What AI Cannot Do (And Where It's Dangerous)

Introduction: Know the Limits

The most dangerous use of AI comes from assuming it can do something it can't. In grant work, this often happens when professionals mistake AI's fluency (its ability to write convincingly) for competence in a domain where it doesn't have competence. An AI can write a grant deadline in a way that sounds authoritative, even if that deadline is completely fabricated.

This lesson catalogs the major limitations. Memorize these. Understand them. Build habits around them. The difference between an AI-powered grant professional and a dangerous one is knowing exactly where the line is and never crossing it.

1. AI Cannot Do Original Research

This is non-negotiable: AI cannot research. It cannot go to the Ford Foundation website and check current grant deadlines. It cannot search scholarly databases for peer-reviewed research on trauma-informed care. It cannot access news archives to learn about recent policy changes. It cannot call a program officer and ask questions.

Why This Matters for Grants

Grant applications require current, accurate information. Deadlines, eligibility requirements, funder priorities, recent policy changes, published research, demographic data—all of this must be current and verified. AI's training data is from the past. It cannot verify against current reality. An LLM trained in mid-2023 will confidently give you 2023 information, even though we're now in 2024 or later.

Critical Rule

Never trust AI for any information that needs to be current or accurate. Never use AI to research grant deadlines, eligibility, program requirements, funder priorities, or any factual information critical to your proposal without independent verification from the official source.

Case Study: The Hallucinated Deadline

A grant professional asks ChatGPT: "What is the next application deadline for the Ford Foundation's Youth Opportunity Initiative?" ChatGPT responds confidently: "The next deadline is March 15, 2024." This sounds authoritative. The date has the right format. But it's likely hallucinated—ChatGPT has no access to current Ford Foundation information and is just predicting a plausible-sounding deadline. If you submit a proposal based on this deadline, you're submitting based on an AI hallucination, not real information. The real deadline might be different. The program might have changed. The deadline might have passed.

The Right Way to Use AI

You find grant opportunities through official sources: foundation websites, grants.gov, grant databases, your networks. Once you've identified a real opportunity with real deadlines and requirements, you can use AI to draft proposals. But never use AI to find or verify grant opportunities in the first place.

Apply This: Build the Habit

Make it a rule: Never ask AI "What grants are available for X?" or "What are the deadlines for Y?" Instead, always research opportunities first through official channels, then use AI for drafting once you've verified the opportunity is real.

2. AI Cannot Verify Its Own Claims

Remember how LLMs work: they predict the next token based on patterns in training data. They have no way to check if what they're saying is true. An AI can confidently state incorrect statistics, misquote policies, attribute quotes to the wrong people, or invent research findings.

Why This Matters for Grants

Grant proposals often include citations, statistics, policy references, or research findings. If you use AI to help you draft these sections and don't verify independently, you might submit false information to funders. This isn't just ineffective—it's ethically problematic and potentially illegal if the information is material to the funder's decision.

Real Example: Fabricated Statistics

You ask AI: "What percentage of homeless youth in the US have a mental health diagnosis?" AI might respond: "Research suggests that approximately 68% of homeless youth have a mental health diagnosis." This sounds authoritative and specific. But did AI verify this statistic? No. It just predicted this is the kind of statistic that appears near questions about homeless youth mental health. The real percentage might be 60%, 75%, 50%, or different studies might give different numbers. You cannot put this statistic in your proposal without verifying it in actual peer-reviewed research.

Key Takeaway

When you use AI-generated content in a proposal, you are personally responsible for the accuracy of every claim. If you use statistics, citations, policy references, or research findings that AI suggested, you must independently verify them. The AI's confidence is not verification. Your verification is verification.

3. AI Cannot Understand Your Community

AI has learned patterns about communities from training data. It can discuss common challenges in urban neighborhoods, rural areas, or immigrant communities based on what it learned from documents. But it doesn't understand your specific community. It doesn't know what the lived experience is like. It hasn't walked your neighborhood. It hasn't met your clients. It doesn't understand the nuanced cultural context.

Why This Matters for Grants

Strong grant proposals demonstrate deep understanding of the community being served. "We serve the homeless population" is weak. "We serve homeless youth ages 16-24 in the downtown corridor who are predominantly LGBTQ+, many of whom have aged out of foster care and face discrimination seeking shelter"—that's specific and demonstrates understanding. AI can write the first version. Only humans can write the second, because it requires lived knowledge of place.

What AI Gets Wrong

You ask AI: "Describe the needs and challenges of immigrant families in [Your City]." AI will generate something based on general patterns about immigration: economic barriers, language challenges, acculturation stress, access to services. This might be true, but it might miss what's specific about your city. Maybe your city has a strong informal support network. Maybe language barriers are less common. Maybe the biggest barrier is actually employment discrimination due to licensing restrictions. AI can't know this. Only community members and organizations serving the community can know this.

4. AI Cannot Make Strategic Decisions

Strategic grant work requires judgment about mission fit, funder relationships, competitive landscape, and organizational capacity. Which funders to pursue? What to emphasize in proposals? How to position your organization? These are strategic decisions, and they require human judgment rooted in understanding your organization and the field.

Why This Matters for Grants

AI might suggest that multiple funders "would be a good fit" because your work touches on their priority areas. But maybe one funder is actually more strategic because you have a relationship with their program officer. Maybe another funder's grants are too restrictive. Maybe pursuing one funder could damage your relationship with another. AI can't know this. These are human strategic decisions.

The Danger

Never let an AI system tell you which funders to pursue or how to position your organization. Grant professionals make these decisions based on strategic judgment, relationships, organizational capacity assessment, and market knowledge. Use AI to draft once you've made the strategic decision. Don't use it to make the strategy itself.

5. AI Cannot Build Relationships

Modern grant work is increasingly relationship-based. Program officers, foundation staff, and fellow grant professionals in your network are key to success. AI can draft professional emails, but it can't build the relationships that lead to funding.

Why This Matters for Grants

The best grants often come from relationships. A program officer who knows your organization, believes in your mission, and has seen your impact over time is more likely to fund you and more likely to provide guidance. AI can help you communicate, but the relationship itself is irreplaceable human work.

What AI Cannot Do

AI can draft an email to a program officer. But it can't develop the trust that comes from months of communication, attendance at funder events, and demonstrated commitment. It can't notice when a funder's priorities shift and proactively reach out. It can't ask thoughtful questions that demonstrate you've read their last few grants. These are human strengths that AI can't replicate.

6. AI Cannot Guarantee Accuracy in Editing

Earlier we said AI is good at copy editing. That's true for clarity and style. But AI is terrible at fact-checking. It can catch grammar errors, but it can't catch factual errors. If you tell AI "Our organization was founded in 1995" and that's actually wrong (you were founded in 2005), AI will happily fix your grammar without catching the error in the underlying fact.

Key Takeaway

Use AI for copy editing (grammar, style, clarity, flow). But never use AI to fact-check. Only humans can read a statement and think "Wait, is that actually true?" and then verify it. AI doesn't do this verification.

7. AI Cannot Understand Your Mission

Your nonprofit's mission is something you've developed over years, something you live every day, something that guides your decisions. AI can read your mission statement, but it doesn't internalize what that mission means. It doesn't feel the passion. It doesn't understand the tradeoffs you've made in how you define your work.

Why This Matters for Grants

The best proposals are mission-centered. They don't just describe programs; they explain why those programs matter, connected to something deeper than funding. AI can echo your mission language but can't advocate for it with authentic conviction. Only a grant professional who believes in the mission can do that.

Summary: The AI Boundaries in Grant Work

Here's a simple rule: Use AI to amplify the work you do. Never use AI to replace the expertise you provide.

Task	Can AI Do It?	Rule
Research grants	❌ No	Verify all grant info against official sources
Verify statistics/claims	❌ No	Check all facts independently
Understand your community	❌ No	Write community-specific content yourself
Make strategy decisions	❌ No	Decide which funders to pursue yourself
Build relationships	❌ No	Develop funder relationships yourself
Fact-check documents	❌ No	Do human review of all claims
Brainstorm ideas	✓ Yes	Use AI to generate options; you evaluate

The Human Edge

AI is a tool. You are the expert. You have:

• Knowledge of your organization and community
• Judgment about what's true and what's false
• Relationships with funders and peers
• Understanding of your mission and values
• Strategic thinking about your organization's future
• Professional responsibility for what you submit

AI amplifies these strengths. It doesn't replace them. Understanding this distinction is what separates effective AI use from dangerous misuse.