You write a claim in your notebook. Three weeks later you come back to it, and instead of polishing the sentence, you write a question in the margin: what do you mean by that? And another: give me three examples. And another: can you think of a counter-example? Then you sit with the questions. The examples you generate don't match the original claim. You write about why they don't match. The claim refines. Karlsson does this in his notebooks as a deliberate practice. It's the painter's loop applied to ideas instead of objects — you're testing your prose against the questions a sharp-eyed reader would ask, except you're doing it yourself before the reader gets there.1
Karlsson describes the protocol in operational detail:
"I'll often interview myself. Like, 'What do you mean? The shadow is yellow — what does that mean? Tell me about that.' I'll literally write these questions to myself in my notebook, in the margin: 'Give me three examples.' I'm interviewing myself. And once I start — okay, so three examples — I'll start thinking about that and then I realize, oh wait, those examples don't actually match what I said there. What's going on?"1
Three moves run through the protocol:
Move 1 — Demand specifics. "Give me three examples." A claim that can't generate three examples is either too abstract or unverified. The forcing of examples extracts the categorical-shadow / actual-shadow distinction. If the writer can produce examples and they match, the claim survives. If the examples don't match, the claim has to refine.
Move 2 — Test against counter-examples. "Can you think of a counter-example of that?" This is the falsification move. A claim that handles its own counter-examples gains weight; a claim that breaks against the first counter-example was overgeneralization.
Move 3 — Interrogate the language. "What do you mean by that?" The habitual phrases — modern society, deep relationships, real freedom — usually mean less than the writer assumed. Demanding the writer unpack them turns vague gestures into specific claims that can then be tested.
The protocol works in writing, not in conversation, for the reasons Karlsson articulates elsewhere: written claims can't morph. The sentence sits on the page as evidence of what you actually thought. In conversation, when challenged, you can say "oh, that's not what I meant" and reshape the claim retroactively. On paper, the claim is fixed. The margin question is a question to a fixed claim, and the writer has to answer it honestly because the claim isn't going to change underneath them.
The margin format also creates a layered text. The original claim is in the body; the questions are in the margin; the answers go below or in a new entry weeks later. The structure makes the interrogation visible to the writer later, when they reread. The writer can see their own thinking refining over time. That visibility is part of what makes the practice generative.
The protocol works best with delay between layers. Karlsson:
"I'll think about something, I'll make some notes about it, then maybe a few weeks later I go back and look at those and question myself."1
The few-weeks gap is doing real work. Immediately after writing a claim, the writer is too close to it — the reasoning that produced the claim is still live in the head, so the writer-interviewer is asking questions the writer-author can deflect by recalling the underlying reasoning. After three weeks, the underlying reasoning has faded; only the claim remains. Now the question lands like a stranger's question. The writer-interviewer is no longer the writer-author's collaborator; they've become the writer-author's reader.
This is also why the protocol scales over months. Karlsson revises claims across multiple notebook entries, going back to the same sentence in different states, asking different questions, accumulating refinements. By the time the claim is ready for an essay, it's been interrogated by the writer-as-stranger many times.
Karlsson's wife plays a role that's adjacent to the self-interview protocol — she's the external version of the same move. She's not a typical editor who crosses out words:
"She'll read a piece and then she'll say, 'Let's throw that out, but that's interesting. Let's have a conversation about that.' And then we'll start — okay, what's a concrete example of this? Or maybe we'll go get some biographies and we'll try to match what I said against the biography and look at: does it fit here? Does it not fit?"1
The wife-as-conversation-editor is the live version of the margin-question. Instead of the writer asking himself in writing, the wife asks him in conversation: what's a concrete example? The protocol is the same — the practice runs toward concrete cases that match or don't match, then refines based on the result.
The advantage of the live version is speed; the writer can iterate faster with a live interlocutor than with notebook entries weeks apart. The advantage of the written version is depth; the temporal gap produces the writer-as-stranger effect that a live conversation can't.
When the margin questions generate examples and counter-examples, those examples don't all end up in the published essay. Most of them stay in the notebook — they're the iceberg underneath the prose. The reader sees the refined claim. The reader doesn't see the seven cases Karlsson tested against, the three that broke the original claim, the new examples that emerged in revision. But those tests are still doing work in the reader's experience — they're the reason the refined claim feels solid. A claim that has survived seven interrogation rounds reads differently than a claim that arrived in the writer's head and went straight to the page.
The protocol is described at lines 158-176 of the transcript. Karlsson gives the operational details (margin questions, demand-three-examples, counter-example tests, repeated interrogation rounds). He cites the wife-as-conversation-editor variant in detail (lines 174-180), including the move to test claims against biographies as external reality-check.
He also names the goal explicitly: "I'm trying to suck intelligence into the piece. I want the piece to end up smarter than I am, you know, because by doing this — that's the beautiful thing about writing, right? Because you have this artifact."1 The protocol is in service of producing prose that contains more intelligence than the writer started with.
The self-interview protocol requires the writer to act as their own adversary, which most writers find aversive. The protocol is structurally identical to the inner critic that Stone & Stone identify as a creativity-blocker — except deployed with different timing. Karlsson is interrogating the claim, not the self. The distinction matters; if the interrogation becomes self-attack ("this is stupid, you're stupid"), the protocol collapses into Critic-shutdown. If it stays focused on the claim ("what do you mean by this, give me three examples"), the protocol generates revision.
The protocol also requires the writer to have decent epistemic standards — to actually accept the verdict when the examples don't match the claim. Writers with strong commitment to existing claims will generate examples that confirm them, ignore examples that don't, and the protocol becomes confirmation theater. The honesty of the practice is upstream of its technique.
Lieberman's diagnostic-interrogation framework (in behavioral-mechanics) and Karlsson's self-interview protocol share an operational structure — both work by asking questions that force specifics ("give me three examples," "what would you do in X situation"). Lieberman applies the structure outward, to read another person. Karlsson applies it inward, to read his own draft. Forced-specification is one of the most reliable methods for surfacing what a person actually thinks, and the method works regardless of whether the target is self or other. Writers who learn behavioral-mechanics interrogation techniques have a sharper self-interview toolkit than writers who don't — they know the exact moves that crack a claim open, because they've watched those moves crack other people's claims.
Orlean's self-editing multi-pass protocol and Karlsson's self-interview converge on temporal-gap as a quality multiplier — both wait for the writer-as-stranger effect before doing the editing/interrogation work. Orlean reads aloud the next day. Karlsson reads margin-questions weeks later. The gap is the operating principle in both. They split on what the temporal gap is used for: Orlean uses it for prose-rhythm and pacing detection; Karlsson uses it for claim-validation. A writer who uses both gets two kinds of revision out of the same temporal apparatus.
Pink's commonplace-book practice (keeping a record of excellent sentences from other writers) sits in interesting relation. Pink collects external claims to study; Karlsson interrogates his own claims to refine. Both build a personal archive of prose-and-thought. They split on whether the archive's primary value is learning from (Pink) or learning what I think by attacking what I wrote (Karlsson). Serious writers maintain layered written archives that accumulate over years, not just drafts. The archive isn't a byproduct; it's the apparatus. A writer with a commonplace book plus an interrogation-notebook running side-by-side for ten years has a different cognitive infrastructure than a writer who keeps only drafts and notes-toward-drafts.
The self-interview protocol is a specific instance of structured self-interrogation in service of better seeing. Look across domains and you keep finding the same move.
Eastern-spirituality: Svadhyaya: Self-Study Through Direct Observation — svadhyaya in the yogic tradition has a textual component: studying one's own writings and journals as evidence of one's own mind. Karlsson's protocol is the operational version of this — using one's own claims as data to be examined rather than positions to be defended. Sit a long-time svadhyaya practitioner down with their notebook from five years ago. They read the old entries as data — what did this person believe, what was missing, what would they ask the writer now. No defensiveness. No identification. Hand the same notebook to someone who has never done contemplative-journaling work, and watch them either cringe at the old self or vigorously defend the positions they held then. The capacity to treat one's own text as data has to be built. Spiritual-traditions-with-textual-self-study lineages contain the craft move; writers without that lineage have to invent it on their own.
Behavioral-mechanics: Elicitation Framework (Hughes) — the elicitation framework's core move is open-ended-questions-that-demand-specifics (Hughes' "Tell me about a time when...", Lieberman's "Give me an example"). Karlsson's "give me three examples" applied to his own claims is elicitation turned inward. Sit a skilled interrogator down with their own notebook. They run the same questions on themselves that they'd run on a target — and they catch their own deflections faster than most people because they know what deflection looks like. The same questions that extract information from a stranger extract information from the self. The asymmetry is that the self can ignore the question, deflect, or generate fake examples — whereas a skilled elicitor working on another person has techniques to detect those moves. Writers who learn elicitation detection techniques catch themselves deflecting their own questions faster, and the self-interview produces more refinement per round.
Psychology: Internal Family Systems — Witnessing the Parts — IFS therapy works by separating the Self (the observing, curious, non-reactive witness) from the parts (subpersonalities holding beliefs, defenses, wounds). Karlsson's self-interview is IFS applied to claims. The writer-interviewer is the Self; the claim is a part. The protocol works when the interviewer maintains witness-distance from the claim instead of identifying with it. Watch a writer who can't dis-identify try to interrogate their own claim — they take the question personally, defend instead of examine, and the protocol collapses into an argument with themselves. Writers who can dis-identify can interrogate them. The skill is identical to the therapeutic skill of self-witnessing under pressure, and writers can develop it through the same psychological work therapists use.
History: Socratic Method and Elenctic Questioning — the Socratic method as Plato preserves it is structurally identical to Karlsson's protocol: a respondent asserts a claim, the questioner demands specifics, the specifics generate contradictions with the original claim, the respondent revises, the questioner demands new specifics. The pattern was already operational in Athens by the fifth century BCE as a method for examining beliefs that initially seem secure. Karlsson reinvents the method privately, with himself as both Socrates and respondent. Forced-specification followed by contradiction-surfacing is one of the oldest known methods for refining belief, and traditions that preserve it (philosophy, law, certain forms of therapy, mathematical proof) all converge on the same operational structure. The protocol is not a private craft trick but the textual instance of a 2,500-year-old epistemic discipline. Writers who cannot tolerate the Socratic interrogation of their own claims are structurally limited in the same ways Plato's interlocutors were — they defend rather than refine, and the prose calcifies around the defended claim rather than developing toward a refined one. Craft and philosophy share the apparatus that produces refined positions across time.
The Sharpest Implication
Most writing advice that targets "clarity" or "rigor" tells the writer to be more careful. Karlsson's protocol says clarity comes from a structural move, not a careful posture. The structural move is to split the writer into two roles (author and interviewer) and let the interviewer prosecute the author's claims after a temporal gap. The writer who tries to be careful in one head, in one session, doesn't actually achieve carefulness — they achieve the appearance of carefulness, which is different. Real rigor requires the role-split and the time-gap. This is why writers who only write quickly, only in one sitting, only with no notebook practice can spend years producing work that seems rigorous but doesn't hold up to outside scrutiny. They haven't installed the interviewer. The work is unchecked. The fix is not effort; it's apparatus.
Generative Questions