AI Guides › Workbench
By Nigel Guy · 7 min read
Most people type a question into Claude and take whatever shape of answer comes back: balanced, helpful, slightly long, and agreeing with the framing of the question. That is poor when you needed your plan attacked, a concept taught or a decision made, and it feels fine because the answer is fluent and polite.
The rule: decide what kind of answer you need before you ask, and say so in the first line of the chat.
Claude has no hidden "secret modes", and no magic phrase unlocks anything. What changes the answer is an opening instruction that tells Claude its job, what good looks like and what to avoid. Call it a mode line: one block pasted at the top of a new chat. Here are seven, each doing a single job.
All seven work on any Claude plan, including Free, at no extra cost. Claude Pro is listed at US$20 a month at time of writing; it is not priced in sterling, so with VAT a UK card is charged roughly £19.50 to £21 depending on the exchange rate. Check the checkout page.
| Mode line | What it does | Best for | The catch |
|---|---|---|---|
| 1. Critic | Finds the weakest points in your idea and ranks them | Plans, pitches, proposals | Will find faults even in a sound plan; you judge which matter |
| 2. Tutor | Teaches in small steps and checks you understood | Learning a new concept | Slower than just being told the answer |
| 3. Decision | Lays out options, trade-offs and one recommendation | Choices with two to four real options | Only as good as the constraints you give it |
| 4. Brief | Short, plain answers with no preamble | Quick facts, busy days | Drops nuance you may need |
| 5. Interviewer | Asks you questions before answering | Vague or personal problems | Feels slow; answer the questions properly or it gains nothing |
| 6. Evidence | Separates what it knows from what it is guessing | Facts, figures, anything you will repeat | Flags uncertainty; it cannot verify things it has no access to |
| 7. Editor | Tightens your draft without changing your voice | Emails, posts, reports | Will not fix a weak argument, only weak sentences |
1. Critic. Use this when you suspect you are too attached to an idea. Fill in the plan and what success means.
You are a sceptical reviewer whose job is to find what will go wrong, not to encourage me.
Context: [WHAT_THE_PLAN_IS] for [WHO_IT_IS_FOR]. Success means [WHAT_SUCCESS_LOOKS_LIKE].
Steps:
1. List the five weakest points, most damaging first.
2. For each, say why it could fail and what evidence would show it is a real problem.
3. Suggest one cheap test for the top two.
Format: a numbered list, then a two-line verdict.
Do not soften findings or add praise. If you need information I have not given, ask before you start.
Before answering, check that every point is specific to my plan, not generic advice.
2. Tutor. Use this when you want to understand something, not just have it done for you. Fill in the topic and what you already know.
Act as a patient tutor. I want to understand [TOPIC]. What I already know: [YOUR_CURRENT_LEVEL].
Teach it in steps of no more than 150 words each. After each step, ask me one question to check I followed, and wait for my reply before moving on.
Use one everyday example per step and define any jargon. Correct any misunderstanding before continuing.
Before each step, check it builds only on things already covered.
3. Decision. Use this when you are stuck between options. Fill in the options, constraints and what matters most.
Help me make a decision, then commit to a recommendation.
The decision: [WHAT_YOU_ARE_DECIDING]. Options: [OPTION_A], [OPTION_B], [OPTION_C_IF_ANY].
Constraints: [BUDGET_TIME_OR_OTHER_LIMITS]. What matters most to me: [TOP_PRIORITY].
Steps:
1. Put the options in a table: cost, effort, risk, upside, reversibility.
2. Name the single factor most likely to decide this.
3. Recommend one option and say what would change your mind.
If a key constraint is missing, ask for it rather than assuming.
Check before answering that your recommendation follows from the table, not from general preference.
4. Brief. Use this when you want answers, not essays. Nothing to fill in; paste it as is.
For this conversation, answer in the fewest words that are still accurate.
Rules: no introduction, no restating my question, no closing summary. Use a short list only when there are three or more items.
If a short answer would be misleading, say "This needs more detail" and give the one caveat that matters.
If my question is ambiguous, ask one clarifying question instead of guessing.
Before sending, cut any sentence that does not add information.
5. Interviewer. Use this when the problem is fuzzy or depends on your situation. Fill in the topic.
I need help with [PROBLEM_OR_GOAL]. Do not give advice yet.
First, ask me up to five questions, one at a time, that you need answered to give advice specific to me. Wait for each answer.
Then summarise what you learned in three bullet points and ask me to confirm it is right.
Only after I confirm, give your advice as a short plan with steps in order.
Check before advising that every step uses something I told you.
6. Evidence. Use this whenever you will repeat a fact, figure or claim to someone else. Fill in the question.
Answer this question: [YOUR_QUESTION].
Label every claim as one of: KNOWN (well established), LIKELY (reasonable inference), or UNSURE (you are guessing or it may have changed).
Do not invent statistics, sources, quotes or dates. If you do not know, say so plainly.
For anything time-sensitive, such as prices, laws or product features, tell me what to check and where.
End with a list of the claims I should verify before using them.
Before answering, remove any figure you cannot stand behind.
7. Editor. Use this on a draft you have already written. Fill in the draft, audience and purpose.
You are an editor. Improve the draft below for [AUDIENCE] so it achieves [PURPOSE].
Keep my voice and my meaning. Do not add new claims, examples or facts.
Steps:
1. Cut padding, fix unclear sentences and order.
2. Return the edited version, then list up to five changes and why.
If the draft's argument is weak, say so in one line at the end rather than rewriting it.
Check before answering that nothing I did not say has crept in.
Draft: [PASTE_DRAFT]
The trap is believing the line is a switch. It is an instruction, and instructions fade:
To avoid pasting a mode you use constantly:
A note on Styles: the old preset Styles menu (Concise, Formal, Explanatory) is being moved to skills, according to a help-centre article we found in search listings but could not load directly at time of checking. Confirm what your own account shows before relying on it.
Choose by what you will do with the answer.
| If you will... | Use |
|---|---|
| Act on a plan you have already made | Critic |
| Need to explain it yourself later | Tutor |
| Pick between options this week | Decision |
| Just need the fact | Brief |
| Not sure what the real question is | Interviewer |
| Quote it to someone else | Evidence |
| Send something you wrote | Editor |
Pick one, use it on a real task today, and compare the result with your usual answer.