Large language models (LLMs) are powerful tools, but their effectiveness often hinges on the quality and completeness of the context they receive. This listicle explores the innovative concept of the interrogatory LLM – a technique where the model itself interviews a human to gather that crucial context. Whether you're building a complex feature, verifying a specification, or helping a colleague articulate their expertise, understanding this approach can transform how you work with AI. Here are ten key insights to get started.
1. What Is an Interrogatory LLM?
An interrogatory LLM is a method in which the language model takes on an active, questioning role rather than passively receiving instructions. Instead of a human writing a long prompt, the LLM asks a series of targeted questions to extract the necessary information. The result is a context document – a summary of the human's knowledge – that can then be used for downstream tasks like generating code, drafting plans, or analyzing data. This turns the LLM into a collaborative interviewer, much like a journalist or a consultant, making the human the expert and the model the scribe. The core idea is to leverage the LLM's conversational ability to surface and structure tacit knowledge that might otherwise remain unspoken.
2. The Challenge of Context in Complex Tasks
When using an LLM for anything beyond trivial prompts, context becomes critical. Designing a new feature, for instance, requires descriptions of user interface expectations, implementation guidelines, integration points with external systems, and business rules. All of this easily fills several pages of markdown. The traditional solution is for a human to write this context manually – a time‑consuming and error‑prone process. Even if the expert knows the material well, translating it into a clear, structured prompt demands careful thought. The interrogatory LLM offers an alternative: let the model elicit the context through conversation, which can be faster and more natural for many people.
3. Human‑Written vs. LLM‑Generated Context
The obvious way to feed context to an LLM is to have a human write a detailed document. This works well for writers who are comfortable articulating their thoughts. But many domain experts are not skilled writers, and even experienced ones may find it tedious. The interrogatory approach flips the dynamic: the LLM prompts the human with questions, and the human responds verbally or in short text. The LLM then synthesizes those answers into a coherent context report. This can be especially useful when the information is scattered across the human's mind or when the task requires cross‑referencing multiple sources. The result may have a slightly “AI‑generated” flavor, but it ensures the knowledge is captured accurately.
4. How the Interrogation Process Works
To put this into practice, you prompt the LLM to interrogate you. You tell it the goal – for example, “Ask me all the questions you need to create a design specification for Feature X.” The LLM then begins asking questions one by one. It should cover not only what you know but also what external sources it might consult (e.g., API documentation, company guidelines). As you answer, the LLM builds up a structured understanding. Once it has enough information, it generates the context document for another session – possibly with a different model – to execute the next step. The key is to let the LLM drive the conversation, making it an active partner in knowledge discovery.
5. The Power of One Question at a Time
Harper Reed’s blog first highlighted a crucial refinement: insist that the LLM ask only one question at a time. This prevents the human from being overwhelmed by a barrage of queries and allows each answer to be considered thoroughly. When I tried it myself, I found the LLM frequently needed reminders to stick to this rule. However, the discipline pays off. Single‑question interviewing leads to deeper, more focused responses. It mirrors best practices in human interviewing, where asking one follow‑up before moving on yields richer data. This technique also makes the conversation feel more natural and keeps the human engaged without cognitive overload.
6. Verifying Documents with an Interrogatory LLM
Beyond building context, interrogatory LLMs can be used for verification. Suppose you have a software specification that captures domain knowledge, but you’re unsure if it’s accurate. Instead of asking a human expert to read the entire document – a task many people find difficult – you can feed the document to an LLM and instruct it to interview the expert. The LLM asks targeted questions like “Is section 3.2 still correct?” or “What would you change in the second paragraph?” The expert answers, and the LLM identifies discrepancies. This turns a passive review into an active conversation, often uncovering issues that would have been missed during a silent read.
7. Expert Reviews Made Easier
Reading and reviewing a document is hard work. Many experts, even when they understand the subject perfectly, struggle to spot errors because they read what they expect to see. A conversation with an interrogatory LLM can be more fruitful. It can ask, “You mentioned X earlier – does this section contradict that?” or “Can you think of a scenario where this instruction fails?” By probing specific points, the LLM helps the human think critically. This is especially valuable when the document is poorly written or ambiguous. The expert may feel more comfortable talking through issues than marking up a page, leading to higher‑quality feedback.
8. Combining Build and Review Phases
Naturally, you can use both approaches in sequence. First, an interrogatory LLM interviews the primary expert to build a context document. Then, a second interrogatory LLM interviews a different expert to review that document. This creates a two‑stage pipeline: creation and validation. Each expert only needs to respond to questions, not write or edit documents themselves. The result is a collaboratively built artifact that has been vetted by multiple minds. This mirrors agile or peer‑review practices but reduces the barrier to participation. It is especially useful in interdisciplinary projects where one person holds design knowledge and another holds implementation constraints.
9. Broader Applications: Helping Non‑Writers
While the above uses focus on LLM context for downstream AI tasks, the technique is broadly applicable. I consider myself a natural writer – I need to write to think. But many people find writing extremely difficult. This can be a real problem when their expertise needs to be captured for teammates, documentation, or future reference. An interrogatory LLM can serve as a bridge. Instead of asking them to write a report, you ask them to talk to an LLM. The LLM asks questions, the person answers, and the LLM produces the written artifact. The final output may have a certain “AI‑writing” tang that purists dislike, but it’s far better than having no information at all – or a rushed, incomplete document.
10. Weighing the Benefits and Trade‑offs
The interrogatory LLM approach offers clear advantages: it reduces the cognitive load on domain experts, surfaces implicit knowledge, and can produce comprehensive context quickly. However, it also has trade‑offs. The generated text may lack a human’s unique voice or stylistic nuance. Some people may feel uncomfortable being interviewed by a machine. There’s also a risk that the LLM leads the witness or asks leading questions that bias the responses. Nevertheless, in many practical scenarios – particularly when time is short or writing skills are scarce – the benefits outweigh the drawbacks. The key is to treat the LLM as a tool for collaboration, not a replacement for human judgment.
In summary, interrogatory LLMs represent a shift from human‑as‑prompter to human‑as‑expert, with the AI acting as an inquisitive assistant. By asking the right questions, they unlock knowledge that might otherwise remain inaccessible. Whether you’re designing a feature, verifying a specification, or just trying to capture what someone knows, this technique is worth adding to your AI toolkit. Start with one question at a time, and see how much deeper your next conversation with an LLM can go.