Your research isn't
a training set.

Overleaf brings the power of AI to your writing - built around a single promise. Your work stays yours. Always.

Your data.
Never theirs.
Zero training on your research · 25M+ researchers trust Overleaf
∂L/∂q = μ M = Σ 2ⁿ ρ∂ū/∂t = −∇p+μ∇²ū
  • Your data never trains our models - unless you opt in
  • Your data is never sold or shared with third parties
  • You choose how and when AI plays a role in your workflow
  • Clear warnings before any third-party AI model sees your work

Your data, your control

AI that works for you,
not the other way round.

Your research won't be used to train our models unless you opt in. You have full control of how and when AI plays a role in your workflow - with no surprises in the small print.

Opt in, always

No AI feature activates without your consent. You decide what's helpful for your workflow.

Zero data sales

Informed at every step

How Overleaf compares

Not all AI writing tools
make the same promises

When your research is unpublished, the stakes are high. Generic AI writing tools weren't built with that in mind. Overleaf was.

Built for LaTeX and scientific writing

Every AI feature is designed for the way researchers actually write - equations, citations, and all.

Research privacy by default

Trusted by institutions

∂L/∂q = μ ρ∂ū/∂t = −∇p
Overleaf
No AI training on your work
Data never sold
Built for LaTeX
Real-time collaboration
Journal submission
Institutional controls
Other tools
Not guaranteed
Often unclear
Generic AI
Not available
Not available
Not available

Your rights as a researcher

The Overleaf AI
Bill of Rights

These aren't aspirations. They're commitments - written into how we build every AI feature.

01.Control

You are in control

Full control over how and when you engage with AI on any Overleaf product - opt in, opt out, your call.

Always opt-in
02.Privacy

Your data is private

Your research will never train our models without your explicit consent. We never sell or share your data for commercial purposes.

Zero data sales
03.Transparency

You are always informed

Before any feature passes your work to a third-party AI model, you'll receive a plain-language warning and the choice to proceed.

Full transparency
04.Institution

Institutional control

Administrators can block access to any AI technologies for their subscription.

For universities
05.Ethics

Ethical AI development

We build to avoid bias and discrimination. We have a clear process for users to raise issues - and we act on them with accountability.

Bias-aware
06.Sustainability

Sustainable by design

Our custom models are more efficient and environmentally sustainable than large-scale models.

Lower footprint
∂L/∂q = μ ρ∂ū/∂t = −∇p w² = f²+c²
01
AI exists to advance research, not replace it
Every feature we ship must tie back to making research faster, more open, more transparent.
02
Ethical AI is ongoing work, not a checkbox
We have a process for users to raise issues with our AI - and we act on them.
03
Gains must be balanced with sustainability
Our purpose-built models are significantly more energy-efficient than large general-purpose AI.
04
Transparency is a living commitment
As AI evolves, our policies evolve - always with clear notice and full transparency.

How we build

AI development
principles

These are the beliefs that guide every AI feature we ship - from the first line of code through to the product you use every day.

Research-first, always

Our mission is to make science faster, more open, and more transparent. Every AI decision is held against that standard.

Accountability built in

This page will evolve

M+
researchers worldwide
documents sold to third parties
%
Opt-in for data training
∂L/∂q = μ T = a+b log₂ 2D/N w² = f²+c²(k²+l²) ρ∂ū/∂t = −∇p+μ∇²ū ∫ fv dΩ = ∮ H₃N⁺ − C
\begin{now}

Write with AI.
Keep your research.

Trusted by 25 million researchers worldwide to protect their work while accelerating it.