Skip to content
ISC ParisAI Use Charter

Tool, bilingual builder, computed locally

AI use statement,ready to paste

You fill it in, the block builds itself as you type, you copy it and paste it at the end of your document. Three templates, taken as they stand from Annex A3 of the charter: short for students, standard for staff, academic for publications.

The builder

Pick the template that matches your work. Fields left empty appear in brackets inside the block: they remind you what is missing instead of quietly vanishing.

Your answers

For example: Claude 3.5, ChatGPT-4, Mistral Large
For example: outline, rewording of the introduction, translation of sources
For example: structuring, rewording, translation, generation of examples
For example: sources checked on Google Scholar, figures confirmed via INSEE, quotations traced
Authorisation level applied
The one printed on the syllabus or on the assessment brief.

Not sure which level applies to your work? Find the level in three questions

The button fills in today’s date.
Your name. The handwritten signature goes on after printing.

The block to paste

7 fields left to fill

──────────────────────────────────────────────
AI USE STATEMENT
──────────────────────────────────────────────
Tool(s) used : [to complete: Claude 3.5, ChatGPT-4, Mistral Large]
Parts of the work concerned : [to complete: outline, rewording of the introduction, translation of sources]
Nature of the assistance : [to complete: structuring, rewording, translation, generation of examples]
Checks carried out : [to complete: sources checked on Google Scholar, figures confirmed via INSEE, quotations traced]
Authorisation level applied : [to complete: level stated on the brief]
Date : [to complete: dd/mm/yyyy]   Signature : [to complete: first name and surname]
──────────────────────────────────────────────

Everything is built in your browser. Nothing you type is transmitted, nothing is stored, nothing is sent to an AI model. Close the tab and nothing remains.

What each level expects from you

The level appears on the assessment brief. It does not change the four mandatory contents, it changes how much detail is expected. Where the brief says nothing, the default level applies: N2 for submitted work.
Permitted AI intensity, from the most closed to the most open.
LevelWhat the level allowsStatement expected
N0No AINo AI assistance whatsoever. Controlled environment.No statement expected, since no use is permitted.
N1PreparationAI for brainstorming and research. Final output produced without AI.Statement expected on the preparation phase.
N2AssistanceAI for specific tasks, rewording, structure. Statement mandatory.Mandatory statement, listing the parts concerned.
N3CollaborationAI embedded in the process. Critical thinking and validation required.Mandatory and detailed statement, including checks performed.
N4ExplorationCreative use of AI. Command of the tool is itself assessed.Mandatory statement, dead ends and failures included.

Common core V3, authorisation levels. Scale aligned with the AI Assessment Scale (Perkins, Furze, Roe and MacVaugh, revised version, 2024).

The bibliography line

It does not replace the statement, it comes on top of it once the use goes beyond minor assistance: translation or language correction stay below that threshold.

Recommended format

[Tool name] (version), [mode of use], “[short description of the task]”, accessed [date of access].

Mode of use

Your line

[to complete: tool name and provider], [to complete: one-off, iterative or automated] use, “[to complete: short description of the task]”, accessed [to complete: date of access].

The line picks up what you typed in the builder. Annex A3 gives the version and the provider depending on the case: keep whichever identifies the tool without ambiguity, and adjust the punctuation to the style your journal requires.

Three examples from the charter

  • Claude 3.5 Sonnet (Anthropic), iterative use, “Help structuring the outline and rewording the introduction of chapter 2”, accessed 15 May 2026.
  • ChatGPT-4 (OpenAI), one-off use, “Generation of illustrative examples for the methodology section”, accessed 23 May 2026.
  • Mistral Large 2 (Mistral AI), iterative use, “Translation and harmonisation of quotations from English into French”, accessed 2 June 2026.

Annex A3, citing AI in a bibliography.

What holds, what does not

An alibi statement costs more than no statement at all: it displays a transparency it does not deliver. Here are three cases, and the exact reason the first one is rejected.
  1. Alibi statement

    This document was written with the help of artificial intelligence.

    Statement that holds

    Tool(s) used: Claude 3.5 Sonnet (Anthropic) Parts of the work concerned: detailed outline, introduction Nature of the assistance: structuring, rewording Checks carried out: sources traced and read, figures confirmed on the INSEE website

    Why the first one is rejected

    None of the four mandatory contents is there: not the tool, not the parts concerned, not the nature of the assistance, not the checks. The charter puts it plainly: “I used AI” is not enough.

  2. Alibi statement

    Tool: a generative AI. Parts concerned: various. Checks: yes.

    Statement that holds

    Tool(s) used: Mistral Large 2 (Mistral AI) Parts of the work concerned: literature review, translation of English quotations Nature of the assistance: translation, harmonisation of terminology Checks carried out: every quotation traced back to the original text

    Why the first one is rejected

    The tool is not named and its version is unknown, “various” designates no part of the work, and a check that amounts to “yes” describes nothing. The charter asks for the name and version, the precise parts, and what was actually verified.

  3. Alibi statement

    Checks carried out: I proofread it.

    Statement that holds

    Checks carried out: bibliographic references traced one by one on Google Scholar, market figures confirmed on the INSEE website, quotations found in the original text before reuse

    Why the first one is rejected

    Proofreading is not verifying. AI invents references with confidence, and the charter asks what the check covered: sources, figures, consistency. A check is described by what was compared against what.

Five habits that keep you out of trouble

The good practices from Annex A3, in the order in which they serve.
  1. Be precise

    “I used AI” is not enough. Describe the tool, the parts concerned, the nature of the assistance.

  2. Be honest

    A sincere statement is received better than a concealment that comes to light.

  3. Verify

    AI can invent references. Confirm every quotation, every figure, every source.

  4. Keep your significant prompts

    They may be requested in peer review or if your work is challenged.

  5. Never an unapproved tool on internal or confidential data

    The list of approved tools is maintained by the AI Transformation Office, Annex A5 of the charter, and reviewed every quarter.

Your case fits no box

A hybrid piece of work, a journal that imposes its own form, a doubt about where minor assistance ends: the charter does not settle everything. This tool will not guess in its place. Write to the AI Transformation Office, and the answer will serve those who come next.