How to build a context document that makes AI sound like you
This guide is based on the We Spent The Last 6 Months Talking to Marketers (Then We Built AI Lab) webinar, published on the AI Lab by ActiveCampaign.
Get the quick-start guide
What will I accomplish with this guide? By the end of this guide, you’ll have a context document: a single living file that gives your AI everything it needs to write, plan, and decide in your voice. It’s based on the workflow Pam, a life and business coach, walked through in this webinar — the same 100-plus-page file that gets her to 75–80% of finished output before she edits a single word.
Before you start, you’ll need:
- An LLM with persistent context: Claude projects, ChatGPT custom instructions or projects, or a system prompt your tool supports
- ActiveCampaign account if you’ll feed the document into your email work
- A repository of your existing material: blog posts, podcast transcripts, past emails, frameworks, brand background
- A few uninterrupted blocks of time over several weeks — Pam’s document took three months to build out
Quick reference
- Total time: Three months for a comprehensive document like Pam’s; a usable first version in two to four weeks
- Tools needed: Your LLM of choice, a document editor (Google Docs or Notion works well), source material from your existing work
- Key output: A 50–150 page reference document that you paste, attach, or load into a project so the AI has the context it needs to act in your voice
Watch this section
For full context on the following topics, watch these sections of the webinar:
- Pam’s introduction and the 142-page document — [11:06]–[11:39]
- Pam in her own words on context engineering and the Black Friday moment — [12:13]–[13:23]
- How Pam uses the document with ActiveCampaign for email campaigns — [13:23]–[14:00]
The workflow
Phase 1: Inventory what your AI needs to know about you
After this phase, you’ll have: a list of every piece of context your business depends on, mapped to the source material that captures it.
- List the topics your AI will need to make decisions on: brand voice, frameworks you teach, your ideal client profile, common objections, current offers, content pillars.
- Find the source material for each topic: podcast transcripts, blog posts, email archives, Google Docs, sales calls, course curricula.
- Note what’s missing: if you have no documented voice description or no client profile, you’ll write those from scratch in Phase 2.
- Decide on a single home for the document: a Google Doc, a Notion page, or a project file inside your LLM. Pam’s lives outside the AI tool so she can update it independently.
Phase 2: Write the foundational sections
After this phase, you’ll have: the irreplaceable parts of the document, the sections that no transcript or blog post can substitute for.
- Write a 1–2 page brand background: what you do, who it’s for, what makes you different. Plain language, first person.
- Write a 1–2 page voice description: how you talk, what you avoid, three or four example sentences in your voice and three or four in a voice that’s clearly not yours.
- Write a 1–2 page ideal client profile: the person you serve, what they’re stuck on, what they’ve tried, what they sound like when they describe the problem.
- Document your frameworks: the named methods, models, or step-by-step approaches you use with clients. Each gets its own section with the steps and an example.
Phase 3: Add your raw material
After this phase, you’ll have: enough first-person source material that the AI can pattern-match your voice without you having to describe it abstractly.
- Paste in 10–20 podcast transcripts or interviews: label each with the topic and date so the AI can cite or reference them.
- Paste in 10–20 of your highest-performing emails: the ones that converted, generated replies, or got forwarded.
- Paste in 5–10 blog posts or long-form pieces: the ones that capture how you teach.
- Add transcripts of customer calls: the ones where prospects describe their problem in their own words. This is where Mariana’s three-question system also lives: pull customer language directly.
I realize most people use AI like a vending machine. They put in a request, receive the answer. In my case, I treat Claude like a thought partner. I make sure I always show up with context, with intention, and with my thinking already in motion.
Phase 4: Load it into your AI and test
After this phase, you’ll have: an AI workspace where the document is loaded and ready, plus a sense of where the document is strong and where it’s thin.
- Choose your loading method: Claude project memory, ChatGPT custom GPT or project, or pasting into a system prompt. Pam used Claude in the browser.
- Run a test prompt that needs deep context: ask the AI to draft an email for a campaign you’ve run before and compare it to what you actually sent.
- Note where the AI hallucinates or sounds off-brand: those gaps tell you what’s missing from the document.
- Iterate the document, not the prompt: when the output drifts, Pam’s instinct is to add to the document rather than rewrite the prompt.
Phase 5: Use the document inside ActiveCampaign work
After this phase, you’ll have: a working pattern where the document feeds your email and campaign planning, getting you to 75–80% of the finished output.
- Use the AI to draft email campaigns: paste the campaign goal, the segment, and the offer; let the document do the rest.
- Move drafts into ActiveCampaign for the last 20–25%: the tone match should be close enough that you’re editing details, not rewriting.
- Plan campaign sequences with the AI as a thought partner: Pam used Claude during her Black Friday event and let it push back when she wanted to quit early.
- Update the document monthly: add new high-performing emails, new customer language from calls, new framing the AI got wrong.
Phase 6: Treat the document as a permanent asset
After this phase, you’ll have: a maintenance habit that keeps the document current, so your AI quality compounds instead of decaying.
- Schedule a monthly 30-minute update: add new transcripts, retire outdated offers, sharpen sections the AI keeps misreading.
- Version the document: a date in the filename so you can roll back if a recent edit breaks output quality.
- Note what changes when context changes: if the AI’s output quality jumps after you add a section, capture that. The notes tell you what mattered.
Related
More data from the AI Lab.