Reference Material

What it does

Reference Material is a plain-text field in your AI Chat node settings. Whatever you paste in — your pricing, your methodology, your onboarding process, your frequently asked questions — Claude has it available for every single conversation that product runs.

It is prepended to the system prompt before every message. Claude treats it as foundational context. It does not need to decide when to use it — it is simply always there.

How much can you put in it?

A lot more than you might expect.

Claude Sonnet 4.6 (the default model on Productised) has a 200,000 token context window. In practice, Reference Material can hold several hundred pages of text before you approach any limit. A full methodology guide, pricing tiers, your FAQ, client types, process steps, and company context combined will almost certainly fit with room to spare.

For comparison: platforms that use RAG (retrieval-augmented generation) store large document libraries but only surface 3–5 text chunks per query — meaning the AI sees a few paragraphs at a time, not the whole document. Reference Material gives Claude the entire thing, so it can reason across all of it rather than guessing which fragment to retrieve.

How to set it up

1. Open your AI Chat node on the canvas and click to open its settings
2. Scroll down to the Reference Material section
3. Paste in your background context
4. Save — it takes effect immediately for all future conversations

Converting a PDF or document

If your methodology, pricing guide, or playbook lives in a PDF or Google Doc, the best approach is to have Claude convert it first rather than extracting raw text:

1. Go to claude.ai and start a new conversation
2. Upload your PDF, document, or paste your URL
3. Send this prompt:

I'm adding my expertise as Reference Material for an AI product. Please read the attached document and restructure all the key information into a concise reference guide.

Keep:
- All specific frameworks, methodologies, and process steps
- Exact terminology and how you define it
- Pricing, packages, and what's included
- Who you work with and the problems you solve
- Important nuances, exceptions, and caveats

Remove: narrative, stories, examples, and redundant explanation.

Format with clear headers and bullet points. The goal is a dense, accurate reference that an AI assistant can use when having conversations on my behalf.

4. Paste Claude's response into the Reference Material field in Productised

Why this produces better results than raw text extraction: A PDF extracted to plain text is full of headers, footers, page numbers, repeated boilerplate, and filler narrative. Claude's compacted version removes all of that and keeps only what an AI needs — dense, structured, and purpose-built for a system prompt. The AI reading it during conversations gets better signal per token.

What to put in it

Your methodology or framework — the specific way you approach your work. Every outcome Claude generates references your process, not generic AI advice.

Pricing and packages — what you offer, at what price, what's included. Claude handles pricing questions accurately mid-conversation.

FAQs — the questions your prospects always ask. Claude handles them correctly without the conversation going off-track.

Company or service context — who you work with, what problems you solve, what makes your approach different.

Terminology — if your methodology uses specific terms or frameworks, define them here. Claude uses your language.

File uploads during conversations

If you want your end user to provide their own context — uploading their CV, brief, or website URL — that is handled separately via the file upload or URL question types in Collection Goals. Your end user uploads the file mid-conversation and Claude reviews it as part of that session.

This is different from Reference Material: Reference Material is your knowledge (always present), file uploads are your end user's content (session-specific).