The best way to use ChatGPT for Model UN, or Claude, Gemini or Perplexity, is as a tutor and sparring partner rather than a ghostwriter: ask it to explain the background, generate the questions you should be researching, summarise long UN documents you paste in and rehearse counter-arguments, then verify every fact against primary UN sources before it goes anywhere near your position paper. Used that way, AI makes you a better-prepared delegate; used to write your paper, it makes you a worse one and a detectable one.
Key takeaways
- AI tools are good at explaining background, generating research questions, summarising documents you provide, rehearsing counter-arguments and checking clause grammar.
- AI tools are unreliable for specific facts, citations, resolution numbers, vote counts and your country's actual voting record; they can invent all of these convincingly.
- Using AI to prepare is not cheating; submitting AI-written work as your own is, and experienced chairs can usually tell.
- Disclose AI use if the conference asks, and never submit an AI-written position paper under your name.
- Every claim that reaches your position paper or speech should be traced to a primary source such as the UN Digital Library, UN Press or an official government statement.
- What AI is good for in Model UN preparation
- What AI is bad for, and why
- Is using AI in Model UN cheating?
- How to use ChatGPT for Model UN research: a workflow with prompts
- Can you use ChatGPT for a position paper?
- The verification checklist against primary UN sources
- Which AI tool should a delegate use?
- Where to go next
What AI is good for in Model UN preparation
AI is good for the parts of Model UN preparation where you need explanation, structure and practice rather than facts you will be held to. Large language models, the technology behind these tools, are strong at rephrasing, organising and questioning; they are weak at knowing.
Used well, an AI tool can:
- Explain the background. Ask it to explain a topic, a treaty or a UN body as if to a sixteen-year-old, then again as if to a diplomat.
- Generate question lists. Ask what a well-prepared delegate for your country would need to know about the topic, then treat the list as a research plan.
- Summarise long documents you paste in. A long report or a dense resolution can be summarised accurately when you provide the text yourself, because the model is reading rather than remembering.
- Rehearse counter-arguments. Ask it to play the delegate of a country that opposes your position and to attack your opening speech.
- Check clause grammar. Preambular and operative clauses follow strict conventions. AI is reliable at spotting a missing semicolon or a verb that belongs in the wrong section.
What AI is bad for, and why
AI is bad for anything you cannot independently verify, because a language model produces plausible text rather than checked text, and it will produce a plausible resolution number, vote count or quotation with the same confidence as a real one. The behaviour is usually called hallucination, and it is a property of how these tools work, not a rare bug.
Be especially cautious with:
- Specific facts and figures. Dates, numbers of signatories and funding amounts are exactly the details a model is most likely to invent or round.
- Citations. A model may give you a document symbol, a title and a paragraph number that do not exist together, or at all. Never cite a source you have not opened.
- Your country's voting record. Whether your country voted for, against or abstained on a particular resolution is a matter of record in the UN Digital Library. A model's answer is a guess dressed as memory.
- Quotations. Attributed quotations from ministers, ambassadors or the Secretary-General should be treated as fabricated until you find the original transcript or press release.
Tools with web search, such as Perplexity or the browsing modes in ChatGPT, Gemini and Claude, reduce this problem but do not remove it. The link they show you is the beginning of verification, not the end.
Is using AI in Model UN cheating?
Using AI to understand a topic, plan research or practise arguments is not cheating; submitting AI-written text as your own work is, in the same way that handing in a paper written by a friend would be. Model UN is an exercise in representing a position you have understood, and a paper you did not write is evidence of a position you have not understood.
A simple honesty framework covers almost every situation:
- Follow the conference's rule. Some conferences now publish an AI policy in their delegate handbook or position paper guidelines. Read it, and if it asks you to disclose AI use, disclose it fully.
- Never submit AI-written work as your own. Ideas, outlines and feedback are fine; sentences you did not write are not, whatever the policy says.
- Assume the chair can tell. Experienced chairs read dozens of position papers on the same topic. AI-written papers share a rhythm, a vocabulary and a vagueness about the country's actual record that stands out, and the mismatch becomes obvious the first time you speak without notes.
- Ask whether you could defend it. If a chair questioned you on any sentence in your paper, could you explain where it came from? If not, it should not be there.
The practical case is stronger than the ethical one: delegates are assessed in the room, and our guide to how awards are decided shows how much of that happens in caucus.
How to use ChatGPT for Model UN research: a workflow with prompts
A sound workflow uses AI at the start of research to orient you and at the end to challenge you, and uses primary sources in the middle for everything you will actually claim. The prompts below work in ChatGPT, Claude, Gemini or Perplexity; replace the bracketed parts with your assignment.
Step 1: Orientation
Ask for a plain-language explanation before you open a single UN document.
Explain the topic "[topic]" as it is debated in the [committee] of the United Nations. Cover: the core problem, the main disagreements between groups of countries, the key treaties or resolutions I should know by name, and the vocabulary I will need. Do not give me statistics or citations; I will find those myself.
Step 2: A research plan
Turn the explanation into a list of questions you will answer from primary sources.
I am representing [country] in the [committee] on the topic "[topic]". List fifteen questions a well-prepared delegate should be able to answer about my country's position, interests, history and allies on this topic. Group them under: national interest, past votes and statements, regional bloc, and likely opponents. Do not answer them.
Step 3: Primary-source reading
Answer the questions yourself, from the UN Digital Library, UN Press releases, official statements from your country's mission and permanent government sources. This is the step no tool can do for you; our module on researching your country explains where each kind of document lives.
Step 4: Summarising what you found
When a document is long, paste the text in and ask for a summary tied to your questions.
Below is the full text of [document title and symbol]. Summarise it in 200 words, then list every paragraph that mentions [country] or [regional group], quoting the paragraph number and the exact wording. Do not add anything that is not in the text.
Step 5: Drafting in your own words
Write the position paper yourself from your notes. Then ask for feedback on structure and clarity, not for rewrites.
Here is a position paper I wrote. Do not rewrite it. Tell me: which paragraph is weakest and why, where my proposed solutions are vague, and which claims a sceptical chair would ask me to source. Answer as a list.
Step 6: Rehearsing the room
Use the model as an opponent.
Play the delegate of [opposing country] in the [committee]. I will give my opening speech; respond with the three strongest objections that delegate would raise, in character, and then ask me one difficult question. Keep each response under 120 words.
Step 7: Clause checks
Check these draft resolution clauses for format only: correct preambular and operative verbs, punctuation (commas after preambular clauses, semicolons after operative clauses, a full stop after the last), numbering and consistency of terms. Do not change the substance.
Our draft resolution guide sets out the conventions the model is checking against.
Can you use ChatGPT for a position paper?
You can use ChatGPT for a position paper in the way you would use a teacher: to explain the topic, question your draft and point out weaknesses, but not to write the paper itself. A ChatGPT position paper submitted under your name is both a breach of the honesty framework above and, in practice, a weaker paper, because the model does not know your country's real record and will fill the gap with generalities.
A useful test: after the paper is finished, close every tab and explain your country's position aloud for two minutes without notes. If you can, the paper is yours; if you cannot, you have a document, not a position. Our free guide to writing a position paper covers the structure chairs expect.
The verification checklist against primary UN sources
Every factual claim in your paper or speech should be traced to a primary source before you rely on it, and the three sources that cover most Model UN research are the UN Digital Library, UN Press and the official statements of your country's government and permanent mission. Treat anything an AI tool told you as a lead until it passes this list.
| Type of claim | Where to verify it | What to record |
|---|---|---|
| A resolution exists and says what you think | UN Digital Library (search by document symbol) | Symbol, date, the exact operative paragraph |
| How your country voted | Voting record in the UN Digital Library | Vote (yes, no, abstain, absent) and date |
| What your country said in debate | UN Press meeting coverage; the mission's own website | Date, speaker, a short exact quotation |
| A treaty's status | UN Treaty Collection | Whether your country signed, ratified or neither |
| A statistic | The UN agency that produced it (its own report or data portal) | Report title, year, page |
| A quotation from an official | The original transcript, press release or speech text | Source and date; if you cannot find it, do not use it |
When an AI tool gives you a lead you cannot find in any primary source after ten minutes, drop it; a missing claim costs nothing, and an invented one can cost a chair's trust.
Which AI tool should a delegate use?
For Model UN research with AI, the differences between ChatGPT, Claude, Gemini and Perplexity matter less than how you use them, and any of the four is adequate for the workflow above. A few neutral distinctions are worth knowing.
- Perplexity is built around web search and shows its sources by default, which makes it convenient for the orientation stage, provided you open the links.
- ChatGPT, Claude and Gemini are general assistants that handle long pasted documents, role-play and drafting feedback well; their web access depends on the version and settings you are using.
- All four can be wrong with confidence, and none of them has privileged access to UN records. The primary sources are the same whichever tool points you towards them.
Whichever you choose, keep the tool out of the last step; the 30-day preparation plan shows how to schedule verification and rehearsal.
Where to go next
The free Youth Impacts training curriculum covers the research that AI cannot do for you: start with researching your country and how to write a position paper. At The Global 60, every delegate sits in one of three chambers of approximately twenty and is assessed on research, diplomacy and drafting in the room; the Assembly page explains the format and admission and fees explains how to apply.
Frequently asked questions.
Is using ChatGPT for Model UN cheating?
Using AI to understand a topic, plan research or rehearse arguments is not cheating. Submitting AI-written text as your own work is, and most conferences that publish an AI policy say so. Disclose AI use if the conference asks.
Can ChatGPT write my position paper?
It can produce a document, but you should not submit it. The model does not know your country’s real record, chairs recognise the pattern, and you will have nothing to say once debate moves past your first speech. Write it yourself and use AI for feedback.
What is AI most useful for in Model UN preparation?
Explaining background in plain language, generating research questions, summarising long UN documents you paste in, role-playing an opposing delegate, and checking the format of resolution clauses.
Can I trust AI for my country’s voting record?
No. Voting records are a matter of record in the UN Digital Library, and a model’s answer is a guess. Look the vote up yourself and record the date and outcome.
Which is best for Model UN: ChatGPT, Claude, Gemini or Perplexity?
Any of them works for the workflow in this guide. Perplexity surfaces web sources by default; ChatGPT, Claude and Gemini handle long pasted documents and role-play well. All four can be confidently wrong, so verify against primary sources whichever you use.
Do chairs really notice AI-written papers?
Experienced chairs read many papers on the same topic and recognise the shared rhythm, generic solutions and absence of specific resolutions or statements. The mismatch becomes obvious the first time the delegate speaks without notes.
Youth Impacts is a Model United Nations organisation. Its flagship New York experience is The Global 60: sixty delegates, three small chambers, 22 – 25 April 2027, admission by application and interview. Youth Impacts Model United Nations (YIMUN) has run since 2025. Questions: info@youthimpacts.com. Published 19 September 2026.