GEO for Nonprofits: When AI Recommends Where to Donate
Donors ask AI now. Your homepage isn't ready.
If a donor asks ChatGPT "where should I give this month," the answer names organizations — and none of the six nonprofits we checked on August 14, 2026 gave AI models anything clean to quote. We fetched the homepages of the American Red Cross, WWF, Doctors Without Borders, Charity: Water, Feeding America, and Habitat for Humanity. Then we checked what a crawler sees: structured data, robots.txt rules for AI agents, and an llms.txt file. Zero of six published an llms.txt. Zero named GPTBot, ClaudeBot, or PerplexityBot in robots.txt. Three — Red Cross, Charity: Water, and Habitat — shipped no structured data at all.
That pattern shows up across the 200+ domains we've audited since April. The average site scores 45/100. Nonprofit sites usually land below that line, because the team runs on donation cycles, not SEO budgets. You don't need a budget. You need six fixes, and all of them are free.
What a crawler found on six nonprofit homepages
We ran each homepage through the same checks as our AI crawler check — raw HTML, robots.txt, the llms.txt path. Here's what came back:
| Organization | Structured data on homepage | AI crawler rules | llms.txt |
| American Red Cross | none | none | missing |
| WWF | Organization, PostalAddress | none | missing |
| Doctors Without Borders | WebPage, Organization | none | missing |
| Charity: Water | none | none | missing |
| Feeding America | Organization, WebSite, SearchAction | none | missing |
| Habitat for Humanity | none | none | missing |
additionalType: "NGO". Nobody used NonprofitType, the schema.org type built for exactly this situation. And even the best-structured page has no llms.txt, so a crawler landing on the homepage has to guess where the impact numbers live.Real talk: none of this is a conspiracy, and none of it is a budget problem. Nonprofit web teams ship what donors see in a browser — hero images, donation buttons, mission paragraphs. Nobody audits what an LLM sees, because until last year nobody had to.
Why the donor question changed
Donors stopped typing "charity + keyword" and started asking for recommendations — "which charities fight food insecurity," "where can I volunteer," "is my donation tax deductible." LLMs answer those questions with named organizations. The second shift is subtler: the model builds its answer from whatever it can lift, and it lifts best from self-contained passages of roughly 134-167 words that answer one question directly, plus structured data that states facts outright.
Nonprofit pages answer "what we do" beautifully and "what will my $50 buy" rarely. Mission statements don't quote well. Impact numbers live in 40-page PDF annual reports. Overhead ratios, program spending, and per-dollar impact — the exact facts donors ask about — sit in documents AI crawlers mostly skip, or buried so deep the model can't attribute them.
The 2025 Ahrefs study of 75,000 brands adds the kicker: media and YouTube mentions drive AI citations, while backlinks barely register. An organization mentioned in the news gets cited. But your own site still has to hold up its end — schema, answer-shaped passages, crawler access — or the model quotes the Wikipedia article instead of you.
The fix, in the order we'd do it
- Publish an llms.txt file. One text file naming your key pages — donate, impact, annual report, programs. The llms.txt checker validates the format in seconds. Zero of the six nonprofits we checked had one, so this is the fastest gap to close.
- Add Organization schema with the NGO type. Name, url, logo, sameAs (Wikipedia, LinkedIn, Charity Navigator), contactPoint, and a donate action. Prefer NonprofitType if your CMS allows it;
additionalType: "NGO"— the Feeding America approach — is the fallback. The schema checker catches syntax errors before you ship. - Put donor answers in plain text, early on the page. "89 cents of every dollar goes to programs. A $50 gift feeds a family of four for a week." A quotable block in the first 134-167 words gets cited. The same sentence inside a PDF doesn't.
- Confirm AI crawlers can reach you. GPTBot, ClaudeBot, PerplexityBot, and eleven more default to allowed unless your robots.txt blocks them. None of the six sites we checked blocked them — and none welcomed them, either. The crawler check tests all 14 user-agents on your domain.
- Write a donate FAQ page. What a gift does, overhead, recurring giving, tax receipts — the questions donors actually type. In our audits, pages with FAQPage schema get cited at roughly twice the rate of pages without it.
The donate page is a citation magnet
Donor questions are recommendation questions, and recommendations name organizations. Every month, thousands of those questions get answered by models that can't read your impact report PDF. The org that publishes an llms.txt, types itself as a nonprofit, and answers "what does my money do" in plain text becomes the safe citation — the one the model quotes instead of a random listicle.
Run the free GEO audit on your donate page today. You'll get the six-dimension score, the schema report, and a fix list ordered by impact. For the mechanics behind it all, how AI search engines cite websites is the deeper read.