Most AI-written posts sound alike for a simple reason: the AI was told almost nothing. "Write a tweet about our product" gives you the average of every tweet about every product. In AImitatio, the fix is in two fields you fill in once: the Context of your app and the prompts it draws from. This guide goes through both, in the order you meet them in the app.
Step 1: Write the Context once, and write it like a briefing
Every app you create in AImitatio has a Context field. Its text is sent along with every prompt for that app, so it is the place for anything that should always be true about your posts. Write it as if you were briefing a new team member:
- Who is speaking: "We are a two-person bakery in Leipzig. We post as 'we', never as 'I'."
- Who is listening: "Our followers are locals who like seasonal baking and want to know what's fresh this week."
- Tone: "Warm, short sentences, a little dry humour. No exclamation marks in every line."
- Hard rules: "Never mention prices or discounts. At most two hashtags. Write in English."
Concrete details beat adjectives. "Friendly and professional" is how everyone describes themselves; "we explain things the way you would to a neighbour over the counter" gives the AI something to imitate.
Step 2: Write several prompts, each with one job
An app can have as many prompts as you like, and on every scheduled run the scheduler picks one of them at random. That is your main tool against repetition. One prompt run fifty times will drift into the same post; six prompts with clearly different angles give you a varied feed without any extra work later.
Give each prompt one clear angle, for example:
- "Share one practical tip about storing bread so it stays fresh longer."
- "Describe one ingredient we use this season and why it's in season now."
- "Ask our followers a simple question about their favourite weekend breakfast."
- "Explain one step of sourdough baking that beginners often get wrong."
Each prompt also has an optional image prompt and a category. The image prompt describes the picture that goes with the post ("close-up of a sourdough loaf on a wooden board, morning light, no text"). The category keeps your own overview tidy when you look back at what went out.
Step 3: Let the built-in safeguards do their part
A few things happen automatically on every run, so you don't need to cover them in your prompts:
- Repetition check: your three most recent posts for that app are passed to the AI with the instruction not to repeat or closely resemble them.
- Length limits: each network has its own limit (280 characters on X, 300 on Bluesky, 500 on Mastodon). If a draft is too long, the AI is asked to shorten it, up to three times, before it is cut as a last resort. Prompts that ask for "one tip" or "one question" rarely run into this.
- Images: if you set an image prompt, the image is generated with FLUX on our own GPU first, with an external image service as a fallback.
Text comes from one of two models, alternating between runs, one of which is our locally hosted Teuken. That's one more reason to put your voice into the Context: it's the part that stays the same whichever model writes the post.
Step 4: Schedule with a little randomness
When you schedule a job you choose a start date and time (UTC), an interval in minutes and a random value in minutes. The scheduler waits a random amount of time between zero and that value before each post, so your posts don't land at exactly the same minute every day. Pick an interval that matches how often you would post by hand. Automation should keep you consistent, not flood your followers, and every network has its own rules on automated posting that are worth reading first.
Step 5: Read what went out, then refine
There's no approval queue. Once a job is scheduled, posts go out on their own. After each run you get an e-mail with the log: the post text that was published, the models that produced it, and any warnings. You can also see past posts with their prompts in your app's post list. If one prompt keeps producing posts you don't like, rewrite or delete it. If every post sounds off, the Context is usually the problem.
A quick checklist
- Context says who speaks, to whom, in what tone, and lists the hard rules.
- At least four or five prompts, each with a different angle.
- Image prompts describe a scene, not a slogan.
- A sensible interval, with a random delay switched on.
- You read the first few log e-mails and adjust.
Want to try it with your own account? Create a free account and set up your first app, Context and prompts.
