The risk of optimizing research for evaluation

I have spent much of my career working precisely on these questions: helping research actors think about impact, co-creation, and responsible innovation in more serious and structured ways. Over the last years, I also worked on the development of AI-supported tools for Horizon Europe proposal preparation. At first, this seemed relatively straightforward. European proposals are highly structured documents, researchers are overwhelmed by administrative complexity, and AI can genuinely help reduce friction, consortia structure their impact pathways more clearly, and save researchers time.

But working on these tools progressively revealed an issue that long predates AI. Because the more research funding systems formalize what “good impact” should look like, the more actors adapt their language, behaviors, and proposal strategies to evaluation frameworks themselves. At some point, one starts wondering whether projects are still being designed for societal relevance, or increasingly for evaluative readability.

When impact becomes a compliance exercise

Most Horizon Europe proposals now require highly sophisticated impact sections: pathways to impact, KPIs, stakeholder engagement plans, dissemination strategies, societal outcomes, environmental benefits and increasingly detailed exploitation plans.

At the same time, calls are extremely competitive, timelines are short, and consortia are often assembled late in the process. Under these conditions, impact quickly becomes something teams must produce rather than something they genuinely have time to collectively design.

The result is a paradox. Systems designed to encourage socially meaningful research can end up generating increasingly standardized narratives about impact instead.

Certain formulations become recurrent. Certain stakeholder categories appear everywhere. Certain concepts become unavoidable. Researchers learn the codes of evaluation. Consultants, templates, and AI tools progressively reinforce these patterns further.

Little by little, projects risk converging toward what is evaluatively legible rather than what is necessarily scientifically or socially transformative. This creates a strange situation where everybody spends enormous amounts of energy producing the appearance of an impactful project while often lacking the conditions for deep collective reflection in the first place.

And everyone knows it.

Researchers feel overwhelmed by administrative layers. Project managers struggle to maintain coherence across large consortia. Evaluators read increasingly homogeneous proposals. Civil society partners are sometimes integrated too late for genuine co-creation. And societal impact itself risks becoming reduced to a formal section disconnected from scientific practice. In other words, ticking the “impact” box does not produce impact.

Research systems shape behavior

Researchers learn what kinds of projects, narratives, partnerships, and impact claims are considered credible or fundable. Institutions organize themselves around success rates and evaluation expectations. Support structures, consultants, templates, and now AI tools progressively stabilize certain ways of presenting research as legitimate.

None of this necessarily happens consciously or cynically. In fact, many researchers genuinely care about societal impact. But over time, repeated exposure to the same evaluation logics tends to normalize particular forms of discourse and project construction.

This is how entire scientific cultures progressively become influenced by metrics and funding structures. The issue is therefore not only administrative overload. It is the risk that research itself becomes increasingly formatted around what can be efficiently evaluated.

The irony is that many actors involved in European funding are fully aware of this tension. Over the past years, I have worked extensively with researchers, project managers, startup founders, innovation support organizations, and public institutions around these questions. Most people do not reject the idea of societal impact at all. On the contrary, many are deeply motivated by it.

What they reject is the feeling that impact is disconnected from actual research practices and reduced to a performative exercise. This matters because meaningful societal impact often requires exactly the things current funding structures struggle to accommodate: time, uncertainty, interdisciplinary dialogue, trust-building, conflict, experimentation, and long-term relationships with external stakeholders.

Real co-creation rarely fits neatly into accelerated proposal timelines.

AI will not solve this problem

The recent rise of generative AI tools in proposal writing makes these questions even more interesting. AI can genuinely help structure information, reduce administrative burden, support consistency across documents, and assist overwhelmed teams. But it also raises an important question: what happens when research proposals themselves become increasingly standardized through optimization tools?

If every consortium uses similar AI systems trained on previously successful proposals, there is a real risk of homogenization. Scientific originality becomes harder to distinguish. Certain ways of framing impact become dominant simply because they are statistically recognizable as successful.

This is why I have become convinced that the question is not whether AI should or should not be used in research funding. The real question is what role we want these tools to play. Should they help researchers think more clearly? (This is the design choice we made at SoScience.) Or should they progressively replace the difficult intellectual and political work of collectively defining what matters?

There is a major difference between tools that support reflection and tools that produce formatted discourse on behalf of users. The first can strengthen impactful research practices. The second risks accelerating the bureaucratization already underway.

Beyond compliance

The broader issue ultimately goes beyond AI, proposal writing, or European funding mechanisms. It concerns the growing tendency of innovation systems to prioritize compliance, sometimes at the expense of excellence. Of course, evaluation frameworks are necessary. Public funding requires accountability, coordination, and some degree of comparability between projects. Large-scale research programs cannot operate without procedures, indicators, and forms of assessment.

The problem begins when these mechanisms stop functioning as supports for meaningful research and start becoming ends in themselves. When societal impact is treated primarily as a section to complete under severe time constraints, it risks turning into a performative exercise disconnected from the actual organization of scientific work.

In that situation, everyone becomes frustrated. Researchers experience impact requirements as bureaucratic pressure rather than intellectual engagement. Evaluators read increasingly standardized narratives. Civil society participation becomes compressed into symbolic consultation exercises. And the very concept of impact slowly loses substance because too many actors learn to reproduce its vocabulary without being given the conditions to genuinely operationalize it.

The challenge is therefore not simply to ask researchers to “integrate impact.” It is to create systems where meaningful societal engagement becomes structurally possible, intellectually legitimate, and institutionally supported. Otherwise, impact risks becoming what many researchers already quietly perceive it to be: a mandatory narrative layer added onto projects whose trajectories were largely defined elsewhere.

[5] Pauline Gandré, « Les sciences : un nouveau champ d'investigation pour les gender studies », Idées économiques et sociales, 2012/1 (N° 167), p. 52-58.