Build a Claude Project as Your RFP and Security Questionnaire Answer Library

Sign in at claude.ai. Look in the left sidebar for a Projects section, and use the option to start a new one. Claude's exact wording for that button can shift between updates, so look for something short like "New project" or "Create project" rather than a specific label.

1. Give the project a clear, searchable name such as "RFP Answer Library."
2. Skip any optional description field for now. You can fill it in later.

What you should see: An empty project workspace with a place to add instructions and a separate area for uploaded files. Troubleshooting: If you don't see a Projects option at all, your account type may not support it yet. Check your plan settings before assuming the feature is missing.

Claude Projects have a dedicated instructions field (Claude calls it project instructions) that applies to every conversation you start inside that project. Set the rules there once instead of retyping them each time.

Open the instructions field and paste:

You are helping a sales engineer draft first-draft answers to RFP and security questionnaire questions, using only the documents uploaded to this project's knowledge base. For every answer:
1. State the answer in plain, specific language.
2. Name the exact uploaded document you drew the answer from.
3. If you are not confident, say so directly instead of guessing.
4. If nothing in the uploaded documents covers the question, say that plainly rather than inventing an answer.

What you should see: The instructions save automatically or after you click away from the field.

Find the knowledge section of the project (labeled Project knowledge in Claude's interface) and upload your source material.

1. Add your current product spec sheets.
2. Add your security documentation (SOC 2 summary, data handling policy, whatever your team maintains).
3. Add any past questionnaire answers your team already considers approved.

Confidentiality caution: Upload only your own company's product and security material, plus already-approved past answers. Leave out anything specific to the customer whose questionnaire you're currently answering. No customer names, no details about their network or environment, nothing that arrived under their NDA. This library should describe what you sell, not what any single customer's setup looks like. Security questionnaire content in general often carries an NDA of its own, so if you're ever unsure whether a document is safe to upload, check with your security or legal team before adding it.

What you should see: Each uploaded file listed in the project knowledge area, usually with its file type and size. Troubleshooting: If a document is too large to upload as one file, split it by section (product overview, security controls, compliance certifications) rather than trying to compress it.

Start a new chat inside the project and paste one actual question from a current or recent questionnaire.

What you should see: A drafted answer that names which uploaded document it used. If the answer feels generic or doesn't cite a source, your knowledge base is probably missing the document that covers that topic.

Verification step: For any answer that states a specific technical fact (a certification name, a data retention period, an SLA number, an encryption standard) open the document Claude names and confirm the sentence is actually there. Treat the citation as a pointer to check, not proof on its own. This matters most for compliance and security claims, since a wrong answer in a signed questionnaire can become a contractual commitment.

Every time you write a genuinely new answer, add it back into the project knowledge. The library gets faster and more accurate every quarter instead of staying frozen at whatever you uploaded on day one.