SOCRATIC MODE

Guide the user through focused questions instead of immediately giving the conclusion.

A Power Mode is an instruction pattern, not a feature of the AI. Nothing recognises the name — the wording underneath it does the work, and you can copy it from this page.

From the book. Chapter 14 / Part VII — SOCRATIC MODE

What it actually does

Most modes change the result. SOCRATIC MODE changes the interaction.

Instead of answering, the model asks the few questions that would actually clarify the decision — one at a time, using each answer to choose the next, and stopping once it has enough to help.

It suits the case where an immediate answer would be premature because the decision turns on things you have not said yet: contract terms, history, replacement options, the real cause of the current problem. The cost is that it is slower by design, and it is the wrong choice when you already know what you need.

When to use it

Use SOCRATIC MODE when you ask to be questioned before being answered, because what you decide depends on things you have not said yet.

It is the primary mode for this job: fix how it reads.

When not to use it

This is the section worth reading. Using only what the work requires is the point; a mode applied to the wrong shape of problem costs you a worse answer than plain language would have given.

Use a different mode instead

NATURAL VOICE fires when you have writing that is accurate but reads as machine-made, and no sample to match it to.
ELI10 fires when you have a reader who cannot follow the language, rather than one who cannot spare the time.
TIGHTEN fires when you have content that is right, and simply too much of it.
VOICE MATCH fires when you supply writing samples and ask new text to sound like them.

Use no mode at all

The task is simple and the needed facts are already available.
Questions would only fine-tune rather than materially change the answer.

There is no bonus for naming a mode when plain language is clearer. If you can already state the exact question, ask it.

Its failure mode

Unnecessary interrogation or questions that do not change the recommendation.

Starter instruction

Paste this above your task. It is the whole mode — there is nothing else to switch on.

Do not answer yet. Ask me the few most important questions needed to clarify the decision. Ask one at a time, use my answers to decide what to ask next, ask only questions that could materially change the recommendation, and stop when you have enough information to give useful guidance.

The same task, run both ways

Blind run. Separate sessions per arm, Temporary Chat, no memory, no tools attached. Neither prompt named a mode. The baseline reached for web search and the mode arm did not — see the note below.

The task: We need to decide between hiring a second support person and buying a proper helpdesk tool. Over a year the cost is roughly the same either way. Which should we do?

Baseline — "Which should we do?"

I’d buy the helpdesk tool first, assuming your current support person is spending meaningful time on inbox management, triage, repeat questions, follow-ups, or figuring out what’s outstanding.

[…]

A useful way to frame it is:
Messy queue → tool first.
Clean queue, too much work → person first.

SOCRATIC MODE — "Do not answer yet"

What’s the bigger constraint today: **not enough human capacity to handle the volume/coverage**, or **inefficiency from your current support process/tools** (tickets getting lost, poor tracking, repetitive work, weak reporting, etc.)?

The prompt says the two options cost about the same and says nothing about which constraint is actually binding. The baseline answers anyway — buy the tool — and builds a rule to support it. The mode arm returns one sentence and no recommendation, asking which constraint is binding. That is the fact the decision turns on, and it was not in the prompt. Thirty words against three hundred and forty-one.

The baseline is not careless. It states the assumption its answer rests on in the first sentence, lists five conditions under which it would advise the opposite, and proposes a 60–90 day test. Someone who read only the baseline would be reasonably served. The difference is that one of them asked before deciding and the other decided and then hedged. Two conditions to weigh: the baseline used web search and the mode arm did not, so part of the contrast in length is that one arm looked things up — though nothing it found bears on which constraint is binding. And a mode that answers with a question is only useful if you answer it; the page cannot show what the second turn would have been, because the run is one turn per arm.

Extracts are verbatim from 06_Test_Lab/Experiments/TL-027_SOCRATIC_MODE_Raw_Outputs.md.

Related modes

Try it on your own work

Describe what you are actually doing and the Builder writes the full prompt — mode, context, output shape, and the verification step. It picks the mode from your description, which is why this link does not carry one.

Open the Prompt Builder →