Table of Contents
You type “write a product description for our new dining table” and Claude gives you three tidy paragraphs. They’re grammatical, upbeat and completely forgettable. You could paste them under any table in any shop.
So you add “make it better.” You get the same paragraphs with a few more adjectives.
Knowing how to prompt Claude is what separates that result from a description that sounds like your workshop and sells the table. The good news is that it isn’t technical. It comes down to a handful of habits you can learn in an afternoon.
In this post, I will walk through seven of them, all drawn from Anthropic’s own prompting guidance, then show you one full prompt that uses all seven so you can see how they fit together.
Why Claude Gives Average Answers to Vague Requests
Claude isn’t being lazy when it gives you a generic answer. It’s doing exactly what you asked, with nothing else to go on.
Anthropic’s guide says current Claude models respond well to clear, explicit instructions, and that being specific about the output you want improves results. It also says that if you want “above and beyond” work, you have to ask for it, because the model won’t always infer it from a vague prompt.
Here’s a useful way to think about it. Anthropic suggests imagining Claude as a brilliant but new employee who lacks context on your norms and workflows. A new hire who’s told “write a description of the table” would come back with the same bland paragraphs. Give them the audience, the details and the tone, and they’d do something much better.
How to Prompt Claude: Seven Habits That Work
1. Pass the Colleague Test
Before you send a prompt, ask whether a smart colleague with no background on your task could follow it. Anthropic calls this the golden rule: show your prompt to a colleague with minimal context and ask them to follow it. If they’d be confused, Claude will be too.
If your prompt says “write something for the launch,” your colleague would ask a dozen questions. Answer those questions in the prompt, and Claude won’t have to guess.
2. Explain Why You’re Asking
Claude does better when it understands the purpose behind a request. The guide’s example is telling it that a response will be read aloud by a text-to-speech engine, so it should avoid ellipses, which works better than a bare rule against them.
For your own work, that might sound like “this email goes to a client who’s nervous about the delay” or “this description will sit on a product page next to the price.” One sentence of reason changes the whole approach.
3. Give It a Role and a Reader
A role focuses Claude’s tone and judgment. The documentation notes that even a single sentence makes a difference.
“You are a copywriter for a small, family-run furniture workshop” is enough. Then name the reader too, such as “writing for homeowners aged 35 to 55 who care about craftsmanship.” The role sets the voice, and the reader sets the content.
4. Show What Good Looks Like
Examples are one of the strongest tools you have. Anthropic says they’re one of the most reliable ways to steer format, tone and structure, and recommends three to five of them for best results.
Paste in one or two pieces of your own writing that you’re proud of. Then say “match this voice.” That’s far quicker than describing your style in adjectives.
Wrap each example in its own tags so Claude can tell it apart from your instructions, like this:
<example>
Our chairs are built in one afternoon and checked for a week. Every joint is cut by hand, because glue alone doesn't last.
</example>
5. Organize Long Prompts With Tags and Put Documents First
Once a prompt includes instructions, background and pasted material, it can get muddled. Anthropic recommends wrapping each type of content in its own tag, noting that this helps Claude parse complex prompts unambiguously. Tags like instructions, context and input work well.
When you’re pasting in a long document, put it first. The guide says to place longform data at the top, above your query and instructions, and reports that putting the question at the end can improve response quality by up to 30 percent in tests on complex inputs.
6. Say What to Do, Not Just What to Avoid
“Don’t use bullet points” is weaker than “write in flowing paragraphs.” Anthropic’s advice is to tell Claude what to do instead of what not to do.
There’s a related trick. The guide notes that the style of your prompt can influence the style of the response, so if you want plain paragraphs back, write your prompt in plain paragraphs rather than heavy lists and symbols.
7. Ask for the Action, Then Ask for a Check
Claude takes wording literally. The guide gives an example: asking “can you suggest some changes” tends to get suggestions, while “change this function to improve its performance” gets the changes made.
The same applies to everyday work. If you want a rewritten email, say “rewrite this email,” not “what do you think of this email?”
Then close with a check. A line like “before you finish, verify the numbers against the document I gave you” catches mistakes before you see them. The documentation recommends asking Claude to self-check, especially for coding and math, and it works well for anything with facts or figures.
A Full Prompt That Uses All Seven Habits
Here’s one prompt that combines everything above. It’s for a made-up family furniture workshop. Swap in your own product details before using the structure.
<document>
Product: the Marlow dining table. Material: solid oak. Size: 180 cm by 90 cm, seats six to eight. Finish: hard-wax oil, which can be refreshed at home. Made by hand in our workshop with a three-week lead time. Price: 1,450 dollars. Comes with a 10-year guarantee on joints and structure.
</document>
<example>
Our chairs are built in one afternoon and checked for a week. Every joint is cut by hand, because glue alone doesn't last.
</example>
<instructions>
You are a copywriter for a small, family-run furniture workshop. Write the product description for the Marlow dining table using only the facts in the document above. It will sit on our website product page next to the price, and our customers are homeowners aged 35 to 55 who care about craftsmanship and will pay more for something that lasts.
Match the plain, confident voice of the example. Write in flowing paragraphs, about 120 words in total. Lead with what makes the table worth the price, mention the guarantee once, and finish with a calm, unpushy line about the lead time.
Don't invent any details, such as tree origin, dimensions or care instructions, that aren't in the document. Avoid the words "stunning," "elegant" and "perfect."
Before you finish, check every number in your description against the document and tell me if anything doesn't match.
</instructions>
Count the habits. The document comes first, with the instructions after it. There’s a role, a reader and a reason. The example sets the voice, the tags separate everything, the formatting is described positively, and the check comes at the end.
Notice how short each piece is. A strong prompt is rarely long. It’s just complete.
Common Mistakes When You Prompt Claude
Sending a one-line request for a big task. Claude fills the gaps with averages. If the task matters, spend two minutes on the prompt.
Piling on rules in capital letters. Anthropic notes that newer models respond closely to instructions, so aggressive wording like “CRITICAL: you MUST” can overdo it. Normal, plain phrasing usually works better.
Asking Claude to write out its reasoning. On several of the newest models, the documentation warns that a prompt asking the model to write out its reasoning, for example in thinking tags, may be declined. If you want to understand an answer, ask for a short explanation of it instead.
Pasting a document and burying the question. Put the material first and the question last.
Trusting facts without checking. Claude is good, not infallible. Verify figures, names and any claim that matters before you publish or send.
Pasting sensitive information carelessly. Check your workplace policy before sharing client or financial data with any AI tool.
How to Keep Improving Your Prompts
Your first prompt is a draft. When the result isn’t right, don’t start over. Tell Claude what’s wrong in specific terms: “the second paragraph sounds like marketing, so cut anything that isn’t a fact from the document.”
When a prompt works well, save it. Keep it in a notes app with one line saying what it’s for and what to paste in. If you use Claude regularly, its settings also let you store instructions you find yourself repeating, so you don’t retype them each time.
The One Habit That Matters Most
If you only remember one thing about how to prompt Claude, make it this: give it what a smart new colleague would need. That means the reader, the reason, the material and the shape of a good answer.
Pick one task you do every week, like a client email, a product description or a meeting summary. Rewrite your usual request using three of the habits above and compare the results.
If you’d like that structure ready-made, JSON Prompt Mastery for Claude is the next step. It gives you 70+ structured prompts organized by business workflow, so you can skip the setup and get straight to work you can use.