You can tell when a shop's social media is written by AI on autopilot: "Indulge in our delightful culinary offerings! ✨" Nobody talks like that, least of all the owner who's been hand-pulling noodles since 6am. The irony is that AI is genuinely excellent at marketing content — if you spend thirty minutes teaching it to sound like you instead of like everyone.
Here's the workflow we set up for clients. It runs on any of the major AI tools.
Step 1: Build a voice file (30 minutes, once)
Open a document. Put three things in it:
- Three real posts or messages you've written that felt like you. Not your best marketing — your most you. The WhatsApp message you sent regulars about the new menu counts.
- Five phrases you would never say. "Delightful", "indulge", "elevate your experience" — whatever makes you cringe. This list does more work than anything else in the file.
- Two lines about who your customers are and how you talk to them. "Mostly regulars from the nearby offices. We joke with them. No hard selling — they hate it."
Every time you ask AI for content, this file goes in first. That's the whole trick. Without it you get the average of the internet; with it you get you, faster.
Want a head start? Build your voice file with our free tool → It runs in your browser — nothing you type is stored or sent anywhere.
Step 2: Batch a month in one sitting (45 minutes, monthly)
Don't generate posts one at a time at 11pm — that's how autopilot content happens. Once a month, list your raw material: new items or services, one behind-the-scenes moment, two questions customers actually asked this month, any promo or holiday. Then ask for the month in one go:
"Here's my voice file and this month's material. Draft 10 short posts. Vary the type: some announcements, some behind-the-scenes, one that answers a customer question. Plain language. No hashtag walls, no emojis unless I use them."
Review all ten at once. Kill two, fix the rest. You now have a month of content and your evenings back.
Step 3: The 20% edit
Before anything goes out, add one concrete detail only you could know — the supplier who was late, the regular who named the new dish, the rain this morning. AI produces the 80% that's structure and clarity; the 20% you add is the part customers actually connect with. If you skip this step, skip the whole workflow — generic content quietly tells customers nobody's home.
Step 4: Photos stay real
AI writes captions. It does not generate photos of food you serve or work you've done. A real, slightly imperfect phone photo of today's actual laksa outsells a flawless synthetic one — and the day a customer realises a photo was fake, you've spent trust you can't buy back. Same rule, more strongly, for reviews: never generate them, never "polish" them. It's the fastest way to torch a small business's reputation, and in many places it's illegal.
Bonus for Singapore: multiply your languages
Draft in English, then ask for Chinese or Malay versions in the same register — this is one of AI's most underused tricks for local shops. One caveat: have a native speaker sanity-check the first few. Translation is where "almost right" is most visible, and aunties notice.
The pattern in all of this: AI supplies volume and consistency, you supply the voice and the truth. Keep those jobs straight and your content gets better and cheaper. Swap them, and you'll sound like every other shop with a subscription.