Most Indian marketers who start using Claude for copy get the same result: technically correct text that sounds like it came from a corporate brochure nobody asked for. The copy is grammatically fine, the structure is logical, and it's completely devoid of the personality that makes customers stop scrolling. The model isn't broken — the prompts are.
Claude responds differently than ChatGPT in a few important ways that affect how you should prompt it. It has a notably longer context window, which means you can feed it substantial brand documentation before asking for output. It tends to be more literal about instructions, meaning vague prompts get vague results. And its training with Constitutional AI means it sometimes defaults to careful, balanced language — which is fine for analysis but terrible for conversion copy that needs an opinion.
Here's how to work with these characteristics rather than against them.
Setting Tone with Role Assignment
The single highest-leverage prompt technique for marketing copy is giving Claude a specific professional identity before asking for anything. "You are a marketing expert" produces generic advice. "You are a senior copywriter who has spent 8 years writing direct response ads for mid-market Indian retail brands, specifically in fashion and home goods" produces something with actual direction.
The specificity of the role assignment calibrates Claude's output register. Compare the difference:
Weak role: "Write an Instagram caption for a saree brand."
Strong role: "You are a copywriter for a premium handloom brand based in Kozhikode. Your audience is women aged 28-45 in Kerala and the Indian diaspora in the Gulf. They are buying sarees for weddings, Onam, and gifting — emotional purchases where authenticity and craft story matter more than price. Write an Instagram caption for a new Kasavu saree."
The second prompt gives Claude the audience psychology, the purchase occasion, the emotional lever, and the product detail. The output reflects all of that.
Injecting Brand Voice with Sample Sentences
Telling Claude your brand is "warm, professional, and approachable" is almost useless — those adjectives describe half the brands in existence. What actually works is giving Claude three sentences your brand has already written and asking it to match that pattern.
Here is a template you can copy directly:
"Before writing anything, study these three sentences from our existing brand communications and identify the voice patterns: [Sentence 1]. [Sentence 2]. [Sentence 3]. Now write [your request] in exactly that voice — same sentence rhythm, same level of formality, same way of addressing the reader."
For a Kerala-based B2B software company, your three sample sentences might be:
- "We build software that doesn't require a manual."
- "Most businesses grow before their systems do — we fix that."
- "No demos until you tell us what's actually broken."
From those three sentences, Claude can reliably infer: short declarative sentences, slightly provocative positioning, buyer-first empathy, and avoidance of tech jargon. Every piece of copy it produces after that will carry those markers.
The RACE Prompt Framework
For marketing copy specifically, the RACE framework organises your prompt into four components that consistently produce better results than unstructured requests.
Role: Who Claude is writing as (the professional identity).
Action: What you need produced and in what format (a 3-sentence WhatsApp broadcast, a 150-word product description, a subject line + preview text pair).
Context: The audience, the occasion, the product, the constraint (character limits, mandatory phrases, tone restrictions).
Example: One sample of output you'd consider acceptable — or one example of what you explicitly don't want.
A complete RACE prompt for a Kerala food brand's Onam campaign:
"Role: You are a copywriter for a premium Kerala snacks brand targeting NRKs in the Gulf and UK who want to send Onam hampers to family back home. Action: Write 5 WhatsApp broadcast message options (under 160 characters each) for an Onam hamper sale. Context: The hampers contain traditional Kerala items — banana chips, achappam, murukku, coconut-based sweets. Price starts at ₹1,499 for delivery within Kerala. Audience is emotionally driven — nostalgia and family connection are the key motivators, not price. Example of tone I want: 'This Onam, send home what they actually miss.' Example of tone I don't want: 'Get amazing deals on Onam hampers now!'"
Writing for Indian Audiences Without American Idioms
Claude's training data skews heavily toward American English, which creates a specific problem for Indian marketers: left unprompted, it will produce copy that reads like it was written in San Francisco for a SaaS startup. The idioms are wrong, the cultural references miss, and phrases like "reach out" and "circle back" land awkwardly in contexts where your audience is communicating in Indian English.
Two instructions that solve this consistently:
First: "Write in Indian English, not American English. Use 'revert' not 'get back to me', 'prepone' where appropriate, and avoid Americanisms like 'awesome', 'touchdown', 'ballpark'. The reader is a business owner in Tier 2 India who communicates primarily in WhatsApp and email."
Second: Explicitly name idioms to avoid. "Do not use the phrases 'at the end of the day', 'game-changer', 'leverage', 'synergy', or 'value-add'. These read as corporate filler to this audience."
For businesses serving Kerala-specific audiences, you can go further: "The reader understands references to Onam, Vishu, KSRTC, NEET coaching, Gulf remittances, and chit funds. You can reference these when contextually appropriate."
Multi-Shot Prompting with Before/After Examples
Multi-shot prompting means showing Claude multiple examples of input-output pairs before asking for the real output. For marketing copy, this is particularly powerful for establishing format constraints that are hard to describe in words.
Suppose you need product titles for an Amazon India listing. Rather than describing character limits and keyword placement rules abstractly, show Claude what you mean:
"I need you to write Amazon India product titles. Study these three examples of titles I've approved, then write titles for the products I give you in exactly the same format.
Approved title 1: 'Kerala Banana Chips | Traditional Coconut Oil Fried | 500g Pack | No Artificial Preservatives'
Approved title 2: 'Handloom Kasavu Saree | Pure Cotton | Kerala Onam Special | Cream & Gold Border'
Approved title 3: 'Coconut Shell Craft | Handmade Kerala | Decorative Bowl | 6-Inch Diameter'
Now write titles for: [your products]"
Claude will replicate the pattern — pipe-separated attributes, specific measurements, occasion tags, origin markers — without needing explicit formatting rules.
System Prompt Templates for Different Copy Types
When using Claude's API or the Projects feature in Claude.ai, you can set a system prompt that applies to every conversation in that project. Here are two templates worth keeping.
For WhatsApp marketing copy:
"You are a WhatsApp copywriter for Indian retail and service businesses. All output must: be under 160 characters unless specified otherwise; use a conversational, direct tone; include one clear action (call, visit, reply); avoid emojis unless explicitly requested; never use corporate jargon. When given a product, offer, or event, produce 3 variations — urgent, benefit-led, and curiosity-led."
For long-form blog introductions:
"You are a content writer for a B2B IT consulting firm serving Indian SMBs. Blog introductions must: open with a specific real-world scenario (not a statistic or question); establish the problem in the second sentence; promise a practical outcome (not 'in this article we will explore'); be 80-120 words; use Indian business context — mention Tally, GST, Jio, UPI, or other India-specific touchpoints where relevant. Never start with 'In today's digital world' or any variation of that sentence."
Iteration Prompts for Refinement
First drafts from Claude are rarely final drafts. The difference between users who get great output and users who settle for mediocre output is usually their iteration discipline. Specific refinement prompts work far better than vague requests like "make it better."
Useful iteration prompts after a first draft:
- "The opening is too generic. Rewrite just the first two sentences to start with a specific scenario a Kerala retailer would recognise immediately."
- "This reads too formally. Cut the sentence length by 30% and remove any word over three syllables that has a simpler alternative."
- "The CTA is weak. Give me 5 alternative closing sentences that create urgency without using the word 'limited' or 'hurry'."
- "Version B: rewrite this for someone who has already tried a competitor product and is evaluating switching. What would change in the argument?"
What Claude Does Best — and Where It Needs You
Claude genuinely excels at long-form content where logical structure and clarity matter: technical explainers, detailed how-to guides, email sequences with multiple paragraphs, and any copy that needs to handle nuance without losing the reader. Its ability to hold a long brief in context without drifting is a meaningful advantage over shorter-context models.
Where Claude reliably needs human editing: very short copy (taglines, slogans, 5-word punchlines) where compression and unexpected word choices matter; copy that requires cultural insider knowledge about a very specific regional context; and anything where the goal is emotional shock or provocation. Claude's Constitutional AI training makes it cautious about edgy or provocative language — it will often soften what should be sharp.
The practical workflow for Indian marketers: use Claude to produce the structure, argument, and first draft; then manually sharpen the opening line, the CTA, and any line that needs to land like a punch. You're not replacing your copywriting judgment — you're removing the blank-page problem and the structural scaffolding work, which together consume most of the time.
Prompt engineering is a skill that compounds. The prompts you refine this month become templates your whole team uses next month. Start with the role assignment and brand voice injection techniques — those two changes alone will produce materially better first drafts than unstructured requests, and everything else builds from there.