Rubber Duck Dictation: Rubber Duck Debugging That Writes Itself Down

Quick answer: Rubber duck dictation is rubber duck debugging with the answer kept. You talk a problem through out loud, the way you would to a rubber duck, and instead of the clarity evaporating the moment you stop talking, you get back clean, structured text. Three parts are required: you explain out loud in order, something captures it without you managing the capture, and what comes back is readable rather than verbatim. Drop the third part and you just have dictation.

Rubber duck debugging has one hole in it, and everybody who uses the method has fallen through it.

You explain the problem out loud, you hear the flaw somewhere in your own sentence, you fix it, you feel clever. Then Thursday arrives and what you have left is the memory that you solved something, not the reasoning that solved it. The clarity was real. It is also gone.

Rubber duck dictation is that method with the answer kept.

The definition

Rubber duck dictation (n.): talking through a problem out loud and getting back clean, structured text instead of a raw transcript.

Three parts, and all three are load-bearing.

1. You explain, out loud, in order. This is the forcing move, borrowed whole from the rubber duck method. Narrating in sequence makes you compare what the thing is supposed to do against what it actually does, and those two versions cannot keep quietly agreeing once they are both out in the air.
2. Something captures it without you managing the capture. Not a recording you have to remember to start, name, and come back to. Hold a key, talk, let go.
3. What comes back is readable. Not a transcript. The false starts, the “um,” the three times you restarted the same sentence: gone.

Drop the third part and you have dictation. Drop the first and you have a voice memo you will never open again. All three together is the thing that needed a name.

Why it is not just dictation with extra steps

Dictation has always assumed you already know the sentence. You have the letter in your head, your hands are busy or slow, so you say the words and the machine types them. Fidelity is the whole job. A dictation tool that changed your words would be broken.

Rubber duck dictation assumes the exact opposite. You do not know the sentence yet. Not knowing it is the point, and groping toward it out loud is the work.

Which means fidelity is the wrong target. Verbatim is the right answer for a deposition and the wrong answer for a thought. If you record someone thinking and play it back word for word, you get a bad transcript of a good idea: seven filler words, two abandoned sentences, and the actual insight buried in the middle of the third attempt. The reorganizing is not a nice-to-have on top of the capture. It is the difference between a recording and a note.

How to run it

Five steps, and the first one is where people cheat.

1. Say the goal before any detail. Out loud. “I am trying to work out why the invoice job double-charges on retries.” Skipping this is the single most common way the method fails, because context you never stated is context you never checked.
2. Go in order, in more detail than feels necessary. The step you almost skip because it is too obvious is reliably the one hiding the problem.
3. Narrate the flinch. When you hear yourself hedge, hand-wave, or say “that part should just work,” say so out loud instead of gliding past it. That sentence is the bug.
4. Do not edit while you are talking. The urge to restart the sentence cleanly is the urge to perform instead of think. Let it be messy. Cleaning up is a job for later and not your job.
5. Read it back once, immediately. Ninety seconds. You are checking whether the thing you concluded survives being read, which is a different test than whether it survived being said.

What it is good for

The parent method was described in terms of code because a programmer in 1999 happened to carry a duck. The listener needs no context, so the subject never mattered.

• A bug you have reread five times. The original case, and still the best one.
• A paragraph that will not come together. Explain what you are trying to say to nobody, and the paragraph is usually sitting there in what you just said.
• A decision you keep circling. Walk a stranger through the case for taking the job. You will hear which reason is doing real work and which one you have been repeating to sound certain.
• A prompt you are about to send to an AI. Speaking makes it cheap to include the context a real explanation needs, which we went into in Is Talking to AI Faster Than Typing. Explain first. Let it answer second.

The honest limits

This inherits the boundaries of the method it is built on, and there is no reason to pretend otherwise.

Thinking out loud reliably helps analytic, step-by-step work and can get in the way of pure-insight problems, the ones that need a quiet unconscious leap. One study found 13 percent of people solved an insight problem while verbalizing against 57 percent in silence, while talking helped on a step-by-step problem. That result has a mixed replication record and small samples, so treat it as a real but contested boundary rather than a law. The practical read is unchanged: aim this at reasoning you can lay out in order, and when you are waiting on a leap that has not arrived, stop narrating.

Two more, specific to the dictation half:

• What comes back is a strong first draft, not a finished document. It is a note you can actually use, not a post you can publish.
• If you need the exact words, this is the wrong tool. Interviews, quotes, anything legal: you want faithful transcription, and you want to keep the raw audio.

Keep what falls out

The method has always been free. A duck costs three dollars and talking to it costs nothing. The expensive part was never the thinking, it was that the thinking did not survive the conversation.

That is the entire reason Ducktate exists. Hold a key on your Mac, talk the problem through line by line, let go, and the walk-through files itself as a clean, searchable note. Transcription runs on your machine, so the messy version of your thinking never leaves it.

You do the explaining. Something else remembers what you worked out.