Writing Bot Scripts in English and Roman-Urdu That Convert

Conversion copy helps the right customer make an informed next decision. It does not pressure, disguise conditions, or promise an outcome the system cannot verify. English and Roman Urdu versions must carry the same meaning, price, consent, and escape routes.

After this lesson, you can create a bilingual script sheet, control risky claims, and test comprehension instead of guessing that casual language always performs better.

Write the Functional Version First

For each message record:

message_id
state
purpose
English
Roman Urdu
allowed replies
data saved
risk owner

Use one action per message. Lead with the required fact, then the choice. Avoid decorative greetings on every turn.

Weak:

Amazing news! Your dream result is just one click away!!!

Controlled:

English: Choose your city so we can check service coverage.
Roman Urdu: Apna shehar select karein taa-ke hum service coverage check kar saken.
Choices: Karachi / Lahore / Islamabad-Rawalpindi / Other

The Roman Urdu is a practical rendering, not a universal translation. Test vocabulary with the target audience and retain English for brand names, reference codes, or technical terms when that is clearer.

Build a Message Pattern

Use four parts when needed:

1. Context: identify the business or current task.
2. Fact: state what is known and what is not.
3. Action: ask for one choice.
4. Escape: offer back, stop, or human help.

For money, repeat the unit and conditions:

Item subtotal: PKR 4,500
Delivery: PKR 250
Total due: PKR 4,750
Payment method: Cash on delivery
Reply CONFIRM to place the order, EDIT to change it, or HUMAN for help.

Do not use countdowns, scarcity, savings, guarantees, or “best” claims unless the business has current evidence and approval. Never let a language model invent a discount, policy, delivery date, or product feature.

Separate Service and Marketing Consent

An order update is not permission for future offers. If asking for marketing opt-in, describe the category and frequency, make refusal easy, and record the answer. The script must honor STOP and equivalent plain-language requests, not only a hidden menu choice.

Worked Example

A Multan training center needs to qualify course enquiries.

EN: Which class are you asking about? Choose a course or tap HUMAN.
RU: Aap kis class ke baare mein pooch rahe hain? Course select karein ya HUMAN dabayein.

EN: The next listed batch starts 12 August, subject to seat confirmation.
Fee: PKR 18,000. Registration does not mean payment is received.
RU: Agla listed batch 12 August ko start hota hai, seat confirmation ke baad.
Fee: PKR 18,000. Registration ka matlab payment receive hona nahin hai.

The team tests the script with five target users. Two do not understand “batch,” so the Roman Urdu becomes class group. Another mistakes registration for admission, so the confirmation explicitly says that staff will verify seat and payment. The revision comes from observed comprehension, not style preference.

Failure Cases to Diagnose

• Roman Urdu is a word-for-word machine translation: test meaning and naturalness with users.
• Different prices in two languages: use one authoritative variable for both.
• Bot says “confirmed” before inventory/payment check: name the actual pending state.
• Buttons contain vague labels: use actions such as CHECK PRICE, EDIT, or HUMAN.
• Marketing consent bundled into checkout: separate purpose and choice.
• STOP works only in uppercase: normalize common opt-out phrases and route uncertain cases to suppression review.
• AI rewrites regulated or contractual text freely: lock approved text and require owner review.

🇵🇰 Pakistan Angle

Roman Urdu spelling varies widely. Do not create a giant synonym list from assumptions; review real anonymized inbound phrases and add tested variants. Offer Urdu script only if the business can maintain it accurately. Always preserve unambiguous dates, PKR amounts, order IDs, phone numbers, and addresses.

Respect formality. Aap is safer than overly casual phrasing for an unknown customer. Avoid gender assumptions and honorifics inferred from names. For address collection, show the captured area and landmark back to the customer, but store only what fulfillment genuinely needs.

Hands-On Exercise

1. Select six messages from one flow.
2. Fill a bilingual script sheet with purpose, choices, saved data, and risk owner.
3. Replace vague labels and unsupported promises.
4. Test both versions with at least three intended users.
5. Record misunderstandings and revise the shared meaning.

Done means: users understand the same offer, amount, state, consent choice, and human route in both versions.

Completion Rubric

• Each message has one purpose and action.
• English and Roman Urdu preserve the same commercial meaning.
• Prices and conditions come from shared variables.
• Consent is explicit and opt-out language is recognized.
• Risky claims and contractual text require approval.
• Revisions are based on comprehension tests.

Sources

• WhatsApp Business — Messaging policy
• WATI Help — Chatbot collection
• Google — Developer documentation style guide
• Microsoft — Writing style guide

Key takeaway: write bilingual bot copy as controlled interface text—same facts, choices, consent, and safety in both languages—then improve it with real comprehension tests.