AI is best used at the top of the social workflow: ideation, caption variations, and creative drafts. Keep a human on final voice and on anything factual — audiences detect generic AI captions quickly, and it costs trust.
A realistic AI social workflow
- Ideation — ChatGPT for content pillars and hook variations from a single brief.Free tier
- Creative — Canva AI for post and carousel design; Ideogram where text-in-image is needed.Free tierTaught in our course
- Video — AI video tools for Reels assets, then human editing for pacing and hooks.Taught in our course
- Captions — AI first draft, human rewrite for brand voice — never ship raw.Taught in our course
Briefing AI so the output is usable
Most disappointing AI output comes from a one-line prompt. "Write an Instagram caption for a gym" produces the same caption every gym already posts. The fix is to brief the tool the way you would brief a junior writer.
Keep a short brand brief saved and paste it at the top of every session. It takes ten minutes to write once and improves everything that follows.
- Audience — Who the post is for, in one sentence — age, city, problem.
- Offer — What the business actually sells and the one thing this post should push.
- Voice — Two or three past posts that performed well, pasted in as examples.
- Constraints — Language (English, Hindi or Hinglish), length, words to avoid, no invented offers.
- Format — Ask for ten hooks or five caption variations, not one finished answer.
What AI does well, task by task
| Task | What AI does well | What a human must check |
|---|---|---|
| Content pillars | Turning a business description into themes and post ideas | Whether the ideas fit what customers actually ask about |
| Hooks | Generating many opening lines quickly | Picking the one that is specific rather than clickbait |
| Captions | First drafts and length variations | Brand voice, facts, prices and offers |
| Carousel scripts | Breaking a topic into slide-sized points | Order, accuracy and the final call to action |
| Reel scripts | Structuring a 30-second script around a hook | Whether it sounds natural when spoken aloud |
| Comment replies | Suggesting polite replies to common questions | Every reply before it is posted — never automate this |
| Monthly analysis | Summarising exported insights and spotting patterns | That the numbers it quotes match the export |
Where AI goes wrong on social
- Invented details — AI will happily add a discount, a timing or a claim the business never made. Check every fact.
- Same structure every post — Left alone, it repeats one caption template. Audiences notice the pattern within a few weeks.
- Weak local context — Festival dates, local references and natural Hinglish often come out slightly wrong.
- Image artefacts — Hands, text and product details in AI images need a close look before posting.
- Passing AI images off as real — Never present a generated image as a photo of the actual product, shop or customer.
An example week for a local business
On Monday, paste the brand brief into ChatGPT and ask for twenty post ideas across three pillars; pick five. Ask for ten hooks for each and choose the best by hand. Build the static posts and carousels in Canva, using Magic Write only for drafts. Script one Reel, record it on a phone, and edit it in CapCut with auto-captions. Rewrite every caption in the brand voice, schedule the week, and at month end paste the insights export back into ChatGPT to summarise what worked — then verify its summary against the numbers.