Vernacular Ad Copy

---
name: vernacular-ad-copy
description: Adapt Meta ad copy into Indian languages — Hindi, Tamil, Hinglish and other code-mixed variants — without losing the angle, the offer, or compliance. Use when asked to translate ad copy, write Hindi or Tamil ads, make Hinglish versions, run ads in regional languages, reach tier-2/3 audiences, localise a campaign for India, or someone says 'English ads aren't working outside the metros'. For the English source copy, see headline-and-primary-text. For the argument itself, see ten-angles-copy.
metadata:
  version: 1.0.0
---

# Vernacular ad copy

Translation is the wrong frame. An ad's angle, hook and offer get **re-expressed**
in the language the audience actually thinks in — which in much of India is
code-mixed, not textbook Hindi or Tamil. A literally translated hook keeps the
grammar and loses the scroll-stop.

Budget note from the corpus: **a Hindi or Tamil caption usually costs more tokens
than the same caption in English** — plan generation volume accordingly.

## Before you start

| Input | Why |
| --- | --- |
| The English copy set, angle named | You adapt an argument, not sentences |
| Target language AND register | "Hindi" is not one decision — pure, Hinglish, or Roman-script Hinglish read as different people |
| Who checks the copy | A native-speaker review is non-negotiable before spend |
| The landing page language | A Hindi ad landing on an English-only page breaks the promise |

If no native speaker is available to review, stop. Machine-only vernacular copy
carries real embarrassment risk at feed scale.

## Step 1 — Choose script and register

| Audience | Register that usually fits |
| --- | --- |
| Metro, young, English-comfortable | Hinglish in Roman script ("Job nahi mil rahi?") |
| Tier-2/3, Hindi-first | Devanagari, conversational — not शुद्ध textbook Hindi |
| Tamil Nadu and similar strong-language states | The state language in its own script; code-mixing varies by state |

This is a positioning decision, not a translation setting: register signals who the
ad is for before a word is read. When unsure, test registers as variants — they are
genuinely different creatives, not the same ad twice.

## Step 2 — Re-express, do not translate

Work angle-first:

1. Take the angle and the persona's pain — not the English sentences.
2. Ask: how does this person *say* this pain to a friend? That phrasing is the hook.
3. Keep untranslatable anchors in English: brand names, "EMI", "placement", tech
   vocabulary the audience uses in English anyway.
4. Numbers, ₹, and offers stay identical to the English set — the offer does not
   change with the language.

Idioms and wordplay from the English hook are rebuilt, never carried: a pun that
dies in translation takes the hook with it.

## Step 3 — Re-check what language changes

- **Length**: scripts run wider; re-check overlay text fits and stays legible at
  phone size (the 2–7 word overlay budget still applies, per-language).
- **Compliance**: the personal-attributes rules apply in every language — "कर्ज़ में
  डूबे हैं?" implies financial status exactly like its English twin. Scrub with
  **copy-compliance-scrub** per language.
- **Landing page**: match, or use the closest matching page, or expect the bounce.

## Step 4 — Native review before spend

The reviewer checks: register (does it sound like a person or a form?), unintended
meanings, region-wrong vocabulary, and script rendering on device. One wrong-register
word reads as an outsider trying — the opposite of the trust the vernacular ad
exists to build.

## Output

```
# Vernacular set: <campaign> × <language/register>

| Field | English source | Adapted | Back-translation |
| overlay | "..." | "..." | "..." |
| headline | "..." | "..." | "..." |
| primary | "..." | "..." | "..." |

Register: <e.g. Roman-script Hinglish>  Anchors kept in English: <list>
Compliance: scrubbed per language — <clean / findings>
Landing page: <matched language? which URL>
Native review: <name/role, date>  ← blank means DO NOT LAUNCH
```

## When it breaks

| What you see | What it means | The fix |
| --- | --- | --- |
| Vernacular ad underperforms English | Literal translation — grammar kept, hook lost | Re-express from the angle, in how they'd say it to a friend |
| Comments mock the copy | Wrong register or region vocabulary | Native review before spend, not after |
| Great CTR, page bounces | Language promise broken at the page | Match the landing page language |
| Overlay text cramped or clipped | Script runs wider than the English layout | Re-fit per language; the word budget is per-language |
| Rejected in one language, fine in another | Compliance scrubbed only on the English | Scrub every language version separately |
| Hinglish ad flops in Chennai | Register imported across states | Register is per-state; Tamil Nadu is not a Hinglish market |

Never ship machine translation unreviewed to save time. The failure mode is not
mediocrity — it is becoming the screenshot in the group chat.

## Rules

- **Native review before spend, always**, because wrong-register copy at feed scale
  is a public embarrassment, not a private draft.
- **Re-express the angle, never translate the sentences**, because the hook lives
  in how the audience actually talks.
- **The offer and its numbers stay identical across languages**, because a language
  variant is not a pricing variant.
- **Compliance is per-language**, because an implied attribute violates in Hindi
  exactly as in English.
- **Register is a targeting decision** — declare it per audience, because Hinglish
  and Devanagari Hindi reach different people with the same words.

## Related skills

- **headline-and-primary-text** — the English source set and its budgets.
- **ten-angles-copy** — the angle being re-expressed.
- **copy-compliance-scrub** — run per language version.
- **bulk-ad-variations** — language registers multiply the matrix; add as a
  deliberate axis, not an afterthought.