How to Talk to AI? A Biologist's Prompting Guide

You’re several minutes into a back-and-forth with an LLM chatbot (ChatGPT, Claude etc.), trying to get a clear answer about interpreting some results. But, it keeps circling back to the same generic explanation. The AI seems confused and continues to apologize and you're frustrated!

I've been there. We've all been there.

These tools don't work like Google. When you type a question into Google, it finds all relevant existing information. It retrieves information and you can choose what you are looking for. But when you ask an LLM something, it's not searching through an index. It's predicting the next most likely word, then the next, then the next building an answer word by word based on patterns it learned from billions of texts. That means every word in your question (also known as prompt) matters.

Think about it this way, How will you explain PCR to a 10 year old and a graduate student ? Same concept but different explanations. You will adjust your language, and examples based on your listener. Similarly AI can change its answers based on cues from you.The key is being clear and specific about what needs to be done, providing context for how it should think, and defining exactly what output you expect.

A Simple Prompting Framework

Let me show you an example for how to prompt LLM for RNA-seq analysis:

---

<Role> You are an expert in RNA-seq and cancer biology

<Task>Interpret the given differentially expressed genes from a cancer vs. normal to extract biologically meaningful insights.

<Inputs> I've uploaded a CSV file containing differentially expressed genes with the following columns:

• gene_symbol

• log2FC

• FDR 

  
  

<Process>

1. Filter genes meeting the criteria (log2FC > 2, FDR < 0.01)

2. Group genes into functional clusters (e.g: immune signaling)

3. Identify enriched pathways for each cluster

   
   

<Constraints>

• Use only the genes provided in the uploaded file.

• Avoid generic pathway names and be specific

• Keep interpretations cancer focused

  
  

<Output format> Return results in the following structure:

1. Functional Clusters: Give 2 clusters with short descriptions

2. Pathway Insights: Give 2 enriched pathways per cluster with short justification

3. Gene-Level Highlights: For each gene, give 2–3 bullet points on biological significance

---

I am assigning a role that gives the model a defined expertise, ensuring its reasons like an RNA-seq expert or cancer biologist instead of a general purpose assistant. A clearly defined task removes ambiguity and tells the model exactly what kind of output is needed,  interpretation, summarization, or comparison.

Using delimiters (<>, ‘’, :) separates raw data from instructions so the model doesn’t confuse the two or pull irrelevant information into its reasoning. A step-by-step process acts like a protocol, guiding the model through the steps to reach the required output. This prevents the issue of AI jumping straight to general conclusions. If relevant, do add sample solutions or examples to improve the results.

Including conditions such as checks, or specific criteria prevents the model from analyzing data that doesn’t meet experimental requirements. Finally, defining the output format ensures the answer is structured, and concise. Together, these rules help the model think methodically making it more specific, and helpful in biological research.

Where AI Actually Helps in Scientific Research

When prompts are structured well, AI becomes useful in the places where scientists spend the most time outside of the bench.

Literature Survey

LLMs can quickly find relevant papers and parse them for summaries, extract key findings, highlight strengths or weaknesses, and quickly compare multiple studies. 

Conceptual Understanding

It can act as your personal tutor while learning new concepts. It can provide explanations, outline assumptions, and clarify any follow-up questions. It can also adapt explanations to the specific biological questions or data.

Scientific CommunicationDrafting introductions, figure captions, or internal reports becomes faster when AI generates structured text 

Data Interpretation SupportWhile AI does not replace statistical analyses, it can assist in framing narratives, suggesting QC, analyses types relevant to given data and interpretation. 

Research and Strategy

When working with multiple datasets, or files, AI can integrate information to identify gaps, untested assumptions, and potential research directions. It can also be a great research partner to just brainstorm a new idea.

Takeaway

AI cannot replace scientific judgment but it can significantly reduce the time spent on literature survey, explanation or interpretation allowing researchers to focus on the experimental decisions where expertise matters the most.

Next time you're stuck in that loop, trying to get a decent answer out of AI, Pause. Take a step back. Think about what you're really asking for, how you want it approached, and what you need in return

You might be surprised at how much it can change the output!

PS: This does not include AI-assisted coding, only scientific reasoning and research applications.

References

• Prompting Guide: https://www.promptingguide.ai/

• Google Cloud: Introduction to Generative AI

• GPT 5 and Sonnet 4.5 for editing