ɫ��ֱ��

Here is the brutal truth about AI discoverability: it is not replacing citation metrics. It is layering itself on top of them.

So, work that is already cited, circulated and sitting inside recognisable debates is likely to become even more visible. The rich get richer, now with a chatbot assistant. If Google Scholar sometimes buries obscure work, AI tools might simply forget it was ever born.

This is grim news for anyone who believes good research should glow naturally in the dark. It does not. Visibility is not a moral reward for intellectual virtue. It is a distribution problem.

• How to get your research found on Google Scholar
• Broken citation or academic misconduct? How to spot the difference
• Campus webinar: How to get your academic work published

Researchers are now using large language models such as ChatGPT and Claude as well as Perplexity, Elicit and other tools to map fields, identify key works, summarise debates and build reading lists. These systems draw on a messy infrastructure of citation data, publisher pages, repositories, search indexes, abstracts, public summaries and online traces. Nobody outside the machine room fully knows how all this works. Anyone claiming to have “cracked LLM discoverability” is probably selling a webinar with suspiciously clean branding.

Still, the direction is clear enough. If you want your work to be found by AI, you need to make it legible, connected and repeatedly encountered. Not hyped. Not gamed. Just findable.

Connect your work to the conversation AI already recognises

AI tools are pattern-recognition machines. They are better at seeing work that sits inside an identifiable scholarly conversation than work floating beautifully alone in conceptual space. So, be explicit about your intellectual neighbourhood. Which debate are you entering? Which field should recognise the contribution? Which concepts should travel with the article? Which authors, journals and literatures form the citation environment around it?

This does not mean cynically citing famous people. It means showing where the paper belongs. If your work contributes to migration studies, political ecology, higher education, AI ethics or public health, say so. Use the names of debates that readers use. Do not invent a private label for an existing field unless you are prepared to drag that label around the internet for a decade.

Interdisciplinary work needs even more care. Academics love saying their work “speaks across disciplines”. Often this means it speaks fluently to none of them. Build bridges, not fog machines. “This article connects political geography and science and technology studies by…” is not elegant but for AI visibility, it is useful.

People are lazy. So are robots.

Write your journal article title like a retrieval device, not a private joke

The title is not where your personality goes to do jazz hands. It is prime metadata. AI tools lean heavily on the words attached to your work: title, abstract, keywords, headings, citations, venue and surrounding online text. If your title is coy, metaphorical or built around an in-joke from a conference panel in 2017, you are making discovery harder.

A good title says what the paper is about using language someone might put into a prompt. Use the core concept, case, method or finding. If the article is about platform labour, border policing, doctoral supervision, lunar governance or teacher burnout, those words need to appear where humans and machines can see them.

Bad: “Broken bridges: rethinking mobility after the storm”

Better: “Climate displacement, urban infrastructure and mobility governance after flooding in Jakarta”

Yes, the second is less poetic. It also has a pulse.

Make the abstract brutally explicit

A useful abstract tells the reader what problem the paper addresses, what literature it speaks to, what case or evidence it uses, what it finds and why anyone should care. It is the compressed operating system of the article.

Use the vocabulary of the field you want to reach, not only the vocabulary of the field that raised you. If you want scholars in climate adaptation, migration and urban planning to find the piece, the abstract cannot only speak of “critical spatialities of unsettled dwelling”. I love that stuff. The robot is already giving you a 6-7 hand sign in response.

Do not spend the first four sentences clearing your throat about complexity, urgency and the fact that “little is known”. Little is always known. That is why we are all still employed.

Fix the housing around your academic paper

A paper locked behind a paywall is not dead, but it is wearing ankle weights. Where legally possible, make a version openly available. Use your institutional repository. Use a preprint server where appropriate. Upload the accepted manuscript if journal policy allows. Add the DOI. Use your ORCID. Keep your institutional profile current. Make sure your title, abstract and author name are consistent across publisher pages, repositories, personal websites and academic databases.

Also fix the housing around the paper. AI systems do not only encounter scholarship through journal pages. They crawl the wider knowledge infrastructure around it: repositories, citation databases, publisher metadata, public summaries, Wikipedia pages, structured records and whatever retrieval system someone has duct-taped together this week.

This does not mean creating a Wikipedia page about yourself like a maniac with an h-index. It means making sure the public pages around your field, concepts, cases and debates are not nonsense, and that reliable scholarship is cited where appropriate.

Do not scatter five slightly different versions of the same work across the internet like a deranged breadcrumb trail. One clean, open version beats three broken records and a PDF called final_final_revised_USE_THIS_ONE.pdf.

You do not need to master publishing architecture but your paper does need a stable home with a recognisable address.

Circulate the argument, not just the article

This is the part academics dislike because it sounds like marketing. Tough. Circulation matters.

If people present, cite, teach, discuss, summarise and link to your work, it becomes easier for both humans and machines to recognise it as part of a live conversation. Online availability alone is not enough. Research still moves through talks, panels, newsletters, reading groups, podcasts, blogs, policy briefs and, yes, those LinkedIn posts everyone claims to hate but reads anyway.

Give the paper a portable description: one sentence, three sentences and one plain-English paragraph that use the same core terms. Build citation hooks by asking not only: “What is this about?” but: “When would someone need to cite it?” Produce three to five accurate claims that can survive outside the PDF: not slogans, not TED-talk confetti, just clean statements of what the work shows and when it matters.

Then translate the finding for adjacent audiences. Teach the work to speak the languages of the neighbouring fields most likely to use it. Ideas usually travel by stepping stones, not rainbows.

This is also where research training needs to catch up. I teach people how to produce articles but rarely how to make ideas travel after publication. That gap is now becoming expensive. In an AI-mediated discovery environment, visibility is not a bonus skill. It is part of scholarly method.

None of this guarantees that AI tools will find your work. Your next reader might not begin with a keyword search. They might begin with a prompt. Make sure your work has a fighting chance of appearing.

Darshan Vigneswaran is associate professor of political science at the University of Amsterdam. He is also founder of Spaceinterface and Research Impact Studio.

If you would like advice and insight from academics and university staff delivered direct to your inbox each week, .