If you explain your niche, audience and style to Claude from scratch in every new chat, the copy comes out faceless — "like a neural network wrote it". The fix isn't a longer prompt but setting up Projects properly: a knowledge base that doesn't reset between chats. Below are three steps and a working prompt with the structure.
The point: a Project is the project's memory, not a one-off chat
An ordinary Claude chat only remembers context inside itself. A Project is a container: the files and instructions loaded into it are available in any new chat inside that project. You set the project up once for a specific job (content for Instagram, say), and after that each new topic is a new chat but with the same base of examples and style.
Step 1 — create a Project with a specific name
Claude → Projects → "Create project". The name should be specific and describe the job: "Instagram content", not "Project 1" or "Test". This isn't cosmetic — Claude uses the name to orient itself on what the project's knowledge base is for.
This will be your permanent working project for content, not a one-off conversation on a single topic.
Step 2 — upload 4 types of file into the project base
In the project settings (or by attaching files in the first chat inside it), upload:
- a portrait of the target audience (who reads you, what their pains and interests are);
- 3-5 examples of your best posts (not any posts — the ones that actually worked);
- a description of the tone of voice (formal/informal, short/long sentences, emoji yes/no);
- a list of the topics you plan to cover with content.
The more precise and specific these inputs are, the fewer boilerplate phrases come out and the more the text sounds like your voice rather than the averaged voice of a neural network.
Step 3 — set the structure with one prompt
The "hook → value → CTA" structure is the working minimum for a short post. The prompt below locks it in so that Claude holds the structure in every subsequent post in the project, not just in one reply:
Write a post using this structure:
1. Hook — a gripping first line, no cliches.
2. Value — 3-4 concrete steps or facts.
3. CTA — one clear call to action, with no extra "and also".
Post topic: [insert the topic].
If this structure has already been set earlier inside the project, for the next posts it's enough to write just the topic — Claude will apply the structure set earlier in that same project.
The full workflow
- Create a project with a specific name.
- Upload the 4 knowledge base files (audience, examples, style, topics).
- Set the post structure with the prompt above — once for the whole project.
- For each new post, write only the topic, referring to the structure already set.
If the copy starts drifting away from the intended style over time, it's worth refreshing the base: add 1-2 new example posts that got the best response.
A level up: turn the structure into a Skill
The hook-value-CTA prompt can be set up not as a prompt you paste by hand but as a Skill — then the
command is invoked in one line (/post-structure), and Claude walks through all three parts itself
without making you copy the instruction text every time. A Skill is created in Settings → Capabilities
→ Skills → Create skill — you paste in the same instruction as in the prompt above, only phrased as a
permanent role ("You are a post writer who always writes to this structure...").
How we do it
The qvib.pro knowledge base has separate cards on setting up Claude Skills and on content structures for different formats — from posts to scripts. We apply the same "lock the structure in once" principle in our own engine for generating websites: the more precisely the inputs are set, the fewer edits are needed after the first result.
Related links
- The "Claude skills" section of the arsenal: qvib.pro/arsenal/skills/
- The "Prompts" section of the arsenal: qvib.pro/arsenal/prompts/
- How the qvib engine works: qvib.pro/