over a decade. DORA, signed by tens of thousands of researchers and institutions, asks evaluators to judge research on its content rather than on counts and venue prestige.

The principle is right, yet it has not slowed the race, not because anyone disagrees with it, but because careful, qualitative judgment does not scale. Reading several papers from each of a hundred applicants, and weighing their significance across subfields, takes time and expertise committees do not have. Even asking for each candidate’s best three papers leaves several hundred to read in a single hiring round.

So, evaluators fall back on what is fast: quantitative measures such as paper counts, venue prestige, citations, and the h-index. These metrics survive not because anyone thinks they are ideal, but because they are cheap to apply at scale.

This is why better evaluation, on its own, cannot fix the problem. DORA asks evaluators to exercise more judgment, but the volume of work makes that hard no matter how willing they are; a cap does not magically make careful reading scale either.

What it does is improve the fallback. Evaluators will still lean on quick metrics, but under a cap those metrics carry a different meaning. A paper count, or a record of where someone published, reflects what a researcher chose to put their name to under a real constraint, not how much they could churn out. The shortcut everyone relies on becomes far less misleading, because the volume that distorted it is gone.

A place to start

A cap is not the final answer, and I do not want it adopted unexamined. It is one concrete response, offered partly to make a larger point. The problems of publish-or-perish are well known, and there has been no shortage of attempts to address them, through better metrics, contribution statements, and new publishing models.

But almost all of those efforts try to measure the paper flood more wisely, and almost none try to reduce it.

A limit on volume is the option the conversation keeps stepping around. There may well be better instruments than a cap, but we will not find them while the profession’s reflex, faced with cheap production, is to publish ever more and lean ever harder on counting it.

I used generative AI tools for the writing, but the ideas, arguments and judgments are mine, worked out through reading, reflection and the feedback of colleagues. The division is deliberate, and it is the article’s point: production can be assisted by machines, while the judgment that matters stays human.

Muhammad Aamir Cheema is a professor and head of the department of software systems and cybersecurity in the faculty of information technology at Monash University in Australia.

This article is a commentary. Commentary articles are the opinion of the author and do not necessarily reflect the views of University World News.