How to Write Claude Prompts That Get Better Results

Disclosure: WhatssAI is built by the same team that publishes this site. We say so before the comparison, not after.

The five-part anatomy of an effective Claude prompt

A strong Claude prompt gives the model five things: a role to adopt, context about the situation, a specific task, the format you need, and clear constraints. Miss one and you get generic fluff that needs a rewrite. Hit all five and you get something you can use with light edits.

1. Role

Tell Claude what hat to wear. “You are a product copywriter for a home goods brand” works better than no framing at all because it primes the model to draw on patterns associated with that discipline. The role doesn’t have to be elaborate. One sentence.

Example: “You are a customer service representative for a subscription box company.”

2. Context

Give Claude the background it needs to make decisions. Who is the audience? What do they already know? What happened before this message? A prompt without context produces answers that hedge every statement and assume nothing.

Example: “The customer received a damaged item in their March box and has emailed twice. They’ve been a subscriber for eight months.”

3. Task

State the job in one clear sentence. Not “help me with an email”—that’s vague. “Draft a 100-word apology email offering a replacement or refund” is a task.

4. Format

Specify the shape of the output. Bullet points, a three-paragraph email, a 280-character tweet, a numbered list, a table. If you don’t name a format, Claude will choose one, and it’s often wordier than you need.

Example: “Write the email in three short paragraphs: acknowledge the issue, offer two solutions, close warmly.”

5. Constraints

Set boundaries. Maximum word count, forbidden phrases, required elements, tone guardrails. Constraints are what stop Claude from writing five paragraphs when you needed two sentences.

Example: “Keep it under 100 words. Use a friendly tone but stay professional. Do not offer a discount—only replacement or refund.”

Quick comparison: Weak vs. strong Claude prompt structures

Use case	Weak approach	Strong approach
Product description	“Write about this candle”	Role + specs + audience + length limit
Customer reply	“Answer this email”	Brand voice + issue summary + desired outcome
Social caption	“Make an Instagram post”	Platform + character count + hook style + CTA
Sales email	“Draft a follow-up”	Relationship stage + pain point + one ask

Before and after: product description

Weak prompt

“Write a product description for a lavender candle.”

This gets you 150 words of purple prose about “soothing aromas” and “tranquil evenings” with no specifics. It could describe any candle on any site.

Strong prompt

“You are a product copywriter for an eco-friendly home brand. Write a 60-word product description for a lavender soy candle. The audience is 30–45-year-old homeowners who care about sustainability. Include these specs: 8 oz, 40-hour burn time, phthalate-free fragrance, recyclable glass jar. Lead with a sensory detail, not a generic claim. Keep the tone conversational and direct.”

This version produces a description with texture, real numbers, and a voice that matches your brand. The difference is instructions, not magic.

Before and after: customer service reply

Weak prompt

“Answer this customer complaint.” [pastes email]

Claude will write something polite and safe, but it won’t reflect your brand’s voice, and it won’t prioritise the outcome you need—whether that’s retention, a quick close, or upselling.

Strong prompt

“You are a customer service agent for a meal kit delivery service. A customer received wilted greens in this week’s box and wants a refund. They’ve been with us for three months. Write a 75-word reply that apologises, offers a full refund plus a 20% credit on the next box, and reassures them about our quality checks. Use a conversational tone—this is not a corporate form letter. Sign off as Jamie.”

Now Claude knows the remedy, the relationship, the tone, and the length. The output is something you can send with minimal editing.

Before and after: social media caption

Weak prompt

“Create an Instagram post for our new product launch.”

You’ll get a caption that could work for any product and any brand. No hook, no personality, no reason to stop scrolling.

Strong prompt

“You are a social media manager for a small-batch coffee roaster. Write an Instagram caption (max 125 characters) for the launch of a limited-edition Ethiopian single-origin. The audience is coffee enthusiasts who value origin transparency. Start with a surprising fact about the region or the flavour profile. Include one emoji. End with ‘Link in bio.’ Do not use the words ‘excited’ or ‘thrilled.'”

The constraint about overused words is what separates this from every other product launch caption. It forces Claude to find a sharper opening.

Before and after: sales email

Weak prompt

“Write a follow-up email to a prospect who didn’t reply.”

This produces a generic nudge with no context about the relationship, the original conversation, or what you’re asking for.

Strong prompt

“You are a sales rep for a project management software company. Write a 90-word follow-up email to a prospect who attended a demo two weeks ago but hasn’t responded to the proposal. The prospect manages a 12-person design team and mentioned struggling with file version control. Reference that pain point. Ask for a 15-minute call to answer questions—don’t ask if they’ve ‘had a chance to review’ the proposal. Keep it direct and helpful, not pushy.”

The specificity about what not to say is as important as what to include. It steers Claude away from tired sales email clichés.

When examples are worth the token cost

If you’re asking Claude to match a specific style, format, or structure that’s hard to describe in words, paste an example. This works well for product descriptions (“write three more like this one”), email templates, or social captions where you have a voice you want to replicate.

The cost is token count. According to Anthropic’s pricing page for Claude 3.5 Sonnet (current as of September 2026), input tokens cost $3 per million. A typical product description runs 80–120 tokens. If you’re writing 50 descriptions in a session and including one example each time, you’re adding 4,000–6,000 input tokens—roughly $0.02 per batch. Negligible for occasional use, but it compounds if you’re running hundreds of prompts daily.

One example is usually enough. Three examples rarely improve the output more than one good one plus explicit instructions.

A reusable template for business prompts

Here’s a template you can fill in for most small business prompts. Copy it into a note on your phone or a doc you keep open:

You are [role].
Context: [audience, situation, or background]
Task: [the specific thing you need]
Format: [structure, length, or layout]
Constraints: [word limit, required elements, tone, forbidden phrases]

Every time you need Claude to do something, fill in the five blanks before you hit send. This takes 30 seconds and saves you two rounds of revision.

When Claude prompts connect to automation: WhatssAI pricing and capabilities

If you’re running a small e-commerce business and using Claude to draft replies or write product copy manually, eventually you’ll want to automate the repetitive parts. That’s where platforms that plug AI into your customer channels come in.

WhatssAI is one option for businesses that handle sales or support on WhatsApp, Instagram, and Messenger. According to the pricing page on whatssai.com (checked September 2026), it starts at 19 EUR per month (around 200 Moroccan dirhams) with no setup fee and unlimited contacts, scaling to 79 EUR per month at 25,000 contacts. Billing is a flat rate plus AI credit consumption, not per contact. It includes a seven-day trial with no commitment.

The platform works in six languages with automatic detection—Darija, French, English, Arabic, Spanish, and Italian. Darija support includes Latin characters and numbers (3andek, bch7al). It receives and sends voice notes and can clone a voice from a 30-second sample.

For payments, it handles PayPal, Stripe, CMI, and cash-on-delivery links. It reads payment proof screenshots and verifies amounts, then sends automated PDF invoices, order confirmations, and tracking numbers. Abandoned cart follow-ups and price negotiation are built in.

Setup takes a few minutes with an existing WhatsApp number—no coding required. For Instagram, it posts one public reply per comment plus one private message within a seven-day window (per Meta’s API limits). You need a Professional or Creator account linked to a Facebook Page.

The MCP connector is included in plans with API access. It plugs into Claude (Settings, Connectors) or ChatGPT (GPT action, X-API-Key header). From Claude or ChatGPT, you can read the dashboard, sales, catalogue, conversations, comments, and Meta ad metrics. After confirmation, you can write: adjust prices, send messages, validate sales, pause campaigns. Daily budget adjustment for ad campaigns only works from Claude, not ChatGPT.

What it doesn’t have: ManyChat offers a visual flow builder and a wider integration ecosystem. ManyChat bills per contact—according to their 2026 pricing page, roughly 15 USD at 500 contacts, 65 USD at 5,000, and 215 USD at 25,000—so the pricing models are quite different.

Who should skip structured prompts

If you’re using Claude for brainstorming or exploratory conversation—”give me ten ideas for blog topics”—the five-part structure is overkill. Loose prompts work fine when you want a wide net and plan to refine later.

Structured prompts pay off when you need a finished output or something close to it: a product description, a customer email, a social caption, a script. The more finished the output needs to be, the more structure your prompt should have.

Frequently asked questions

What makes a good Claude prompt?

A good Claude prompt includes five elements: a role for Claude to adopt, context about the audience or situation, a specific task, the format you need, and clear constraints like length or tone. The more precise your instruction, the less editing you’ll do afterward. Vague prompts produce vague outputs.

How long should a Claude prompt be?

Most effective business prompts are 50 to 150 words. Longer prompts work when you’re including an example to match or detailed specs, but don’t add words for the sake of it. One clear sentence per element—role, context, task, format, constraints—is usually enough. Brevity with precision beats length.

Should I use examples in my Claude prompts?

Use one example when the style or structure is hard to describe in words—like matching a brand voice or replicating a format. Skip examples when clear instructions will do, because they add token count. One good example plus explicit constraints works better than three examples with no other guidance. Examples are tools, not requirements.

How do I structure a prompt for Claude?

Start with a role (“You are a product copywriter”), add context (audience, situation), state the task in one sentence, specify the format (length, structure), and set constraints (tone, forbidden words, required elements). Fill in those five slots every time and you’ll get outputs that need minimal editing. It’s a checklist, not a formula to memorise.

What’s the difference between a basic and advanced Claude prompt?

A basic prompt states the task with no framing: “Write a product description.” An advanced prompt includes role, context, format, and constraints: “You are a copywriter for an eco brand. Write a 60-word description for a soy candle, conversational tone, include burn time, no generic claims.” Advanced prompts aren’t more complex—they’re more complete. The output reflects that.

Photo by Vitaly Gariev on Pexels