The AI that interviews you until you know what you think

You tried talking it out once. You had a decision to make, or a reaction you couldn’t quite name, so you opened Voice Memos and talked for four minutes. Then you read the transcript and closed the app. It read like someone talking, because it was: loops, half-sentences, and no clear sign of which version you actually believe.

The problem isn’t that talking is the wrong instinct. It’s that a monologue is the wrong format. Nobody rambles their way to a clear position. But almost everyone gets clear when a good interviewer keeps asking the next question.

Journalists have known this forever. The subject doesn’t write the profile. They answer questions, and the questions do the work of pulling a shape out of a mess of thoughts.

Run the interview method manually

You can do this today with no special tools:

1. Record a two-minute ramble on the thing you’re chewing on, in Voice Memos or anything else.
2. Transcribe it. Apple’s built-in transcription or any dictation tool works.
3. Read it and write down the three questions a skeptical friend would ask. This is the hard part: you have to interrogate your own thinking, which is the known weak point of the Socratic journal method too.
4. Record yourself answering those questions, one at a time.
5. Transcribe again and read it back. Your answers, not the original ramble, are the closest thing to what you actually think. The sentence you defended hardest is usually the belief that was buried under the noise.

The method works. The catch is step 3: writing good questions about your own thinking is exactly what your brain avoids doing. You already agree with yourself, which is why the useful question is whether an AI can ask the follow-ups for you. And the loop is 30–45 minutes of app-juggling, so most people do it once.

The collapsed version

Xtraktr runs that whole loop as a single conversation, so steps 1 through 5 happen without any app-juggling. You talk through the rough version, and it picks the next question from what you just said: where the claim is vague, what the example actually shows, where a skeptic would push. Each question is built on your last answer rather than read off a list, so the thread keeps moving instead of resetting.

You can steer it while it runs. Nudge it to dig deeper when the interesting part is sitting under the claim, to slow down and help you say a fuzzy point more plainly, or to pull the thread together once you’ve talked enough. What you end up with is your own position, tested: the things you can back, the one you couldn’t, and the question you still haven’t answered. That second, cross-examining pass is the whole idea behind critical journaling, and it’s the half most notebooks quietly skip.

Because the AI runs on your iPhone, not in a data center, the half-formed things you say out loud never touch our servers, which matters when what you’re working out isn’t ready for anyone else yet. Here’s what on-device actually means.

Why this beats asking AI what to think

You could paste your dilemma into a chatbot and get an answer in ten seconds. It will be smooth, reasonable, and the same answer it would hand anyone who typed the same prompt. It draws on the average of the internet and knows nothing about your particular situation.

An interview can’t do that, structurally. It has nothing to hand you, because every sentence in the result traces back to something you said: your example, your phrasing, the objection you couldn’t quite answer. The AI’s only job is to ask and to keep the cross-examination moving, so the conclusion stays yours. If you want to see where you sit before you start, the two-minute quiz places you against the method, and the free Critical Journaling book walks through the full five-question version, with no email required to read it.