Adapted from Jefferson Pooley's essay "The Matthew Effect in AI Summary," LSE Impact Blog, May 2026. Expanded and humanised for general readership by Pawan Maheshwari.
Imagine you're a junior researcher in Lagos, Nairobi, or Lima — brilliant, rigorous, publishing in peer-reviewed journals. Then a colleague in Boston opens an AI research tool, types your exact topic, and the tool returns a neat summary built almost entirely from scholars at Oxford, Harvard, and MIT. Your work doesn't appear. It might as well not exist.
This is not a hypothetical. It is, increasingly, the lived reality of what happens when we hand the task of synthesising knowledge over to AI — and it has a name borrowed from the Bible.
What Is the Matthew Effect?
In 1968, Columbia sociologist Robert Merton coined the Matthew Effect, borrowing from the Gospel of Matthew: "For to every one who has will more be given, and he will have abundance; but from him who has not, even what he has will be taken away."
In plain terms: famous scholars get more famous. Their papers get cited more, which makes them appear more authoritative, which leads to still more citations. The less-cited scholar drifts toward obscurity — not necessarily because their work is worse, but because it never entered the visibility loop in the first place.
Visibility doesn't just reflect quality. It creates it. The more a paper is cited, the more credible it appears — and the more it will be cited again.
The Matilda Effect — A Bias With a Name
In 1993, historian of science Margaret Rossiter named a parallel phenomenon the Matilda Effect — after Matilda Gage, a nineteenth-century suffragist who documented how women's intellectual contributions were systematically attributed to their male colleagues. The Matthew Effect doesn't just disadvantage isolated individuals. It amplifies patterns already inscribed in the system — shaped by colonialism, patriarchy, and the dominance of English-language publishing in the Global North.
Enter the AI Research Tool
Over the past eighteen months, Google, OpenAI, Anthropic, Microsoft, and Perplexity all launched "deep research" tools with near-identical promises: ask a question, receive a researched summary with citations. Academic publishers followed — Elsevier, Clarivate, and Digital Science all released AI-powered literature assistants.
The tools are impressive. But underneath the polished output lies a structural problem: when we ask an AI to summarise "the literature," the tool doesn't read papers the way a human does — it operates by predicting, statistically, which content is most likely relevant based on patterns learned from a vast training corpus. That corpus is the accumulated record of published scholarship — including all of its biases.
"The models may act as laundering machines — context-erasing abstractions that disguise their probabilistic reasoning and give old inequalities a fresh coat of algorithmic legitimacy." — Jefferson Pooley, LSE Impact Blog
The Privilege Multiplier
The old Matthew Effect was traceable. You could follow citation trails and document patterns of exclusion. Scholars could investigate and push back. The new AI-mediated version is different: it operates inside a black box. Early empirical evidence confirms this worry — AI research tools systematically surface papers from high-prestige institutions and English-language publications at rates disproportionate even to those papers' citation counts. The tools are acting as "privilege multipliers."
Consider a researcher in Brazil studying urban poverty. The global academic literature is stratified: US and UK journals receive more citations, are included in more databases, and carry more prestige metrics. An AI tool trained on this literature will weight those sources more heavily. The Brazilian researcher who has documented ground-level realities in a Portuguese-language regional journal may simply be invisible to the tool — not because their work is less rigorous, but because the system never valued it equally.
What This Means — and What Comes Next
For individual researchers: treat AI summaries as a starting point, not a verdict. Seek literature databases that prioritise inclusivity over citation prestige. Make a habit of searching for voices from outside the dominant publishing centres.
For institutions and funders: the AI research tools being deployed at scale are not neutral. Demand transparency from AI tool providers, invest in diverse citation databases, and recognise that a tool being widely used does not make it equitable.
For AI developers: invest in bias audits, diversify training data, and create mechanisms for users to surface under-represented perspectives. These are not optional extras — they are the price of deploying these tools responsibly in high-stakes epistemic environments.
"The researcher in Lagos who didn't make the summary didn't fail to deserve inclusion. The algorithm simply never learned to see her."
AI & ScholarshipKnowledge EquityBiasMatthew EffectAcademic AI
