# Persona & prompt guidelines Setting up your AI’s persona is more about precision than giving it a personality. Because Ebbot uses a highly focused language model tied to your specific company documents (a RAG setup), how you write your instructions directly impacts how well the AI performs. Think of it less like talking to a human and more like programming with plain English. #### Start small and keep it organized The most common mistake when setting up a persona is writing a massive paragraph of instructions. This is known as "prompt bloat," and it dilutes the importance of everything you write. Aim to have as little information in the persona as possible. Start with a short, general baseline. #### To help the AI process your rules, use a clear structure. Treat the AI like it’s reading a manual by using clear, distinct headings for things like Role, Tone, and Boundaries. This makes it much easier for the model to navigate its instructions than sifting through a wall of text. {% hint style="info" %} We use Markdown format for the text structure of our prompts. This means you can easily create headings by using hashtags (for example, typing `## Role` or `### Rules`). {% endhint %} #### Beware the "Helpful" trap It is tempting to use vague instructions like, "You are a helpful and friendly support agent." In the AI world, instructing a model to be "helpful" can backfire. It could cause the AI to prioritize giving a helpful-sounding answer over an authentic, factual one. An AI told to be purely "helpful" might hallucinate a refund policy or agree with a frustrated customer just to please them. Instead, replace vague adjectives with concrete directives. Tell the AI to be "accurate, objective, and polite," and explicitly instruct it to base its answers solely on your uploaded knowledge base. #### Guide the output without forcing it to count A frequent pitfall is giving the AI strict numerical constraints, like "Keep your answer below 150 tokens" or "Write exactly 50 words." Language models do not plan their sentences ahead of time; they generate text word-by-word. Forcing the AI to act as a word counter drains its reasoning capacity which can result in degraded answer quality. If you want shorter responses, use formatting constraints instead. Tell the AI to "be concise," "respond in a single short paragraph," or "use a maximum of three bullet points." #### Show, don't just tell If you want the AI to behave a certain way, the fastest way to achieve it is by giving it a concrete example. Include one or two ideal Question-and-Answer pairs right in the persona setup. Showing the AI exactly what a good response looks like is far more effective than spending three paragraphs trying to describe it. You also need to establish what the AI should *not* do. Defining this "anti-persona" is crucial for setting boundaries. Explicitly state things like, "Never offer discounts," or "If the answer is not in the provided documents, say you do not know." #### Treat tweaks like a science experiment Setting up the perfect persona requires trial and error, but it needs to be methodical. If the AI gives a bad answer and you rewrite the entire prompt in response, you won’t know which change actually fixed the problem, or what you might have accidentally broken in the process. Make a habit of changing one variable at a time. If the answers are too long, adjust the length instruction and test it. If the tone is wrong, tweak the tone instruction and test it. Making small changes and immediately testing the effect will save you a lot of frustration in the long run. --- # Agent Instructions: Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter: ``` GET https://docs.ebbot.ai/ebbot-docs/core-capabilities/ebbotgpt/ebbotgpt-best-practices/persona-and-prompt-guidelines.md?ask= ``` The question should be specific, self-contained, and written in natural language. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.