Before AI Helps Nonprofits, It's Going to Sort Them.

“What we experienced just this week was overwhelming demand.”

Michael Flood, LA Regional Food Bank - Source

That’s Michael Flood, who runs the LA Regional Food Bank, in a clip that stuck with me. His food bank had just run out of food. Not run low, run out, shut the event down early and sent people home.

He’s been doing this a long time, and he says the same thing three different ways. The demand is overwhelming, and it’s not letting up.

And listen to why. He doesn’t point at one thing, he points at three. The wildfires, the government shutdown, and prices, food and gas and rent. Three completely separate forces, all landing on the same line of people outside the same building. And that line is going to get longer. I’m pretty sure about that part.

Here’s where I’m supposed to say “and that’s why nonprofits matter” and move on. I’m not going to, because everybody already knows nonprofits matter. It’s roughly 5% of the economy¹ and it’s the thing that catches you when the market and the government don’t. Fine. We agree.

What I actually want to talk about is the part nobody’s saying out loud. The whole sector is being told AI will let it do more with less, and maybe. But the first thing AI actually changes for a nonprofit is how people find it, and that’s where this gets dangerous. AI search is replacing regular search, and the big platforms have all but confirmed that’s the direction. Old search handed you a list, ten options, and you scrolled. AI hands you the answer, one name, maybe three. And the second the result collapses from “here are some orgs working on this” to “give to this one,” a huge number of smaller nonprofits just fall off the map. Not because they got worse. Because they were never in the summary.

That’s the first move in something bigger, and it’s the thing I want to talk about. My read is that before AI helps these orgs, it is going to SORT them. Decide who gets seen, who gets funded, who gets trusted, and the sorting runs straight against the orgs in that food bank line. So yes, we are going to need nonprofits more than ever, and at the very same time we are handing the sector a tool that quietly thins it out first.

I spent a while building a data platform for this world, and writing about why it’s so broken, that’s parts one and two of this series. Which is why I think the AI story everyone keeps repeating has the order of operations backwards.

First, the squeeze

Start with demand, because that part isn’t really up for debate. Three things are pushing it up at the same time, and they don’t have much to do with each other.

AI and automation are coming for jobs. Forrester puts it around 10 million US jobs gone by 2030.² You can argue with the exact number, the direction is the point, and every one of those people is somebody who might need help next year that they didn’t need this year.

Weather is getting more expensive and more frequent. Back in the 80s the country averaged a billion-dollar disaster every 82 days. The last few years it’s been closer to every two weeks.³ And disasters are exactly when nonprofits show up first and the government shows up later.

And government money is getting pulled back. About a third of the nonprofits that actually run services lost some government funding in early 2025, and plenty of them weren’t sitting around waiting to see how it ends, 29% had already cut staff and 21% were already serving fewer people by the time anyone asked.⁴

That LA food bank from the top of the page is not one of these three. It’s all three at once, standing in the same line.

Now look at where the money is going, because it’s going the other way. Total giving looks fine if you only read the headline number, but underneath it the donor base is quietly falling apart. In 2000 about two thirds of American adults gave to charity. By 2020 it was under half.⁵ The dollars held up only because the people who still give are writing bigger checks, so you end up with a sector funded by fewer and richer donors, which is exactly the kind of funding base that snaps the moment a few of them change their mind.

And don’t assume the people at the very top are quietly covering the difference. The Giving Pledge, the whole billionaires-promise-to-give-most-of-it-away thing, is losing steam. Fifteen years in, the share of US billionaires who’ve even signed it has gone down instead of up, last year just four new people joined, and by one count exactly one living couple has actually finished giving away what they promised. Meanwhile billionaire wealth has more than doubled over the past decade. The money at the very top is growing a lot faster than it’s getting given away.⁶

And trust, the one thing this whole sector actually runs on, is falling faster than trust in almost any other institution we have.⁷

So that’s the setup. More people needing help, fewer people paying for it, and a trust problem sitting on top of both. When you get squeezed like that you really only have two moves, serve fewer people or get a lot more efficient, and nobody in this world is going to raise their hand for serving fewer people. So they reach for efficiency, and right now efficiency has a name, it’s AI. Do more with less. Which is where I get off the train, because I don’t think AI is about to hand this sector efficiency. I think it’s about to hand it a sorting machine. Here’s how, in order.

Who the machine can see

So back to that AI answer. The obvious question is, how does it decide who to name? And mostly it works a lot like Google always has. It ranks. It leans on what gets repeated across the internet, the blogs, the articles, the ratings, the mentions, and the orgs that show up over and over are the ones it pushes to the top. So when an AI tells you the “best charity for wildfire relief,” what it’s really handing you is the most-written-about, most-repeated, most-online charity for wildfire relief. And that almost always turns out to be the biggest one. Not the most effective one, not the one closest to the people in that line, the biggest.

And here’s the thing worth sitting with. The LA Regional Food Bank is not some tiny operation. It’s big, it’s well known, it has a name and a CEO who gets booked on the news and people whose actual job is talking to reporters. That is a big part of why you saw that clip at all. Now picture the small pantry two towns over, going through the exact same thing, the same overwhelming demand, the same shelves going empty, except no camera crew ever showed up. It doesn’t have a PR team, and it probably can’t afford much of a marketing department either. It posts on Facebook every now and then, maybe emails the people who’ve given before, and that’s about it. Same crisis, no coverage. And an AI picking who to name works exactly like that newsroom did. It surfaces the one that was already big enough to be seen. The small one barely exists.

Here’s a small thing that stuck with me. A while back I was using AI to pressure-test some ideas for a nonprofit data platform, and the AI itself basically told me not to bother, that in a couple of years a platform like that would be in danger, made pointless by how fast AI is moving. It’s a fair worry. I also think it’s at least somewhat wrong, and here’s why.

Most people assume the AI is reading the actual source, the Form 990 every nonprofit files with the IRS. It isn’t. Nobody’s model is sitting there cracking open millions of tax PDFs. When it says anything concrete about a nonprofit, it’s pulling from a handful of platforms that already did that work, the ones that scraped the filings, cleaned them up and made them readable by a machine. So the AI only knows what those few middlemen know. Whatever they don’t cover, or haven’t refreshed in a couple of years, the AI can’t see either. There’s a gate in front of the gate. And that is exactly why I don’t think a good data platform becomes pointless, because that layer, the clean and current data sitting underneath the answer, is the thing almost nobody is building well for this world. The moat was never the model. Everybody has the same model.

So the first thing AI sorts is visibility. And it sorts it upward, toward the orgs that were already big enough to be written about, and already covered well enough to make it into the extract.

What happens when one org wins

Say you get past the visibility problem and your org actually gets seen. The next thing AI decides is who gets the money, and that sort runs the same direction. Using AI well isn’t free. It takes clean data, people who know how to wrangle it, a budget for the tools, and time to sit there and experiment. Big nonprofits and the foundations behind them have all four. The small org running on fumes has none of them. So the efficiency everyone keeps promising flows to the orgs that were already strong, and those orgs then look even better to funders, who already lean toward the big and the safe when things get tight.⁸ AI just pours gas on a fire that was already burning in one direction.

And here’s where I think people’s instincts are actually ahead of the conversation. Look at how angry everyone is getting about concentration everywhere else. There’s a whole genre of stories now about some single company quietly becoming the infrastructure underneath policing, or hiring, or healthcare, the one vendor that ends up shaping decisions about people’s lives at a scale nobody really voted for. You don’t have to follow any particular case to feel the instinct underneath it. We get nervous when one player owns a whole domain, because there’s no exit, no alternative, nobody keeping them honest. The same thing is about to happen quietly in the nonprofit world, and almost nobody is watching for it there. If AI funnels all the disaster-relief money, or all the homelessness money, to the one org it keeps surfacing, that org becomes the only game in town for its cause. The default. Unchecked. With no real pressure to stay good.

And that last part should bother even the people who don’t care about fairness. A field with a lot of organizations in it is basically a portfolio. Lots of approaches, lots of variance, which is exactly how you get both the failures you can learn from and the occasional breakthrough you can copy. Collapse a cause down to one or two dominant orgs and you collapse all that variance with it. The impact regresses to the median. No experimenting at the edges, no competition keeping anyone sharp, no different approaches for different communities, just one big average way of doing it, frozen in place because nothing is pushing on it anymore. Monoculture is efficient right up until the moment it isn’t. We’d be optimizing the safety net for tidiness and quietly trading away the thing that makes it get better.

I’ve watched the opposite of this work, in about the highest-stakes setting there is. In the war back home, one of the things that has helped Ukraine the most is how decentralized everything is. A huge number of small teams making their own calls on the ground, running their own fundraising, sourcing their own gear, deciding fast without waiting for permission from the top. It has been far more effective than the old one-big-organization, top-down way of doing things. Lots of small independent actors beat one centralized one, not in a feel-good way, in a results way. So when I hear “let the AI pick the single winner for each cause,” it reads to me as exactly backwards.

And the sector wants in, badly. Applications from AI-powered nonprofits to one accelerator went from 13 to 122 to 379 in three years. But 84% of them say the thing they need most is funding for the tools and the talent, and nearly half say adopting AI actually raised their costs.⁹ Everybody wants the efficiency, almost nobody can afford it, which is the definition of a sorting machine. And the part I find almost funny, the big-money AI-philanthropy crowd spends a lot of its breath worrying out loud about AI concentrating power at the scale of the whole civilization, while the exact same thing runs unnoticed in their own backyard.¹⁰ AI doesn’t lift all the boats. And a sea with one boat in it isn’t a safety net, it’s a single point of failure with a halo.

When every application is perfect

There’s a quieter version of this that happens inside the funding process itself. Grant-making has always run on a kind of proxy. A nonprofit writes a proposal, the funder reads it, and uses how good the proposal is as a stand-in for how capable the organization probably is. It was never a perfect signal, but it sort of worked, because writing a clear, well-argued, well-researched proposal actually took something. Time, focus, people who knew the work well enough to explain it simply.

AI quietly kills that proxy. When any org can generate a clean, polished, perfectly on-the-rubric proposal in an afternoon, the proposal stops telling you who’s capable and starts telling you who’s good at prompting, and those are very different things. So picture a program officer with a few hundred applications in front of them, and they all read well now, every single one. The signal they used to lean on just went to zero.

And if you want to see where this road ends, look at hiring, because it already happened there, and it’s fucked.

A Wharton professor said it about as plainly as anyone could.¹¹ He’s hired research assistants for fifteen years, and the way he used to find the good ones was simple, he read the cover letters and the people who actually meant it stood out. Now every single one is perfect. Polished, personalized, name-checking his own papers, and by his own count all of the best ones he’s ever gotten showed up in the last twelve months. So he stopped trusting them. He told a reporter he doesn’t really use them anymore, he goes off who his colleagues can vouch for instead. The signal didn’t get sharper, it died, and he fell straight back on the names he already knew.

And it isn’t just one professor’s inbox. Every resume is perfect now, polished and stuffed with the right keywords by the same few tools everybody uses. So companies fought back by automating their side too, AI writes the applications and AI screens them, and a bot decides which humans another bot’s writing gets to meet. What comes out the other end isn’t the best person for the long run, it’s the best person on paper, and those were never the same thing. There’s just a whole layer of software now making sure nobody notices the gap. Grant-making is walking straight into the same wall.

And when the signal goes, people don’t stop deciding. They just decide on something softer instead, whatever feels safe. The name they recognize, the org they funded last year, the one somebody on the board already trusts.

And to be fair, this cuts both ways. AI also lets a tiny org with one overworked director put together a proposal it never would have had the hours to write, and get a shot it wouldn’t have gotten otherwise. That’s real, and it’s good. But step back and look at the whole field at once. When everybody’s writing levels up at the same time, the difference between proposals collapses, and the small org ends up with a better proposal and a worse chance, at the same time.

That’s the third sort, and it might be the sneakiest one. It doesn’t really pick a winner, it quietly blinds the people doing the picking, one perfect application at a time.

When you can’t tell what’s real

There’s one more sort, and it sits underneath all the others. This whole sector runs on trust. People hand money to strangers so those strangers can go help other strangers, and the only thing holding that up is a basic belief that the org on the other end is real and will do what it says. That belief is about to take a beating, and not in the way people expect. Everyone worries about the big AI scam, the one deepfake charity, the fake disaster fund that takes off and makes the news. That will happen, sure. But that’s not the real damage.

The real damage is volume. AI makes everything almost free to produce, at any scale you want. The heartfelt appeal, the glossy impact report, the urgent email, the entire fake organization with a website and a backstory and photos, all of it, infinite, instant, basically free. Donors can’t keep up with that, and nobody can verify anything at the speed this stuff gets generated. And it doesn’t take a scandal to do the work, it takes a thousand small moments of doubt. The email that’s a little too polished. The org you can’t quite confirm is real. The story that might be true. Each one shaves a little off, and after enough of them your default quietly flips, from “this is probably fine” to “this is probably not.”

I wrote a whole piece in this series about fraud, and the point of it was that fraud was never really the rare scandal, it’s an ambient tax everybody pays. This is that, turned up to a level the sector has no tools for. And here’s the cruel part. The one corner of the world that runs entirely on trust is also the one with the least built to defend it, and it’s getting hit with the most trust-corroding technology we’ve ever made, at the exact moment it needs people to give more, not less. The honest little food bank ends up paying the tax for every cheap fake it had nothing to do with.

Trust in nonprofits was already sliding faster than trust in almost anything else, before any of this started.⁷ Something like two thirds of people say they need to trust an organization before they’ll give to it, and fewer than one in five actually do.¹² Now pour an ocean of synthetic everything on top of that. When you genuinely can’t tell what’s real, you do one of two things. You give less, or you give only to the names you already recognize.

The wave and the line

So put the four together. AI decides who gets seen, and it picks the biggest. It decides who gets the money, and it picks the biggest. It quietly takes over how funders choose, and it rewards the most familiar name. And it floods the whole thing with so much fake that people retreat to the names they already know. Four different mechanisms, in four different corners of the sector, all pushing in the exact same direction. Toward the orgs that were already big, already known, already fine. And away from everybody else.

Now hold that next to the story the smart money is telling right now. There’s an essay going around, “The Third Wave of American Philanthropy”, and it’s sharp, you should actually read it. The argument is that AI is about to turn hundreds of billions of dollars into new charitable money, something like $37 to $100 billion a year on top of the $600 billion Americans already give, and that the bottleneck won’t be the money, it’ll be organizations. Her read is that we don’t have nearly enough great ones to absorb what’s coming, so we need to go build thousands of new ones. A Silicon Valley for public goods, run by tech-caliber people, fast and ambitious.

I don’t think she’s wrong about the money. I think she’s got the organization part backwards. We are not short on organizations doing great work. There are thousands and thousands of them already, and a huge number of them are starving for funding right now, today. Do we also need new ones to handle the new things coming at us? Absolutely, we will. But “go build a Silicon Valley of new philanthropic startups” quietly steps right over the thousands that already exist, already work, and just need someone to actually fund them. And it’s worth looking hard at where this new money is built to go, because by its own description, the wave is aimed at new, startup-shaped organizations run by people who come out of tech, and it’s openly skeptical of the slow, unglamorous, traditional parts of the sector. Which means the food banks and the small local groups doing the actual catching are not who this is for.

So here’s the whole picture, pulled all the way back. A historic wave of money is coming in from the top, and by design it routes around the bottom. And at the very same time, AI is quietly erasing the bottom from view. The two things everyone is most excited about, the windfall and the technology, both point away from the people standing in that line.

And look, maybe she’s right that the distribution layer needs rebuilding. The whole setup that’s supposed to move money from the people who have it to the work that actually needs it is clearly not doing its job, and I’m open to the idea that it has to be torn up and redone. There’s even a fix sitting right there underneath all of it, the unglamorous part nobody wants. Clean, current, open data about who these organizations actually are and what they actually do, so the machine has something true to work with instead of just ranking whoever is loudest. But it’s slow, there’s no demo day in it, and none of the money that’s coming is pointed at it. So mostly, it won’t get built. It’s just not sexy enough. So I’m building it. That’s Vianido, the boring layer underneath everything that everyone else is too busy chasing demo days to touch. I think it’s the part that actually decides how all of this plays out, and I plan to be the one who gets it right.

And that’s the part the whole third-wave pitch keeps stepping over. You cannot build a new distribution layer on top of a broken one. If the data underneath stays fragmented and stale and easy to game, then every shiny new institution you stack on top doesn’t fix the sorting, it just industrializes it. You move the money wrong, faster, and at a bigger scale than before.

Meanwhile, the line outside that food bank in LA is still there. It’s going to be longer next year, and longer the year after that. We are going to need these organizations more than we have in a long time, and we are handing the whole sector a machine that quietly decides, at scale, which of them we ever hear about again.

“Roughly 5%” refers to the nonprofit sector’s share of US GDP. Independent Sector and the Bureau of Economic Analysis put nonprofits at about 5.6% of GDP (roughly $1.5 trillion), and the sector employs close to 10% of the private workforce. https://independentsector.org/resource/health-of-the-u-s-nonprofit-sector/ ↩
Forrester, “AI and Automation Will Take 6% of US Jobs by 2030” (2025): projects roughly 10.4 million US jobs lost to AI and automation by 2030. https://www.forrester.com/blogs/ai-and-automation-will-take-6-of-us-jobs-by-2030/ ↩
NOAA National Centers for Environmental Information and Climate Central: US billion-dollar weather and climate disasters averaged one every ~82 days in the 1980s, versus roughly one every two weeks across 2016-2025 (about every 10 days in 2025). https://www.ncei.noaa.gov/access/billions/ ↩
Urban Institute, via The Chronicle of Philanthropy, “A Third of U.S. Nonprofits That Serve Communities Lost Government Funding in Early 2025”: survey of 2,737 nonprofits; of those hit by disruptions, 29% had reduced staff and 21% were already serving fewer people. https://www.philanthropy.com/news/a-third-of-u-s-nonprofits-that-serve-communities-lost-government-funding-in-early-2025/ ↩
Lilly Family School of Philanthropy data, via The Chronicle of Philanthropy, “Donors Are Down, but Dollars Are Up”: share of US adults who give to charity fell from ~66% in 2000 to ~46% in 2020. https://www.philanthropy.com/news/donors-are-down-but-dollars-are-up-how-u-s-charitable-giving-is-changing/ ↩
CNBC, “After 15 Years, the Giving Pledge Yields Mixed Results” (Aug 2025), citing the Institute for Policy Studies report “The Giving Pledge at 15”: signatories now represent ~12% of US billionaires, down from ~14% in 2010; just four new signers in 2024; IPS identifies one living couple as having fulfilled the pledge; billionaire wealth has more than doubled over the decade. https://www.cnbc.com/2025/08/07/giving-pledge-buffett-gates-billionaires-philanthropy.html ↩
The Chronicle of Philanthropy, “Financial Climate and Trust Issues Trouble Nonprofit Sector, Report Finds”: trust in nonprofits has fallen faster than trust in any other institution (down 7 points since 2020). https://www.philanthropy.com/news/financial-climate-and-trust-issues-trouble-nonprofit-sector-report-finds/ ↩ ↩²
Urban Institute, via The Chronicle of Philanthropy, “Emergency Funding Takes On New Urgency Amid Government Shutdown”: in funding crises, funders gravitate toward larger organizations, leaving smaller nonprofits effectively shut out. https://www.philanthropy.com/news/emergency-funding-takes-on-new-urgency-amid-government-shutdown/ ↩
Fast Forward accelerator data, via The Chronicle of Philanthropy, “Nonprofits Are Embracing AI, but Many Struggle to Find Funding”: AI-powered applicants rose from 13 (2024) to 122 (2025) to 379 (2026); 84% cite funding for tools and talent as their top need; nearly half say adopting AI raised their costs. https://www.philanthropy.com/news/nonprofits-are-embracing-ai-but-many-struggle-to-find-funding/ ↩
George Rosenfeld (Coefficient Giving), “A personal letter on transformative AI,” Multiplier (May 2026): names “extreme concentration of power” among the top institutional risks of advanced AI. https://multipliercg.substack.com/p/a-personal-letter-on-transformative ↩
Ana Altchek, “RIP cover letters,” Business Insider (June 1, 2026): Wharton economist Judd Kessler reports AI-generated cover letters have become uniformly polished, with “all of the best cover letters” arriving in the last 12 months, leaving him to rely on faculty recommendations and referrals instead; Google, Amazon, McKinsey, BCG, and Cisco no longer require cover letters. https://www.businessinsider.com/rip-cover-letters-generative-ai-hiring-2026-6 ↩
Give.org (BBB Wise Giving Alliance), Donor Trust Report 2026: about 67.7% of Americans say trust is essential before giving, while only ~18.3% report high trust in charities. https://www.give.org/ ↩