Modes/Custom mode — write a prompt

How to write a great promptThe Custom mode is only as powerful as your prompt. This guide explains every element that makes a system prompt effective for multi-AI debate.

The AI scoring system (0–100)Before launching a Custom debate, Alethe AI evaluates your prompt using Gemini Flash across 4 criteria — each worth 25 points. The score gives you instant feedback on whether your prompt will generate a rich, multi-perspective debate.

≥ 80 — Excellent

55–79 — Good

< 55 — Weak

The 4 scoring criteria

25 pts

Multi-AI relevance

Is the topic rich enough for multiple AIs to bring distinct perspectives? An overly factual prompt ("What is the capital of France?") generates no useful disagreement. A good prompt leaves room for analysis, interpretation, or judgment.

GOOD

Give me a comprehensive analysis of the rise of populism in Western Europe since 2015.

WEAK

Who won the 2022 World Cup?

25 pts

Debate potential

Does the question allow for multiple legitimate positions? The best prompts pose a problem where reasonable experts can diverge — not necessarily a moral controversy, but a topic where models can hold different angles.

GOOD

Should open-source AI models be regulated? Analyze the arguments for and against.

WEAK

What is 12 × 14?

25 pts

Flow

Is the prompt structured, clear, and unambiguous? A well-formulated prompt states the context, the objective, and the expected format. It avoids contradictory instructions, vague phrasing, or constraints that are impossible to satisfy simultaneously.

GOOD

You are a senior financial analyst. Evaluate the growth strategy of a 10-person B2B SaaS startup that just raised €2M. Focus on 18-month viability.

WEAK

Answer very briefly but also in great detail and exhaustively without being too long.

25 pts

Language clarity

Is the prompt written in clear, correct language? Spelling errors, language mixing, or cryptic phrasing degrade AI response quality. A prompt that is precise in its language produces precise responses in that same language.

GOOD

Analyze the ethical implications of using AI in judicial decision-making.

WEAK

u AI analyze this for me pls thx

The 5-element structureA complete Custom prompt combines these 5 elements. You do not need all 5 every time — but the more you include, the more precisely the models will behave.

The roleWho are the AIs?

Start by defining the identity you assign to the models. A precise role anchors the register, expertise level, and attitude. The more specific the role, the more relevant the responses.

"You are a lawyer specializing in French corporate law with 20 years of experience."

The domain & contextWhat is the framework?

Specify the domain of expertise, industry, geography, or time period involved. Avoid generalities — the more precise the context, the less the models have to guess.

"Context: French deep-tech startup, Series A, 15 employees, European B2B market."

The constraintsWhat is forbidden or required?

Explicitly list what the models must do or avoid. Examples: cite sources, use specific vocabulary, avoid recommendations outside your sector, structure in bullet points.

"Always cite legal statutes. Never give medical advice. Respond only in English."

The expected formatHow should the answer be presented?

Indicate whether you want a structured analysis, a step-by-step reasoning, a comparison table, a concise answer, or a long development. Models adapt to your format if you specify it.

"Structure your response as: (1) analysis, (2) risks, (3) recommendation in max 3 points."

The toneWhat communication register?

Define whether models should be formal, direct, educational, provocative, supportive, or neutral. An unspecified tone leaves the models to choose — often too academic or too generic.

"Tone: direct, no unnecessary diplomacy. Get to the point. Challenge if needed."

5 common mistakes

Being too vague

"Tell me about economics."

Specify the sub-domain, the concrete question, the desired angle.

Contradictory instructions

"Be very concise AND very exhaustive."

Choose one objective: speed or depth. Not both.

Purely factual question

"What year did the French Revolution happen?"

Turn it into analysis: "What conditions made the French Revolution inevitable?"

Forgetting user context

"Advise me on my business."

Give your sector, size, stage, specific problem. AIs don't guess.

Generic role

"You are an expert."

Expert in what? For how long? For what type of client?

Real examples with scores

ExcellentScore / 100

You are a partner at a strategy consulting firm (McKinsey-type). Your client is a mid-sized French company (500 people, food & beverage sector) considering expansion into sub-Saharan Africa. Analyze the opportunities and risks of this expansion across 4 axes: market, operational, regulatory, and financial. Constraints: cite quantitative data where possible, structure your response in clear sections, and conclude with a 3-point actionable recommendation.

GoodScore / 100

You are a cybersecurity specialist. Explain the main threats facing an SME in 2025 and how to protect against them. Be concise but complete.

WeakScore / 100

You are an expert. Talk about technology and give advice. Thanks.

Ready to try?Launch a Custom debate on Alethe AI. The AI evaluates your prompt before you start — you get a score and tips to improve it.

Launch a debate

Conversation modes