The creative engine, rewritten: posts that stop the scroll
We tore the writing engine down to its foundations and rebuilt it. New creative mission. Emotional hooks. An explicit anti-patterns blacklist. A smarter model. The Alchemist's output just got noticeably sharper.

If you used the Alchemist last week, you might notice something different: the captions feel more like something a person would actually write. Less "corporate wellness poster," more "friend who's good at this." Here's what changed under the hood.
The creative mission
Every caption the Alchemist writes now starts from a creative mission — a set of principles baked into the engine before it sees your prompt. The mission is simple: write something a real person would stop scrolling for. Not something that sounds like marketing. Not something that ticks SEO boxes. Something with a hook, a point of view, and a reason to keep reading.
The mission includes a few concrete rules: lead with the most interesting thing, not the brand name. Use specific details, not vague claims. Keep it conversational. Never open with a question that nobody actually asks ("Ready to take your business to the next level?").
Emotional hooks, not buzzwords
The engine now tags each caption with an emotional register — curiosity, urgency, nostalgia, surprise, pride, relief. Not as a label the reader sees, but as a writing constraint. A caption tagged "curiosity" opens with something unexpected. A caption tagged "urgency" leads with a deadline or a scarcity signal. The hook isn't bolted on afterward; it shapes the entire structure.
This matters because the single biggest predictor of whether someone reads past the first line is whether that first line triggers an emotion. "Our new fall menu is here" triggers nothing. "The dish that sold out in 40 minutes last October is back — and we added a twist" triggers curiosity and urgency at the same time.
The anti-patterns list
We built an explicit blacklist of phrases and patterns the engine is not allowed to use. It's long. Here are a few entries:
- "Exciting news" or "We're thrilled to announce" — the reader decides if it's exciting; telling them it is means it probably isn't.
- "Take your [X] to the next level" — meaningless without a specific "next level."
- Starting with a rhetorical question — "Ever wondered what it'd be like to…?" Nobody wonders that.
- Generic calls to action — "Check it out!" says nothing about what "it" is or why someone should.
- Corporate hedging — "We believe that…" Just say the thing.
The engine checks its own output against this list before returning a caption. If it catches a match, it rewrites the offending section. You never see the draft that had "Exciting news!" in it.
Micro-arc structure
Longer captions (2–6 sentences) now follow a micro-arc: hook, tension, resolution. The hook is the first line — the one that has to earn the "more" tap. The tension is the middle — a contrast, a complication, a detail that adds depth. The resolution is the close — a specific call to action or a punchline that ties back to the hook.
This sounds like creative-writing-class stuff, and it is. It also works. Captions that follow a micro-arc hold attention longer than captions that are just three disconnected statements stacked on top of each other.
The engine now runs on Claude Sonnet by default with a higher temperature setting (0.9), which gives it more creative range while the anti-patterns list keeps it from going off the rails.
Topic diversity and dedup
If you generate a week's worth of posts at once, the engine now tracks what it's already written and actively avoids repeating angles. The deduplication window is 200 posts deep — it compares every new caption against the last 200 to make sure the opening, the hook, and the structure aren't too similar to something recent.
It also knows what day it is. A post scheduled for a Monday reads differently from a post scheduled for a Friday afternoon. Date-aware writing was always possible in theory; now it's built in.
The "go deeper" button
Sometimes the first pass is fine but you want more. The "go deeper" option tells the engine to take the same topic and rewrite with more specificity, more research, and a tighter hook. It doesn't just rephrase — it digs into details the first pass skipped, pulls in context, and delivers a caption that reads like you spent twenty minutes on it instead of twenty seconds.
Tone and caption length settings
Two new controls in your campaign settings: tone and caption length. Tone options include professional, casual, witty, bold, inspirational, and educational — and the choice sticks across every post in the campaign. Caption length lets you pick concise (1–2 sentences), standard (2–4), or detailed (4–6), depending on the network and the content.
Both settings save to the campaign and apply to every regeneration, so you don't have to re-pick them each time.
Sharper writing, zero effort.
The creative engine does the heavy lifting. You review, edit, and post. The Free plan includes 30 credits a month.


