Prompts, workflows, and tools from our workshops. Everything you need to start using AI across fundraising, marketing, research, and daily operations.
An LLM is a frozen state: trained on massive data and locked in place. The model itself doesn't change. But the products built on top of it (Copilot, ChatGPT) add layers that make it powerful: your prompt gives it context and instructions, tools let it search the web and create files, skills give it specialized know-how, and memory lets it remember your preferences over time. The LLM predicts the next word based on patterns. Everything else is scaffolding that makes that prediction remarkably useful.
Incredibly capable, but needs clear direction. Vague instructions produce vague results. Be specific about what you want, who you are, and what the output should look like. If you can write in English, you can do everything in this toolkit.
Complete these steps to unlock Copilot's full potential for your work at Hartman.
Copilot lives inside Outlook, Word, Excel, PowerPoint, and Teams. Look for the Copilot icon 
in each app's ribbon.
Summarize email threads, draft replies to donors, prioritize your inbox, and coordinate meeting schedules across teams.
Generate grant proposals, donor reports, program summaries, and board memos. Drop in data and get a formatted document.
Use Agent Mode (under Tools) to analyze donor data, build dashboards, and create reports in plain English.
Generate presentations from data or Word docs. Ideal for board presentations, program overviews, and donor stewardship decks.
Get meeting recaps, action items, and transcripts. Ask "What did I miss?" and Copilot summarizes the conversation.
Connect Monday.com and other platforms to Copilot. Chat with your project boards in natural language.
NotebookLM is a free tool by Google, completely separate from Microsoft. Upload documents and have AI conversations grounded only in those sources. No hallucinations from random web data. Google's guide →
Start every day with a clear picture of what needs your attention. Set up a scheduled prompt that runs automatically.
Generate professional donor communications that maintain Hartman's voice while saving hours of drafting time.
Upload a podcast transcript to NotebookLM or Copilot and extract multiple content pieces from a single episode.
Use Copilot to generate first drafts of grant proposals. Always review and refine. AI gets you 80% there in 10% of the time.
Speaking prompts is faster than typing. Install Wispr Flow and start dictating your prompts to Copilot instead of typing them. Most people speak 3-4x faster than they type.
There is absolutely no doubt that we are working with agents now. Everyone in the tech industry already has multiple agents working alongside them. Right now, knowledge workers sitting in front of a computer are starting to work with agents because productivity gains are too significant to ignore. The good news is Hartman is way ahead of the curve.
Go to m365.cloud.microsoft. In the Copilot chat, click the "Create Agent" button, or go directly to copilotstudio.microsoft.com.
Give your agent a clear name (e.g., "Hartman Email Writer" or "Podcast Repurposer") and a brief description. Upload a logo or icon. The description helps you and your team understand what the agent is for.
This is the most important part. Write detailed instructions covering: the agent's role, tone and style guidelines, output format rules, and any constraints. System prompts are far more detailed than regular chat prompts.
Upload documents the agent should reference: Hartman's website, program descriptions, past proposals, brand guidelines. Connect Outlook and OneDrive so the agent can access your files and email history.
Choose what the agent can do: search the web, read your emails, access your calendar, create files. Enable only what the agent needs for its specific job.
Create 3-4 starter prompts that appear when users open the agent. Examples: "Draft a thank-you letter to [donor name]" or "Turn this podcast transcript into a blog post."
Use the test panel to try your agent. Compare output to regular Copilot. Adjust the system prompt until output is consistently good. Once working, share with your team via the agent library.
The Excel Agent is a prebuilt Microsoft agent embedded in Excel. Because Microsoft owns LinkedIn, the Copilot Excel agent can synthesize LinkedIn data — which is blocked to all other AI systems. Combined with web search, it turns a raw event list into a researched, ranked prospect list in minutes instead of days.
Take an event registrant list and identify who has donor capacity and philanthropic propensity. The Excel agent cross-references each name with public web data and LinkedIn profiles to produce a ranked, enriched prospect list.
Download the sample prospect list (or use your own event registrant sheet). Upload it to OneDrive, open in Excel, and click the Copilot icon. Select Claude from the model dropdown for best results.
Paste this prompt to identify individuals with demonstrated philanthropic backgrounds:
Once you have your ranked list, generate personalized briefings for leadership meetings:
Use the Excel Agent to build formulas, create dashboards, run data analysis, and generate visualizations — all in plain English. No need to know Excel functions.
Agent Mode requires your data to be on OneDrive. Upload your spreadsheet first, then open it in Excel.
Before building the agent, go to plain Copilot chat and ask it to write a professional email about a topic of your choice. Screenshot the result. This is your "before" — the baseline you'll compare your agent against.
In Copilot Chat (Work tab, DeepThink mode), run this prompt. It analyzes your sent emails and generates a system prompt that captures your personal style.
In Copilot, click "New Agent" in the left sidebar. Toggle to "Configure" (not "Describe"). Fill in:
Add these rules to the system prompt generated in Step 1 to fine-tune behavior:
Ask the agent: "Draft a reply to Anna about the donor research." Does it sound like you? If it repeats mistakes (too formal, wrong sign-off, uses "I hope this email finds you well"), go back and add explicit corrections to the system prompt.
Once built, your agent is available everywhere in Microsoft 365. In any app — Outlook, Word, Excel, Teams — type @ in the Copilot sidebar and select your agent by name. In Outlook, the agent sees the email thread you're looking at and can draft a reply in context. Copy-paste the draft into the email (sending from Copilot is not yet available, intentionally for safety).
Build agents by program (e.g., one for each initiative). The donor name and time period are always passed as inputs at runtime. One well-configured Report Writer agent serves all donor reports for that program.
Upload an existing report or raw notes to plain Copilot and ask it to write a donor report. Screenshot the result — this is your "before."
Take an existing report you want to replicate. Paste it into Copilot chat and ask AI to extract the structure:
In Copilot, click "New Agent" → "Configure":
Upload raw materials (meeting notes, WhatsApp exports, staff updates — multiple documents at once). Type: "Produce the report." The agent will ask for the donor name and time period, then generate the full structured document.
Chain agents for a full production workflow:
The Report Writer agent is shareable across the team. Click the three dots next to the agent, click Share, and select specific team members. Build one well-configured agent per program and share it, so everyone producing reports for that program uses the same trained agent.
You can't get frustrated if it hallucinates or gives you an inaccurate answer. You need to help it. Inject more context. That's how you work with AI. You're fact-checking instead of researching from scratch.
What's next: Start with one agent this week. Pick the task you spend the most time on and build an agent for it. Test it for a few days, refine the instructions, and then share it with your team. The goal is to have 3-4 specialized agents covering your most repetitive work within a month.
GEO is the next stage after SEO. While SEO was about ranking on Google's page 1, GEO is about influencing what AI systems say about you when users ask. 60% of searches are now zero-click — the answer is consumed without ever visiting your site.
Put a 50-word direct answer at the top of every blog post. AI scrapes the opening paragraph first. Don't bury the lede.
Post on YouTube, Reddit, LinkedIn. Publish in third-party publications. AI prioritizes these sources over brand websites.
Include author bios, expert quotes, statistics, and citations in all content. AI prioritizes credibility markers.
Clear headers, FAQ format, short paragraphs. Make content easy for AI to parse and lift into answers.
Put yourself in the shoes of donors. Ask the questions you think donors are asking about your organization before they engage and donate. Then go ask those questions in ChatGPT, Claude, and Perplexity. What do you see in the answers? Are you mentioned?
Vibe coding means describing what you want in plain language and letting AI generate the code. The technical substrate is still code, but the user never writes it. We're going to be able to use basic simple language to build software and applications.
Gemini Canvas is Google's free tool for creating interactive visual outputs. Go to gemini.google.com.
They're going to want to understand impact in a visually friendly, accessible way. In one to two years, interactive visual reports will be the standard for donor communications.