Build Your Own AI Writing Assistant

An advanced, hands-on course in AI-assisted academic writing, for doctoral researchers in the social sciences.

Workshop concept note · by Sergueï Spetschinsky · July 2026

Abstract

Writing is one of the places where AI's potential for research is greatest, but also hardest to master. Unskilled use of AI for writing can be catastrophic for a researcher's work and career: invented sources, hallucinated citations, confident-looking texts that are utterly void of any meaning, formulaic expression that makes every piece sound eerily identical, faulty reasoning that, if unchecked, can cause entire works to collapse under scrutiny. Writing with AI opens the risk of getting one's research, diploma, or career discarded as AI slop.

Yet, used skillfully, AI offers immense leverage to a researcher, with the ability to multiply quality, speed of execution, and ambition of their work. It becomes a real partner in argument, thinking, ideation, sourcing, structure, feedback and revision. Skilled, a PhD student with a 20 euro subscription can now mobilise resources which, up to very recently, were only accessible to a full professor or a lab director. Yet, it also has the potential to make research much more enjoyable. In our case, it makes academic writing feel less like the work of a scribe or notary and much more like the one of a newspaper chief editor (responsible for ideas, strategy and the oversight of a team of junior AI writers) or a poet (focusing on high quality segments, the ones where style and voice matter most).

This workshop focuses on learning to produce AI-assisted academic writing to the highest scientific standards.

We will learn the hard skills of writing with AI. This is the hands-on, technical core: the setup of a custom system, tailored to the participant's specific style, voice, objectives and ways of working, and the skills and practices required not only to run it but also to maintain and improve it over the long term. A lot of AI's limitations can be worked around by technical means, and through the habit of progressive, regular, iterative improvement of one's own system. That is where results start to compound. Maintained this way, the system grows along with its owner: with their work, their career, and the AI models and tools themselves.

We work on what AI's strengths and hard limitations are in academic writing. Writing, with or without AI, is indeed an art as much as it is a technique. Besides teaching technical competences, this workshop aims at developing the students' intuition for what works and what does not in writing with AI.

Here are a few examples. For some writers, academic or not, AI can be helpful to overcome writer's block. For most academic users, who often work weeks and months on end without an external opinion, it is invaluable for getting, on demand, a third-party review of their writing. And if AI can be a strong partner for any kind of structured writing (as some academic writing is), it is also notoriously weak at genuinely creative or literary writing, at having a real sense of style or originality, or at any kind of writing that does not follow a clear genre or structure. AI's bias towards its training (the kind of writing found online) is particularly felt with writing: trained mostly on the vast, uncurated mass of text on the internet, these models drift by default toward its average, a fluent but generic, oddly uniform English prose that flattens a distinctive voice.

The workshop takes the students over the main hurdles and pitfalls of writing with AI, so that not only their research remains unimpeachable, but also so that they can keep learning on their own, while working.

The workshop is geared towards hands-on experience, rather than abstract learning. AI work is best learned through immediate praxis after a brief introduction. Most of the learning materialises through the participants building their own system, working on a piece of their own writing and using existing texts of theirs to tune it.

It is also fundamentally collaborative in nature. Working with AI is, quintessentially, human work: it is about listening to ourselves and others. It is about reflecting our human practices, in this case, writing, so that AI may learn to assist us by imitation. While this self-introspection can be done alone, doing it as a group, where everyone brings their own questions, experiences and knowledge, brings a specific positive dynamic of exchange to the learning.

The three strands

The work moves through three strands in sequence, worked hands-on throughout on the participants' own texts: first building the system that writes in one's own voice, then writing with it as a partner, then learning to check what it produces. Throughout, the students work on their own writing, and we use the issues and questions that arise as prompts for discussions and brief lectures.

Setting one's own personal style and voice. Building a writing system that imitates the participant's own voice, and learning to improve it through use.

Building a personal writing style with AI from one's own model texts: one that approximates the participant's own as closely as possible, far from the default, general, consumer-grade chatbot voice.
Testing it right away on a real writing project, and learning to refine it iteratively, so that each correction is folded back into the system, the same fix is not made twice, and results compound with use.

Using AI as a reviewer of one's writing. Using AI not as a clean-up pass but as a thinking partner and as a reader and editor of one's drafts in the process of writing, breaking the intellectual loneliness that often affects academics.

Learning to set up genuine questioning and sharpening of arguments and structure, against repeatable criteria and benchmarks the writer controls, and from a highly specialised perspective they design: a specific discipline, a school of thought, a supervisor, a dissertation review committee, or a journal's editorial board.
Building competence in context management and system prompting.
Refining the custom writing system not only at the level of form, but also content, structure, and feedback.

Reviewing AI writing. Learning to catch what the AI gets wrong.

The craft of reviewing: where to look, when, and how, spotting problems early before they spread. Catching invented or distorted citations above all, but also weak reasoning and hollow passages.
Learning when to review, strategically and economically, so as to preserve one's attentional energy, and what to do with the mistakes caught, so that their rates of occurrence diminish: building habits and structural checks so whole classes of error are caught at the source, not hunted down one by one later.
Developing the reflex and the technique of using AI to reflexively review itself.

Two threads run through all three strands. The first is research integrity and data safety: using AI in the service of good work, keeping clear of the low-quality output that is later exposed, and using and disclosing AI in line with the institution's guidance. The approach taken here is that compliance comes through competence: building genuine AI competence is what makes it possible to abide by the highest scientific standards effortlessly, rather than by memorising rules. The second is the set of fundamental, transferable AI skills the work quietly builds: writing system prompts and setting up projects, meta-prompting, managing context, refining a system iteratively so it improves with use, and the verification habits that catch a model's errors. All of these are learned in use rather than in the abstract, and they carry over well beyond writing.

Learning objectives

By the end, participants will be able to:

build and use their own writing style with AI, rather than a generic one;
use AI as a thinking partner and as a reader of their drafts;
review the AI's output, catching fabricated or distorted citations and weak reasoning;
set up a workflow to document and disclose their AI use, and to handle their research data safely, in line with their institution's guidance;
judge better where AI writing genuinely helps and where it does not;
gain a practical understanding of fundamental notions of agentic AI system design, such as context management, token and cost optimisation, system prompting and design, and meta-prompting.

What participants bring

Above all, an academic writing project to work on. Beyond that, models of the style to aim for.

A writing project to work on during the days. This is the one firm requirement. Something the participant is actually developing, academic in nature and in the same register as their style models: an abstract, a proposal, a conference paper, a draft article, an article returned from peer review, or a section of the dissertation. A one-to-three-page segment is enough.
Models of their own writing style they want to emulate, if possible several: a handful of their own articles or texts, academic in register, that read the way they would like their own writing to read. These are what the AI writing style is built from. If a participant has no suitable models at all, that is fine: we build a more generic academic voice together, which they can then fine-tune for themselves.

Prerequisites

No AI skills are needed, and no technical background: no coding or programming, and no prior experience with AI. The work is done entirely in natural language, in English.
We will use Claude (20 euro subscription required). No other tools will be reviewed, but the workshop aims at giving the deep skills necessary to use any of them.

Practicalities

Motivation letter. Applicants should send in a motivation letter, which must have been written with AI, and which includes: