How to Use an LLM as an Interviewer: A Step-by-Step Guide to Gathering Context Through Conversation

Overview

When you need a large language model (LLM) to tackle a complex task—like designing a new software feature or drafting a policy—you usually have to feed it a mountain of context: user requirements, implementation guidelines, system dependencies, and more. This can run to several pages of dense technical writing. The conventional approach is to have a human author that document, but there’s a faster, more conversational alternative: turning the tables and letting the LLM interview you.

An interrogatory LLM asks a person a series of focused questions to extract the exact information it needs. The process is guided by an initial prompt that tells the LLM to act as an interviewer, ask one question at a time, and then synthesize the answers into a structured context report. This report can then be used by another LLM (or the same one) to carry out the actual task. The technique was popularized by Harper Reed, who emphasized the crucial rule of one question per turn—something the LLM often needs to be reminded of.

Beyond building context from scratch, you can also use an interrogatory LLM to review existing documentation. Give it a specification or knowledge base, then have it interview a human expert to check for accuracy or gaps. This is often more effective than asking the expert to read and critique a document, especially if the document is poorly written.

This guide will walk you through everything you need to set up your own interrogatory LLM session, complete with prompts, pitfalls to avoid, and real-world examples.

Prerequisites

• Access to an LLM interface (e.g., ChatGPT, Claude, or an API). You’ll need a model capable of following multi‑step instructions and maintaining a conversation history.
• A defined task that requires contextual information you can’t easily write down. Examples: designing a feature, auditing a process, or creating a policy document.
• A human expert (yourself or a colleague) who has the knowledge the LLM needs to extract.
• Basic prompting skills—you should be comfortable writing a few sentences that set the LLM’s role and constraints.

Step‑by‑Step Instructions

1. Clarify the Goal and the Context You Need

Before you prompt the LLM, decide what the final context report should contain. For a feature design, you might need: user stories, acceptance criteria, technical constraints, external API details, and implementation notes. Write a short bullet list of the sections you expect in the output.

2. Write the Initial Interview Prompt

The prompt should establish the LLM’s role as an interviewer, define the goal, and list the information categories to cover. Include a clear instruction to ask one question at a time. Example:

You are an expert interviewer for a context‑gathering session. Your goal is to create a comprehensive context report for designing a new user dashboard feature. You need the following information: user personas, key functionality, technical stack, security requirements, and external integrations. Ask me one question at a time. After I answer, ask the next question. Do not proceed until you have all the necessary details. Once finished, summarize everything in a structured markdown report.

3. Start the Conversation

Paste the prompt into the chat interface. The LLM should respond with its first question. Answer honestly and as completely as you can. If the answer is a short “yes” or “no”, consider adding a brief explanation—the more context you give, the better the final report will be.

4. Enforce the One‑Question Rule

LLMs often forget and ask multiple questions in one turn. If that happens, politely remind them: “Please ask only one question at a time.” You may need to repeat this after every few turns. It’s a small price to pay for keeping the conversation focused.

5. Let the LLM Consult External Sources (Optional)

Some LLMs can browse the web or access internal databases. If your context requires data from a specific website (e.g., API documentation), you can say: “After you have my answers, consult https://example.com/api-docs to verify the endpoints I mentioned.” This step is not essential but can enrich the report.

6. Generate the Context Report

Once the LLM indicates it has enough information, ask it to produce the final output. You can say: “Now please write the complete context report in markdown, using the categories we discussed.” Review the report for completeness and accuracy.

7. Use the Report for the Next Stage

Copy the context report into a new conversation with a different LLM (or the same one) to perform the actual complex task—e.g., generating prototype code or drafting a specification. The report serves as a ready‑to‑use input.

8. Reverse the Process for Document Review

If you have an existing document (say, a software specification), feed it to the LLM and instruct it to interview an expert to verify its accuracy. Example prompt:

Read the attached specification. Then, act as an interviewer and ask me questions to determine if each section is correct and complete. Ask one question at a time. After my answers, produce a list of corrections and additions needed.

This approach turns a tedious reading task into a dynamic conversation, often revealing hidden issues.

Common Mistakes

• LLM asks multiple questions at once. This overwhelms the human and scatters the context. Remind the model after each infraction.
• Human answers too briefly. A single‑word answer may cause the LLM to misinterpret or miss nuances. Provide complete sentences or references.
• Ignoring the initial prompt instructions. If the LLM deviates (e.g., starts writing code instead of interviewing), re‑prompt with the original role and constraints.
• Trying to write the document yourself first. The whole point is to let the LLM do the writing. Resist the urge to pre‑write paragraphs; let the conversation drive the content.
• Skipping the final summary request. Without an explicit command to produce the report, the LLM may leave the information scattered in the conversation history.
• Using the technique for trivial tasks. It’s best suited for complex, multi‑faceted subjects. For simple questions, a direct prompt is faster.

Summary

The interrogatory LLM approach transforms context gathering from a solo writing chore into a guided, conversational process. By prompting the LLM to ask one question at a time, you can extract deep expertise from a human without requiring them to write a long document. The resulting context report is structured and ready for downstream use—whether for generating code, reviewing specifications, or building knowledge bases. This technique is especially valuable for people who find writing difficult, as it leverages natural conversation rather than formal documentation. The key is to enforce the one‑question rule and to stay engaged as the interviewee. With practice, you can cut hours of documentation work down to a short, productive chat.