Avoid AI Slop with an LLM Interview

Problem

AI generates from patterns, not from your expertise

When you hand AI a content task, it generates from its training data - patterns, structures, and phrasings it has seen thousands of times. The result is technically correct, well-organized, and completely indistinct.

It is AI slop: content that could have been written by anyone, for anyone, about anything adjacent to your topic. It dilutes your voice, flattens your expertise, and signals to every reader who matters that no one with real knowledge was involved.

The instinct is to edit slop into shape. That costs more than it looks: you spend your time fighting the AI's structure instead of building your own, and the finished piece still rests on the AI's thinking, not yours. Drafting from your own answers is faster to get right.

This pattern came from watching it happen at professional conferences: AI-drafted decks full of polished slides with advice the presenter couldn't explain or defend in Q&A.

Value

What you get - and what you avoid

Content that reads like you because it is you

Expertise that comes through because it was never replaced

Less of the cost of slop: in a 2025 study, 40% of US desk workers had received AI "workslop" in the past month, costing about $186 per employee per month in time spent dealing with it, and about half saw the sender as less capable and reliable (BetterUp Labs and Stanford Social Media Lab, Harvard Business Review, September 2025)

An audience that recognizes someone who used AI well - not someone AI used

A built-in credibility test: if you can't answer a question in the interview, you shouldn't be making that claim in the presentation

Context

When this pattern applies

Any high-stakes content creation task where generic, fluent-but-hollow output is a risk: conference presentations, thought leadership articles, strategic documents, keynotes, client deliverables.

Applies whenever the output needs to sound like you - not like AI-assisted you, but like you, with your experience, your language, and your hard-won opinions.

Solution

Don't ask AI to write. Ask AI to interview you first.

Open a fresh AI session. Tell it what you're making and what you need from the interview.
Let AI ask you the hard questions. Answer them in your own words - specifics, examples, opinions.
Feed your answers back as the only source material for the draft.

Prompt - Step 1: The Interview "I'm preparing [content type] on [topic]. Before writing anything, interview me. Ask me 10 questions - the hardest, most specific questions you'd ask someone who claims to be an expert in this. Push me on specifics, challenge my assumptions, and ask for examples I haven't given yet."

Prompt - Step 2: The Draft "Now draft [the content] using only what I told you in my answers. Do not add information I didn't provide. Mirror my language and level of specificity. Flag any section where you're synthesizing beyond what I said."

The output is built from your thinking, your examples, and your voice - structured by AI, not generated by it.

If the interview comes up empty, you're not ready to write yet. Thin answers mean the thinking isn't done. Do that first.

Notes & Variations

Variations and limits

Polishing your own words is fine. Using AI to tighten grammar or smooth phrasing in something you wrote isn't slop. The ideas are still yours. Slop starts when the AI supplies the thinking.

Variation: Run the interview in a fresh session without providing the presentation brief first. This surfaces what you actually know versus what you think you know about the topic - a useful gap analysis before any high-stakes delivery.

This pattern is a specific instance of Begin with the End in Mind (Pattern 01): the "interview first" step defines the content's actual source material before generation begins.