There’s no shortage of writing on which AI tools nonprofit leaders should consider. Most of it is either a feature comparison that doesn’t account for how work actually gets done or a list of recommendations that treats all tools as interchangeable. Neither is particularly useful if you’re trying to build a working practice that increases your analytical capacity and saves time.

What follows is more specific: the tools I use regularly, the tasks I’ve matched them to, and the reasoning behind those matches. I’m not covering every AI tool I’ve tried — I’m covering the ones that have settled into my workflow because the results justify the time and cost.

The context is a hybrid role: I lead development and technology strategy for a small nonprofit while maintaining a consulting practice in nonprofit strategy and AI adoption. The tasks that matter most involve research, writing, data, and the kind of analytical work that used to require either a larger team or more hours than most nonprofit leaders have.

Claude: Editing, Analysis, Discussion, and Records Work

Claude is my primary tool for work requiring sustained analytical engagement with a document, a dataset, or a problem. I pay for a $20-a-month Claude account.

• Editing and writing. I use Claude for drafting and refining analytical writing — the kind of piece that requires argument, not just information. Where other AI tools tend to flatten individual voices in favor of something generic, Claude can mirror nuance and tone. It’s also the tool I trust most with editorial feedback — I use it to scan professional writing for hyperbole, overstatements, and jargon, and it will tell me when a sentence is performing conviction rather than demonstrating it. This is feedback that improves the work.
• Analysis and discussion. When working through a strategic problem, such as how to frame a program for a specific funder, how to interpret a set of data, or how to approach an organizational decision, Claude serves as a thought partner. It doesn’t default to agreeing with my framing, which makes it useful for stress-testing an argument before you put it out there.
• Records processing and data management. This is where Claude’s Projects feature is my go-to. I’ve built projects that maintain persistent context — organizational documents, grant records, donor data structures, program frameworks — so each session starts with the relevant background already loaded. With those formats set up, I can ask Claude to analyze a draft grant proposal against the current RFP and materials from past grantees, or organize data I’ve gathered into formatted spreadsheets, turning narrative into cells without entering every line myself (though I always vet and track the outcomes).

For nonprofit leaders: investing in a well-structured Project pays off quickly. The first few sessions feel slow because you’re establishing context. After that, the tool operates at a level of specificity that a generic session can’t match.

• What Claude is not good at. Real-time information. Claude’s knowledge has a cutoff date, and it will tell you so — but more importantly, it won’t seek out current data unless you explicitly use a web search tool. For anything requiring current funder intelligence, recent policy developments, or up-to-date organizational information, I move to Perplexity.
• Perplexity: Research, Citations, and Background Intelligence

Perplexity.ai: Research, Research, Research

Perplexity is built differently from Claude. While Claude is a language model you converse with, Perplexity is a research engine that cites its sources. That distinction matters more than it sounds.

• General research. For any question that requires current, sourced information — what a foundation has funded recently, what a policy development means for workforce programs, what other organizations are doing in a particular space — Perplexity is faster and more reliable than asking Claude or running manual searches. It pulls from live web sources, cites each claim, and lets you verify directly.
• Funder and donor records analysis. When I’m building a funder profile or researching a prospective donor’s giving history and interests, Perplexity surfaces public information — IRS 990 filings, foundation websites, news coverage, and grant announcements — faster than any manual research process I’ve used. The output needs verification, but it compresses a two-hour research task into twenty minutes.
• Background research for writing and presentations. Before I write anything analytical, I use Perplexity to check what’s been published recently on the topic, identify data points I might have missed, and verify claims I’m making from memory. It’s the fact-check layer that sits between my knowledge and anything I publish.
• Deep Research for complex landscape work. Perplexity’s Deep Research feature — available on Pro and Max plans — runs parallel searches across multiple sources and synthesizes findings into a structured report. I use it for funder landscape scans, policy environment updates, and research on peer organizations. The following posts in this series cover those use cases in detail.

What Perplexity is not good for. Analytical depth. Perplexity is excellent at finding and organizing information; it’s not a substitute for working through what that information means. When I need to move from research to analysis, I take Perplexity’s output into Claude and work from there. The two tools are complementary: Perplexity fills the context, and Claude interrogates it.

Perplexity Computer: Multi-Step Autonomous Research Projects

Computer is Perplexity’s agentic feature — available on Pro and Max plans — and it operates differently from the two tools above. Rather than responding to a single prompt, Computer runs a multi-step research process autonomously: launching parallel research threads, pulling from multiple sources simultaneously, synthesizing findings, and producing a structured deliverable. You define the task, set it running, and return to a finished output.

I’ve recently used Computer for projects where the research is complex enough that managing it manually would take most of a day: multi-geography program research, comprehensive funder landscape analyses, and briefing documents that require synthesizing a large number of sources into a structured format.

The autonomous framing is accurate but worth qualifying. Computer doesn’t run well on vague tasks. Having experience writing multi-step prompts, specifying clear deliverables and outcomes, and being specific about scope, audience, and data presentation formats will help streamline the work. The preparation — establishing a clear document structure, precisely scoping the geography and source types, and defining the output format — determines the quality of the output. What it does is execute a well-defined research plan faster and more thoroughly than a single researcher could manage manually.

One problem that warrants attention before you use it: Computer’s default behavior is to take actions on your system — uploading files, accessing connected apps — without prompting for permission at each step. That’s a default worth correcting before your first session. Once I learned it (because it uploaded files to my Google Drive without asking), I told it never to do that again. A best practice is to tell it to check with you before taking any action outside the research itself.

The practical division of labor. In a complex research project, I typically use Computer for information gathering, bring the output into Perplexity for follow-up searches or verification, then move everything into Claude for analysis, synthesis, and any written deliverables the project requires. That sequence — Computer for scale, Perplexity for verification, Claude for analysis — covers most of what a small nonprofit’s research and writing work requires.

What This Costs

A rough monthly picture for the workflow above:

• Claude Pro runs $20/month and covers the editing, analysis, and Projects work without additional credit costs for most tasks.
• Perplexity Pro runs $20/month and covers standard research and a meaningful volume of Deep Research queries. Heavy Deep Research use — multiple landscape scans per month — may warrant moving to Max at $200/month, which includes 10,000 credits for autonomous tasks. The nonprofit Enterprise Pro discount (contact enterprise@perplexity.ai) brings that down for eligible organizations. So far, I’ve done one Deep Research project that burned through my credits, so I bought more for another $20, a one-time purchase.
• Perplexity Computer sessions can cost credits beyond the base subscription because they are very token-intensive. A complex 90-minute research session runs roughly $12–15 in credits at current rates. For occasional large projects, the pay-as-you-go credit model makes more sense than a higher-tier subscription.

Total for a month with regular Claude and Perplexity use, plus one or two Computer sessions: approximately $50–60. For the research and analytical output that replaces it, it’s a reasonable unit cost for most nonprofit budgets. I haven’t committed to the higher tiers yet and may not need to.

The Underlying Logic

Each tool has a focus area because it has a structural advantage: Perplexity finds and cites current information; Claude analyzes and refines; Computer runs multi-step research autonomously when the task is too complex for a single session. Using each for what it’s built for yields better results than using a single tool for everything.

The more consequential point concerns workflow design. AI tools don’t improve work by their mere presence. They improve work when they’re integrated into a practice with clear handoffs, defined use cases, and clear limits. The organizations that get the most value from these tools aren’t the ones that have tried the most tools. They’re the ones who are specific about what they’re trying to accomplish and match the tool to the task.

That’s the frame for everything that follows in this series. 

Working with these tools

 If you’re interested in building a Perplexity research space for funder prospecting or donor research, I offer short-term project work — typically five to thirty hours over a month or two — to help organizations set up and test these workflows or to run the research directly. I also work with development staff who want to learn the methodology and build it themselves.

 Feel free to reach out if either sounds like a fit: get in touch