Putting together a high-quality proposal is a constant tension between adding as much value as possible while managing the application timeline. With these processes usually involving multiple collaborators and different skillsets, getting started and following through is a major challenge.

You can turn these pain points around with AI. In this guide, you’ll learn how to process an RFP to develop a first draft that you can use as a base for collaboration, speeding up the start of the process.

This tool will analyze the RFP document, ask questions to help you fill the requirements, while serving as a brainstorming assistant as you do so. When finished, it’ll export it to your favorite workspace platform, so you can continue refining.

Add an input for the RFP

• Add previous grant proposals

• Configure the LLM

• Connect an LLM export tool

• Set up the chat interface

• Share, track analytics, and improve

Add an input for the RFP

The starter configuration of a Stack AI project already includes a Text Input, an LLM, and an Output node. We’ll add a Files input to help you upload RFPs, making it easier to get started. Expand the Inputs section, drag and drop a Files node onto the canvas.

Connect the Files to the LLM node.

Add previous grant proposals

Help the LLM understand what good grant writing means by adding a Knowledge Base to the project. When connected, Stack AI will search for and retrieve the most relevant chunks of previous grant proposals based on your inputs and the uploaded RFP.

Expand the Knowledge Bases section on the left side, drag and drop a Documents node onto the canvas.

Click the Document node to open the settings, and upload previous proposals into the knowledge base.

Stack AI supports multiple data sources: instead of uploading files manually, you can connect it to the platform you’re storing your documents. Browse the list of integrations in the Knowledge Base section to find yours.

Connect the Input and Files inputs to the Knowledge Base input handle: this will help search for the most relevant information for each instruction and RFP document.

Connect the Documents to the LLM node.

Let the LLM choose when to read your previous grant proposals

In the architecture we just set up, the workflow will always search for the most relevant proposals and pass them on to the LLM for every message you send. This is useful if you predict that this data will be necessary for every step of the grant writing process within the AI tool.

However, Stack AI also supports a method that lets the LLM decide whether it needs to consult your previous grant proposals or not. It analyzes the content of each of your prompts and, based on your intent, it starts the search, retrieves the relevant chunks, and uses those to generate the answer.

If you’d prefer this option, use the tool feature, accessible by clicking the LLM node and selecting the corresponding button.

This opens a list of actions the LLM can execute both within and outside of Stack AI. To enable the LLM-driven search decision, select the Search Knowledge Bases tool.

Click the Settings icon to configure the knowledge base search.

Click Add Knowledge Base to select the knowledge base you’d like to connect to this LLM tool.

If you configured the Documents node previously, a new Knowledge Base was automatically created to contain all your uploaded documents. You’ll notice that you can reuse it here.

Once you select the knowledge base, the LLM will now selectively retrieve (or not) the most appropriate chunks of past grants depending on the content of your prompts.

Configure the LLM

While OpenAI’s models are generally flexible for most tasks, we’ll use Anthropic Claude for its longer conversational memory, a lean for thorough responses, and overall better writing abilities.

Click the OpenAI node on the canvas to open its settings, open the Provider dropdown, and select Anthropic.

Select Anthropic’s latest model at the time of writing, the powerful yet fast Claude 4 Sonnet.

Stack AI has a Data Protection Addendum with Anthropic, which means that Claude won’t retain any of your information on its servers, increasing privacy and security.

The model needs to understand its role. Paste the following system instructions into the Instructions input field.

Role: You are an expert grant writing assistant trained to help users craft high-quality, funder-aligned proposals using AI. You have access to a knowledge base of previously successful grant proposals to inform tone, structure, and formatting best practices.

Your Objective: Guide the user through the process of creating a complete, compelling grant proposal based on a specific Request for Proposal (RFP). Ask targeted, sequential questions to collect all required inputs, then generate a polished proposal using the funder's language and expectations.

Step-by-Step Process:

1. Analyze the RFP. Extract and document the following:

• Required proposal sections or components

• Evaluation criteria, scoring rubrics, or priority areas

• Expected language, tone, and style preferences

• Page limits, word counts, formatting or submission constraints

• Any other funder-specific instructions or deadlines

2. Use the extracted details to guide a structured, conversational intake.

• Ask one question at a time

• Begin with foundational info (e.g., project title, purpose, target population)

• Progress into more specific areas aligned with the RFP’s structure (e.g., goals, activities, partnerships, budget)

• If the RFP lacks detail, default to standard grant proposal components:

  • Executive Summary

  • Statement of Need

  • Goals and Objectives

  • Methodology or Work Plan

  • Evaluation Plan

  • Budget and Justification

  • Organizational Capacity

  • Sustainability Plan

3. Adapt your language and approach:

• Mirror the terminology and tone used in the RFP

• Follow established stylistic norms from the knowledge base of successful proposals

• Maintain a professional, persuasive, and supportive tone

• Where appropriate, suggest phrasing or offer examples based on prior proposals

4. Maintain clarity and usability throughout.

• Keep each question focused and jargon-free

• Add brief context or examples when necessary

• If the user is unsure, offer suggestions, templates, or explain what’s typically included in that section

5. Proposal generation and delivery:

• Once all necessary information is collected, confirm with the user before generating the full draft

• Structure the proposal according to funder expectations and best practices from the knowledge base

• Preserve any formatting, length, and content constraints

• Save the final output as a new Notion page (or designated workspace), clearly titled and organized

Tone and Ethics Guidelines:

• Be supportive, concise, and professional

• Never fabricate information — rely solely on user inputs or provide options for them to choose from

• Always prioritize clarity, compliance, and alignment with the RFP and funder's intent

Next, add the following to the Prompt section:

User Message:

RFP content:

Knowledge base:

Type a forward slash to open the node variables, choosing the Text Input for User Message, Files for RFP content, and Documents for Knowledge base, as shown in the image below.

To fully leverage the model’s context window, scroll down the settings list and set the Window Size to 100.

Connect an LLM export tool

Your AI-powered grant writing assistant will help you get started with a generated first draft, but you’ll still want to make changes before sending. At the end of the process, the LLM can automatically export its work to an external platform such as:

• ClickUp, Notion, Coda for storage, editing, and team collaboration

• Gmail or Outlook to quickly share via email with stakeholders

• Zapier webhooks to send it to over 7,000 apps you use for work

• Or a custom platform previously set up via API integration

In the LLM settings tab, click Add Tool to begin.

Search for your desired integration on this list. For this tutorial, we’ll be exporting the first draft to a new Notion page.

Stack AI supports easy no-code connections with other work software. Click New Connection and follow the instructions on how to integrate both platforms.

Once connected, scroll down to the Configurations section and select the Parent Page Id. This will place the new RFP within a specific page within the Notion workspace, making it easier to find later.

The tool is now ready to use. Notice that we’ve added the following sentence to the system instructions of the Claude LLM before:

5. After you generate the full grant proposal, create a new Notion page and save the content there.

Once the LLM finishes asking you all the questions you need to consider based on the RFP, it will automatically export the first draft to Notion, so you can open and polish it there. It’s important to review the output of the tool and improve it with more concrete details before sending.

Configure the chat interface

Click the Export tab at the top left of the screen.

Stack AI offers many user interfaces to export your workflows. For this use case, we’ll use the Chat Assistant: select it from the dropdown.

This tab offers a wide range of controls that affect the interface’s user experience, such as style, security, and subdomain settings. We’ll focus on the Fields section. Expose the Files input to the user by ticking the doc-0 checkbox.

Notice that the icon in the preview changes to a paper clip, indicating the functionality to attach files is active.

Click Save at the top right to save these changes to the exported interface.

Share, track analytics, and improve

Use your assistant by clicking the Published Interface link. Share it with your team by copying and pasting this link in an email or internal communication channel.

Track your tool’s performance in the Analytics tab, where you can see a breakdown of core metrics, download logs, and get an AI report on usage.

Review all chats from all users in the Manager tab, helpful to troubleshoot issues and understand how your team is using the tool.

Faster writing, more funding

Building a winning proposal is complex and time-consuming. The workflow doesn’t have to be: focus on what’s strategic and high-value, while AI takes care of analyzing requirements, asking the core questions, and producing the first draft for you.

Automate this and more workflows with a free Stack AI account.