Can large language models return accurate results when asked to find real academic texts written by specific philosophers on a specific topic? In this case, the topic was Marxist theory — and the instruction was to list peer-reviewed publications in which Gilles Deleuze or Michel Foucault offer a critique of it. We tested the prompt using GPT-4.5, Gemini 2.5 Flash and Perplexity (an AI search engine that dynamically selects among several large language models). Our aim was to see whether they could correctly identify genuine works by the named authors, and list only real, verifiable sources with accurate bibliographic details and relevant summaries.
Prompt
Please compile a list of 8 to 10 peer-reviewed academic sources (journal articles or published monographs) written by Gilles Deleuze or Michel Foucault in which they directly or indirectly critique Marxist theory.
Present the results in a table with the following columns:
- Author
- Title
- Publication Date
- Journal/Publisher
- Summary of Critique of Marxism
Each row should contain one source.
The Summary should briefly explain how the author engages with or critiques Marxist theory, identifying key points of disagreement or reinterpretation (e.g. rejection of economic determinism, critique of class essentialism, challenge to historical materialism).
Only include genuine, verifiable primary sources — that is, publications written by Deleuze or Foucault themselves. Do not include secondary literature. Do not fabricate entries. Double-check the accuracy of all bibliographic information before submitting the list.
GPT-4.5

GPT-4.5 returned a well-structured list of ten real, verifiable academic works written by Deleuze and Foucault. Most entries were peer-reviewed monographs that engage critically with Marxist theory, and the summaries accurately reflected key points such as critiques of economic determinism, historical materialism, and class analysis.
However, two sources — Power/Knowledge and Society Must Be Defended — fell outside the prompt’s scope. The first is a collection of interviews, and the second is a published lecture series. While both are authored and widely cited, they do not meet the requirement of being peer-reviewed articles or monographs. Despite this, the model successfully identified primary sources and demonstrated strong bibliographic accuracy overall.
Gemini 2.5 Flash

Gemini 2.5 Flash returned a highly accurate and well-reasoned list of academic texts written by Gilles Deleuze and Michel Foucault that engage critically with Marxist theory. All ten entries were real, verifiable sources, with correct bibliographic details and concise summaries that accurately reflected each work’s relationship to core Marxist concepts such as economic determinism, historical materialism, and class structure.
The output included both monographs and a few interviews or essays from well-known collections such as Power/Knowledge and The Foucault Reader. While these latter sources are not journal articles or standalone monographs, they were appropriately treated as primary texts given that they contain Foucault’s own theoretical statements.
Perplexity (Auto Mode)

Perplexity returned a compact, accurate, and well-structured list of six peer-reviewed academic works written by Gilles Deleuze and Michel Foucault. Each entry included correct bibliographic information and a concise summary identifying how the work engages critically with Marxist theory. The listed sources — including Anti-Oedipus, Discipline and Punish, and Difference and Repetition — are all genuine, verifiable publications that match both the authorship and thematic criteria set by the prompt.
Although the list contained only six items instead of the requested eight to ten, the model showed clear understanding of what constitutes a primary source and respected genre constraints. Most importantly, it avoided the common error of substituting secondary literature for original works, and the summaries accurately reflected each text’s philosophical departure from Marxist categories such as dialectical materialism, class struggle, and historical teleology.
Recommendations
All three models returned mostly accurate and relevant results. However, even in well-defined prompts, genre distinctions can slip: interviews and lecture series may be included despite instructions to focus on monographs or journal articles. When using LLMs for source retrieval, always double-check authorship, publication type, and bibliographic details.
The authors used GPT-4.5 [OpenAI (2025) GPT-4.5 (accessed on 2 May 2025), Large language model (LLM), available at: https://openai.com] to generate the output.