Guide · 5 min read
10 Words That Make Claude Dramatically Better
Most prompt engineering advice tells you to write longer, more detailed prompts. But some of the biggest improvements come from a single word or short phrase that changes how Claude approaches a task entirely.
These aren't tricks — they're legitimate instructions that activate specific behaviors Claude is capable of but doesn't default to. Here are the ten most useful, each with a before and after.
1. ultrathink
Adding "ultrathink" to any prompt triggers Claude's extended reasoning mode. Instead of producing the first plausible answer, it reasons through the problem more carefully before responding. Use this for decisions, analysis, and anything where you want Claude to show its work rather than just produce a quick output.
Before: "What's the best pricing strategy for a new SaaS product?"
After: "ultrathink — what's the best pricing strategy for a new SaaS product targeting mid-market HR teams?"
The difference in depth and nuance is immediately obvious.
2. skeptically
Claude's default is to be agreeable — show it a plan and it finds reasons the plan is good. "Skeptically" flips that: it looks for what's weak, missing, or wrong before it says anything nice. This is the single fastest fix for AI that just validates whatever you already decided.
Before: "Review my business plan."
After: "Review my business plan skeptically."
If the decision really matters, go further — the anti-sycophancy guide has a full prompt for it.
3. mimic
Claude's default writing voice belongs to nobody. Describing your voice ("professional but friendly") barely helps. What works is one word and a sample: tell Claude to mimic, then paste something you actually wrote.
Before: "Write a LinkedIn post announcing our new service."
After: "Mimic the voice in the sample below, then write a LinkedIn post announcing our new service. Sample: [paste two of your old posts]"
The sample does the work. The word tells Claude the sample is the assignment, not decoration.
4. concretely
Ask a broad question and you get advice-shaped filler: "communicate clearly, set expectations, follow up regularly." "Concretely" bans it. Claude has to name specific actions — things you could actually start on Monday.
Before: "How can I improve my client onboarding?"
After: "Concretely, how can I improve my client onboarding?"
If an answer still comes back vague, reply with one word: "Concretely?"
5. tersely
AI output is padded by default — warm-ups, transitions, summaries of what it's about to say. "Tersely" changes the register entirely: short sentences, no padding, nothing repeated. It's not rude. It's how someone writes when they respect your time.
Before: "Summarize this report."
After: "Summarize this report tersely."
Word limits ("under 100 words") control length. "Tersely" controls tone. Use both.
6. plainly
Jargon is Claude's comfort zone. "Plainly" forces the explanation a smart person outside the field actually needs — without the condescension of "explain like I'm five," which produces answers you can't use in a meeting.
Before: "Explain what this contract clause means."
After: "Explain plainly what this contract clause means."
Plainly keeps it adult. It removes the jargon, not the substance.
7. verbatim
When you ask what a document says, Claude paraphrases — and paraphrase is where errors hide. "Verbatim" means word-for-word. It pins Claude to the actual text, so you can check the answer against the source in seconds.
Before: "What does the contract say about cancellation?"
After: "Quote verbatim what the contract says about cancellation."
Use this any time the exact wording carries the meaning: contracts, policies, prices, deadlines.
8. quantify
Ask whether something is a good idea and you get adjectives: "could significantly impact margins." "Quantify" demands numbers — estimates with the arithmetic shown and the assumptions stated, which you can then check.
Before: "Is offering a 20% launch discount a good idea?"
After: "Quantify the impact of a 20% launch discount on our margins. Show your assumptions."
Estimates can still be wrong. But a number with visible assumptions can be argued with — "significantly" can't.
9. steelman
A strawman is the weakest version of an argument. A steelman is the strongest. When you're leaning toward a decision, don't ask Claude what it thinks — ask it to steelman the option you're about to reject. You learn more from the best case against you than from another voice agreeing.
Before: "I'm thinking of dropping our cheapest plan. Thoughts?"
After: "Steelman the case for keeping our cheapest plan."
If the steelman doesn't move you, decide with confidence. If it does, it just saved you.
10. audit
The habit that catches mistakes before they cost you: after any answer containing facts, figures, or claims, send one word. "Audit" makes Claude re-examine its own response — flagging what it's least sure of and what you should verify before acting.
Before: (accepting the answer as-is)
After: "Audit your last answer."
This pairs with the guide on stopping AI from making things up — audit is the quick version, that guide is the full system.
Use them in pairs
One word changes one behavior, and they stack: "skeptically + concretely" for reviewing plans, "tersely + plainly" for anything you'll forward to someone else, "quantify + audit" for decisions with money in them.
The most important rule stays the same: specificity beats length. These ten words are specificity — one word each.
For teams
Want us to build this for your whole team?
We run AI workshops and full implementations for teams of 5–200. Custom prompts, workflows, and training built around your actual work.