Feeding LaTeX sources (.tex) to an LLM

MakeItMarkdown · July 2026

LaTeX sources are the native format of papers, theses and course notes — and the best possible input when you have it, because the structure a PDF destroys is still explicit in the .tex. Converting the source beats OCR-ing its output every time.

The element mapping

In the .tex	In the Markdown
`\section / \subsection`	`## / ###` outline
`\title, \author, abstract`	Document title, authors, an Abstract section
`equation / align` environments	`$$ … $$` fences, math body verbatim
`tabular`	GFM pipe tables
`\includegraphics + \caption`	`[Figure: name.png]` placeholder with caption
`\cite / \ref`	`[key]` / `(ref: label)` — readable citation anchors
`thebibliography`	A References section, keyed
Preamble (packages, macros)	Dropped, with a macro-count warning

Before → after

In the file

\section{Method}\label{sec:method}
Throughput $T$ follows
\begin{equation}
T = \frac{U}{\Delta t} \cdot \eta
\end{equation}
as shown in~\cite{lee2024}.

In the Markdown

## Method

Throughput $T$ follows

$$
T = \frac{U}{\Delta t} \cdot \eta
$$

as shown in [lee2024].

Honest limits

Not a TeX engine. Custom macros (\newcommand) are not expanded — text using them converts literally, and the fidelity report counts the definitions it saw.
Math is preserved, not rendered — bodies stay verbatim inside $$ fences, which is exactly what LLMs read best anyway.
Figure images are referenced by path, not extracted — the .tex doesn't contain them.
Multi-file projects (\input/\include) convert one file at a time — batch-drop the parts.

FAQ

Why not just paste the PDF? Because the PDF is the painted output; the .tex is the structure itself.

BibTeX .bib files? Not yet — thebibliography environments convert; request .bib with a sample.

Drop a .tex and check the outline, fenced math and keyed references.