references/caption-craft.md

# Caption Craft

Everything here serves one goal: copy that reads as though a real person typed it on a phone between two tasks.

## Contents
1. The seven caption patterns
2. The one-word swap
3. Language rules that make it sound typed
4. Dialect notes (Egyptian and Saudi/Gulf)
5. Patterns that mark copy as machine-written

---

## 1. The seven caption patterns

Pick by what the moment is doing, not by what is easiest. Over-using pattern 1 makes a whole calendar feel the same.

### Pattern 1 — The "لما" trigger
```
لما + موقف محدد + :
```
The workhorse. Puts the reader inside the situation as a condition, so their brain finishes the sentence before the image arrives.

> لما تقعد شهور تعمل كونتنت وريلز وميديا باينج وبرضو مفيش ريتش ولا مبيعات:
> لما يجيلى تعديل على ديزاين قديم:

Gulf: `لما` becomes `لين` or `يوم`, sometimes `إذا`.
> يوم يجيك تعديل على ديزاين عمره سنة:

### Pattern 2 — Two-party juxtaposition
Two lines. The first gives the general state of the world, the second names the character whose reaction you are about to see.
```
سطر 1: السياق
سطر 2: الشخصية دي تحديدًا + :
```
> قولت لمشتركين الكورس مليون مره مينفعش يشاركو الماتريال مع حد
> أسألة المشتركين بعدها بدقيقتين:

The strongest pattern available, because the irony is built into the situation rather than into the wording. Reach for it when the moment has two parties in it.

### Pattern 3 — Direct quote setup
The caption is a line someone said to the audience, quoted back.
> مش انت قولتلي هنبدأ بالساليري الصغير دا وهتزودني بعد تلات شهور:
> يعنى انت مش شايف حاجة غلط فى الديزاين؟

Works because the reader hears the manager's or the client's actual voice in their head, and the voice does the emotional work for you.

### Pattern 4 — Confession
First person, stated flatly, slightly too honest.
> والله ياجماعة الوحيد يستاهل اقول عليه البيست فريند هو اللابتوب بتاعي

### Pattern 5 — The accusation
Second person, aimed at someone, usually with an exclamation.
> داخل تشقط من لينكد ان يا وغد!!
> وكمان جونيور؟

### Pattern 6 — Labeling a sub-audience
Names a narrow group out loud. Drives tagging, because people mention each other to say "this is you".
> للـ ai designers الPentool دي حاجة فى الفوتوشوب:

### Pattern 7 — Reverse brag
A benefit disguised as a complaint or a flex the character cannot help making. This is the pattern to use when the brand needs to be visible, because the selling happens sideways.
> لما تستخدم الميم ماركتينج وكل الريتش والمبيعات بتاعتك تبقى اورجانيك

---

### The colon
Roughly two-thirds of captions end with `:`. It is a door, not punctuation. It creates the half-second pause where the reader shifts from reading to looking.

Use it for patterns 1, 2, 3 and 6. Skip it for confessions and reverse brags, where the caption is a complete thought on its own and a colon would feel like the writer is nudging the reader.

---

## 2. The one-word swap

A useful technique for a familiar, correctly licensed scene. Keep it optional: the supplied deck also contains pure reactions, original dialogue, literalization and affectionate observations.

When the audience genuinely knows a short line and you have the rights to use the source material, consider a one-word swap. An original line, a reaction without words or a staged screenshot may work better.

> Original register: a man at an ecstatic street celebration
> Swap: "انا فى اليوم اللى اتعلمت فيه الـ Pentool"

> Original register: a boss reassuring a new hire
> Swap: "متقلقـش الـ Vibes هنا حلوة والـ Environment لذيذة"

> Original register: an old craftsman talking man-to-man
> Swap: "إركـن Chat gpt على جنب وكلمنى صنايعى لـ صنايعى"

**Why one word and not three.** The rhythm of the original line is the reference. The audience's memory supplies the timing, the accent, the face. Rewrite the whole line and the memory has nothing to hold on to, so the reference evaporates and you are left with a caption on a stranger's photo.

**How to pick the word.** It is almost always the noun that names the world. Swap the object, keep the emotion: a business becomes a design platform, a car becomes a laptop, a bride becomes a client brief.

**When the scene has no quotable line**, write a fresh one in that character's voice, matching their register and their length. A tired old man does not speak in marketing vocabulary.

---

## 3. Language rules that make it sound typed

### Specific details carry recognition
> 1200 جنيه ووجبة · 16 ساعة · 2700 جنيه · 20 ريل ولوجو وايدنتتي · بعدها بدقيقتين · متأخرة ساعتين

A number or timestamp can make a scene concrete when the brief supports it or it is clearly fictional comedy. Do not fabricate a client result, a market figure or a user's lived experience to get a numeric receipt. A named tool, overheard phrase or oddly specific object may do more work than a number.

### Script follows the audience, not the dictionary
Write each borrowed word in whichever script the audience actually types it in.

Stays in Latin script (tool names, jargon, identity words):
> Pentool · Layers · Chat gpt · Ai designer · GEN Z · Vibes · Environment · Date · CRM · Brief

Written in Arabic script (already absorbed into speech):
> كونتنت · ريلز · ميديا باينج · كوتيشن · كاليندر · فيدباك · تاسكات · ريتش · اورجانيك · ديزاين · لينكد ان · بوست · سيشن

If you are unsure, picture the word in a WhatsApp group chat between two people in that job, and write it the way it appears there.

### Leave the seams showing
- No period at the end of a sentence
- Stretched letters carry tone: `تاسكات كتيييييييير` does the work of an exclamation mark
- Casual capitalization in Latin words (`Chat gpt`, `ChAT GPT`) is fine and reads as fast typing
- Minor spelling slips that everyone makes are fine; do not manufacture them, but do not hunt them down either

Over-correcting is the most common way to accidentally turn a good meme into a bad ad. Balanced, fully punctuated, grammatically immaculate copy is the tell.

### Spoken connective tissue
Egyptian memes are held together by: لما، اللى، بقالك، عشان، برضو، مفيش، بتاعي، يعنى، ده، دي، كده، أصل.

Using عندما، الذي، لذلك، ليس instead breaks the spell in one word.

### Keep it under one breath
A caption is one or two lines, rarely three. If it needs a third line, the moment probably contains two ideas and should be split into two memes.

---

## 4. Dialect notes

### Egyptian (default)
Vocabulary of the agency floor: المدير، الكلاينت، الأكونت، الكاليندر، التاسك، الريفيو، الديزاين، البريف، الريتش، الفيدباك، الجروب، الكوتيشن.
Tone: self-deprecating, fast, comfortable with complaint, fond of exaggeration (`بقالي سبعين سنة`).

### Saudi and Gulf
Swap the vocabulary **and** the cultural references together.

Words: `لما`→`لين/يوم`، `عايز`→`أبي/ودي`، `ازاي`→`كيف`، `دلوقتي`→`الحين`، `كده`→`كذا`، `مفيش`→`ما فيه`، `بتاعي`→`حقي`، `جامد`→`فخم/طاغي`، `يا جدع`→`يا بعد حيي`.
Work life: الدوام، الاجتماع، الإدارة، الموارد البشرية، الترقية، نهاية الخدمة، العميل، الوكالة، التقرير، الميزانية، الربع.
Texture: الاستراحة، القهوة المختصة، الدوام الصباحي، الزحمة، رمضان، موسم الرياض.

Tone runs slightly drier than Egyptian. Less exaggeration, more deadpan. Complaint is more indirect, so the reverse brag and the two-party juxtaposition tend to travel better than the open confession.

### English
Same engine. Receipts become salary figures in the local currency, Slack, standups, "circle back", Q4. Formats come from the global section of the format bank.

---

## 5. Patterns that mark copy as machine-written

Any of these in a draft means rewrite the line from scratch, not patch it.

**Denial then reveal.** `ده مش X ده Y` · `هذا ليس X بل Y` · `مش مجرد X` · `not just X` · `this is not X, this is Y`. The structure announces that a writer is performing. The meme already carries its contrast in the gap between caption and image; saying it out loud is doing the picture's job badly.

**Explaining the joke.** A clause whose job is to unpack the joke: عشان كده، والنتيجة، ودي المشكلة اللي، and يعني when it introduces an explanation. Conversational يعني as filler in a character's mouth is ordinary spoken Egyptian and stays, which is why it appears in the connective-tissue list above. If the meme needs the explanation, the meme is broken.

**Punchline restating the caption.** The most common failure. It comes from one mouth writing both lines. See the two mouths rule in SKILL.md.

**Forced puns and wordplay.** Reject wordplay that needs an explanation. Keep a short familiar pun when this specific audience recognizes it immediately.

**Generic nouns everywhere.** الشغل، الحياة، الناس، الأمور، التحديات with no specific object anywhere in the meme.

**Emojis, hashtags, CTAs inside the image.** All three convert a meme into an ad on sight.

**Perfect symmetry.** Two clauses of identical length balanced against each other. Real speech is lopsided.

**Corporate vocabulary smuggled into a character's mouth.** A tired employee does not say "نحقق نتائج ملموسة".

**Generic wisdom without a scene.** An evergreen household situation can work for years when it has a precise recognizable trigger.


## 6. Expanded writing grammar

The source deck favors outside setup plus inside-character response. This remains a strong default for reaction memes, but see references/humor-mechanics.md for labeling, screenshots, original scenes, multi-panel, irony, text-free reaction and affectionate recognition. A colon is optional. Do not manufacture typos or force every post into لما. For a user voice sample, preserve documented rhythm rather than borrowing the deck creator's quirks. If the caption and image intentionally repeat a phrase, check whether the second context changes its meaning; intentional echo is allowed.

Platform affects the amount of text and the reference pool. On X, test whether the copy works beside the image in the timeline. On Instagram, the image often needs to survive without the full caption. On LinkedIn, use occupational familiarity and avoid fabricated anecdotes or fake performance claims. A local reference should be checked with the actual intended cohort, not assumed familiar to every national audience.