The big picture: We asked AI models from OpenAI, Anthropic, and Perplexity 12,000 questions about nonprofits in U.S. cities. If you ask about a small nonprofit by name with web search on, then AI describes it accurately. Ask which local nonprofits to support, however, and we found that AI strongly favors the large ones.
People used to search “food banks in Denver” in a search engine and scroll through ten links to find information. Now, people are asking AI models like Claude, ChatGPT, and Perplexity and getting back a short list of organization names.
A nonprofit that shows up on that AI-generated list might get a new participant, donor, or volunteer. A nonprofit that’s left off, however, never realizes the question was even asked.
So we got curious: Which local nonprofits make the cut on these AI-generated lists? Is there a pattern, and is it equitable? And is there anything small nonprofits can do about it?
Here’s what we found (and how).
Key findings: Nonprofit AI visibility statistics (2026)
- When asked for nonprofit recommendations in a particular city, AI named local nonprofits with $20 million or more in annual revenue at 3.7 times their proportional share.
- Nonprofits with $100,000 to $1 million in annual revenue make up 66% of local nonprofits but only received 32% of AI recommendations.
- While the median local nonprofit in this study had $578,000 in annual revenue, the median nonprofit recommended by AI had $2.5 million in annual revenue (over 4x more).
- The bias for recommending $20 million+ orgs held steady across all three AI platforms: 3.3x for OpenAI, 3.9x for Claude (Anthropic), and 4.2x for Perplexity.
- Without web search enabled, Claude and Perplexity knew something specific about the largest nonprofits more than three times as often as the smallest (74% vs. 20% for Claude, and 71% vs. 23% for Perplexity).
- Without web search enabled, Claude said it didn’t know 58% of the smallest nonprofits, compared with 11% of the largest.
- Without web search enabled, Perplexity’s answers contradicted the nonprofit’s own website for 17% of the smallest nonprofits, compared with only 5% of the largest.
- With web search enabled and when asked about a specific organization by name, all three platforms identified the correct nonprofit 98–100% of the time, at every organization size.
- Across all nonprofits, 31% of nonprofit homepages don’t clearly state all three key details: what the organization does, who it serves, and where it’s located.
Cite this study: Pixel Lighthouse, “2026 Nonprofit AI Visibility Study,” October 2026, https://pixellighthouse.com/nonprofit-ai-visibility-study/.
What we did to find out which nonprofits get recommended by AI
To run this study, we asked three AI platforms 500 open questions like, “What are good food nonprofits in [city]?” Then we matched the 14,540 names they gave against IRS records and compared them with every eligible nonprofit within 20 miles.
We also asked about 1,700 specific nonprofits by name (with and without web search enabled), and checked each answer against the organization’s own homepage to compare.
The following points outline the major takeaways from the data.
Important note: We excluded big-name national organizations from our analysis, so this is a comparison of local nonprofits against other local nonprofits. The national organizations AI named (such as the Red Cross, Feeding America, the Salvation Army, and United Way) are among the largest in the country, so including them would likely make the skew even larger than what our data showed.
Finding #1: AI recommends the largest local nonprofits at nearly 4x their share
Among local nonprofits with at least $100K in revenue, two out of three bring in under $1 million a year. Despite making up the majority of nonprofits, they only received one in three AI recommendations.
The largest nonprofits took the difference. Organizations with $20 million or more in annual revenue represent 4% of local nonprofits and yet received 16% of AI recommendations.
Revenue | Share of local nonprofits | Share of AI recommendations | AI preference ratio |
|---|---|---|---|
$100K–$1M | 66% | 32% | 0.5x |
$1M–$5M | 22% | 32% | 1.5x |
$5M–$20M | 8% | 21% | 2.5x |
$20M+ | 4% | 16% | 3.7x |
The median nonprofit recommended by AI had $2.5 million in annual revenue. The median local nonprofit, however, had $578,000, meaning that AI answers strongly skew toward the larger orgs.
Every platform showed the same bias. The $20M+ AI preference ratio was 3.3x for OpenAI, 3.9x for Claude, and 4.2x for Perplexity.
Looking for possible alternative hypotheses, we checked whether something other than size could explain the difference in AI recommendations:
- Cause: Some causes do have bigger nonprofits than others. Comparing each recommended nonprofit only with local nonprofits in the same focus area, however, still revealed a 3.1x bias in favor of large orgs.
- Matching errors in the data: Dropping uncertain organization matches leaves the $20M+ group still receiving 15.9% of AI mentions. Using only exact name matches resulted in a slightly lowered 14.9%.
- Random chance: Even at the low end of the 95% confidence interval, the $20M+ group comes in at 14% of all AI recommendations, still more than three times the 4% baseline you’d expect based on the number of orgs of this size.
We also tested it the other way around: for 564 nonprofits in our sample, we asked for recommendations in their own city and cause area, then checked whether the organization was named by AI. The smallest orgs were named 12–27% of the time, depending on the platform. The largest were named in 31–44% of responses. However, only 77 nonprofits were in the largest group, so we treat this as supporting evidence rather than a headline statistic.
Finding #2: AI knows far more about the largest nonprofits than the smallest
So why does AI favor bigger nonprofits in its mentions? To be recommended, an organization has to be either already known to the AI or findable through a web search. We looked at the first of those criteria (what AI “knows” on its own) by turning web search off and asking each AI a simple question about every nonprofit in our sample: “What does [nonprofit] in [city, state] do? What is its mission?” Without search, AI can only answer from what it learned in training.
Then we checked each answer against the nonprofit’s own homepage. We counted an answer as “knowledgeable” when it included at least one specific detail the homepage confirms, such as a program, a population served, or a service. The grader was instructed not to count details that could be guessed from the name alone.
Perhaps unsurprisingly, the bigger the nonprofit, the more AI knew. Without web search, Claude and Perplexity identified something specific about the largest nonprofits more than three times as often as the smallest.
How often AI knew something specific about a nonprofit (by revenue, with web search disabled)
Platform | $100K–$1M | $1M–$5M | $5M–$20M | $20M+ |
|---|---|---|---|---|
Claude | 20% | 40% | 54% | 74% |
OpenAI | 53%* | 71% | 83% | 95% |
Perplexity | 23% | 40% | 54% | 71% |
*Note: A hand-check of a sample of OpenAI’s answers suggests the grader was generous marking responses as knowledgeable for the smallest nonprofits, so OpenAI’s true rate there is likely well below 53%. The rise in knowledge with size holds for all three AI platforms regardless. See Limitations paragraph below.
When presented with an unknown organization, we found that the platforms handle the ambiguity differently.
Claude openly admits the uncertainty, saying it didn’t know 58% of the smallest nonprofits, compared with 11% of the largest.
Perplexity, on the other hand, tended to guess (sometimes incorrectly). For 17% of the smallest nonprofits, Perplexity’s description contradicted the organization’s own website. For the largest orgs, the contradiction rate fell to 5%.
Much like a person might, the wrong answers supplied by the platforms weren’t randomly generated. They were plausible missions based on a reasonable guess from the organization’s name.
A few incorrectly guessed mission examples:
- A fly-fishing retreat program for men with cancer was described as a volunteer-pilot medical air service.
- A youth breakdancing program was described as a disability services nonprofit.
- A youth tennis and tutoring program was described as a housing provider for people experiencing homelessness.
A donor reading answers like these from a confident-sounding AI would have no reason to doubt the accuracy of the mission description.
Finding #3: Ask about a small nonprofit by name, and AI gets it right
So what happens on the second path, when AI can search the web? When asked about a specific nonprofit by name with web search enabled, organization size makes almost no difference.
All three AI platforms identified the correct organization 98–100% of the time, across every revenue band. They also described an organization’s mission accurately 94–100% of the time. The answer quality was also stable when repeated. We asked about 100 nonprofits a second time to check answer stability; 297 of 300 repeat answers received the same identity grade.
So AI can identify and describe even the smallest nonprofit well. But that accuracy only helps someone who already knows the nonprofit’s name.
A donor asking for general recommendations won’t hear about many worthy orgs local to them. Neither will a family looking for help, or a volunteer looking for a place to spend an afternoon.
Notably, the recommendation answers in Finding #1 also used web search, and the skew toward large nonprofits held anyway. Search makes AI accurate about a small nonprofit once it’s named, but in our data it didn’t stop AI from favoring large ones.
Finding #4: Small nonprofits’ homepages give AI less to work with
If AI can describe small nonprofits accurately but rarely brings them up on its own, part of the answer may be in what those nonprofits tell it.
We used one AI model (Anthropic’s Claude Haiku) to read each homepage and judge whether it plainly states three things: what the organization does, who it serves, and where the organization is located. The same model read every homepage, so the comparison across sizes is consistent.
Homepage states… | $100K–$1M | $1M–$5M | $5M–$20M | $20M+ |
|---|---|---|---|---|
What they do | 96% | 97% | 98% | 98% |
Who they serve | 80% | 83% | 85% | 88% |
Where they’re located | 82% | 88% | 90% | 93% |
All three | 66% | 74% | 77% | 82% |
As shown in the table above, nearly every nonprofit says what it does. Fewer nonprofits say where and for whom.
Across all nonprofits (weighted by size), 69% of homepages clearly state all three. That means 31% leave at least one unclear or unstated.
For OpenAI, directories such as ProPublica’s Nonprofit Explorer and Charity Navigator appear to fill in some of the gap: a nonprofit directory was the first source cited for 15% of the smallest nonprofits, compared with 4% of the largest. Claude and Perplexity didn’t show this pattern, however.
The platforms also differ in how often they send people to the nonprofit’s own site. OpenAI cited an org’s site in 92–98% of answers and Claude in 75–83% of responses. Perplexity cited it only about 60% of the time, mostly because Perplexity often cited no source at all.
Why does AI favor larger nonprofits?
This study reveals there’s a disparity in AI recommendations between large and small organizations. However, it doesn’t prove the cause.
The likeliest explanation is how much each nonprofit shows up across the web. Larger organizations get more news coverage, more partner links, more directory listings, as well as many more mentions. AI learns from all of it, and searches through those references when it generates answers.
The homepage findings (Finding #4) point in the same general direction. Small nonprofits often give AI less to work with on their own sites in general, so AI leans more on what others say about the organization.
What can small nonprofits do to show up in AI search?
A small nonprofit can’t outspend a large one, realistically. However, even the smallest nonprofits do have at least one highly effective strategy available: making themselves easier to describe by others.
- Say who you serve and where you’re located on your homepage: If you do one thing, this should be it, as it’s a very light lift. Ideally, this information would be repeated on several pages, such as your About page. Use regular, straightforward language a stranger would understand.
- Keep your directory profiles current: Candid, Charity Navigator, and ProPublica’s Nonprofit Explorer are among the directories AI cites. ProPublica in particular shows up more often for small nonprofits.
- Get named wherever your work happens: Local news, partner websites, and community calendars all tell AI that your organization exists and contribute helpful details. Mentions on other websites also act as endorsements that may help an organization get referenced.
As a bonus, all of the above strategies will improve a regular (human) visitor’s understanding of a nonprofit’s work, too. AI models learn from human-written content, and anything that makes your mission or focus clearer will help people and robots alike.
Methodology
We collected data from October 18th to October 21st, 2026, through the APIs for three models: OpenAI’s gpt-5.5, Anthropic’s claude-sonnet-5-5, and Perplexity’s sonar. A fourth model (Anthropic’s Claude Haiku) handled supporting tasks: reviewing homepage text (Finding #4), pulling organization names out of the recommendation answers, and helping match those names to IRS records. In total, we analyzed 12,000 answers.
The panel: We drew 1,700 U.S. nonprofits from IRS Business Master File records, sampled across four revenue bands: $100K–$1M, $1M–$5M, $5M–$20M, and $20M+.
We posed three kinds of questions:
- Without search: We asked each platform about each nonprofit by name with web search set to “off.”
- With search: We asked the same question with web search turned on.
- Discovery: We asked 500 open questions in the form, “What are good [cause] nonprofits in [city]?” across 397 cities.
Grading: Claude graded all three platforms’ answers against each nonprofit’s publicly visible homepage text. An answer counted as “knowledgeable” (Finding #2) only if the nonprofit’s homepage had enough readable text to check it against. We asked about 100 nonprofits twice to check grading consistency; 297 of 300 repeat answers received the same grade, showing high consistency. The grader was Claude Sonnet (claude-sonnet-5-5), the same model tested as the Claude platform. Links were removed from answers before grading. Claude did not grade its own answers more favorably: for the smallest nonprofits, it rated 20% of its own answers as knowledgeable, compared with 53% for OpenAI’s and 23% for Perplexity’s. For nonprofits without usable homepage text, answers were checked for contradictions against IRS records instead.
Homepage review: Claude Haiku (claude-haiku-5-5) read each nonprofit’s homepage text and judged whether it stated what the organization does, who it serves, and where it is located. Homepages with fewer than 50 words (166 in all) were excluded, and a detail counted only if it was stated explicitly.
Matching recommendations: The platforms gave 14,540 organization names across all responses. Claude Haiku extracted these names from each answer, and we matched them to IRS records by exact name where possible, with Haiku judging the closest candidates where there was no exact match. 20% weren’t nonprofits (government agencies, schools, businesses, congregations), and we excluded 614 organizations presented as national (e.g., the Red Cross). We matched 69% of the remaining nonprofits to IRS records. Based on their names, the 31% we couldn’t match included both small and large organizations. Eligible nonprofits had at least $100K in annual revenue and were not hospitals, schools, congregations, or private foundations. We compared the 5,560 mentions of matched, eligible nonprofits (3,575 distinct organizations, since some were named more than once) with all eligible nonprofits within 20 miles of each city. “Recommended” means that the organization was named in the answer.
Weighting: Results are shown by revenue band and are unweighted unless stated. Figures described as “across all nonprofits” are weighted to the actual mix of nonprofit sizes in the United States.
Limitations: API results can differ from what a person sees in each consumer app. AI answers change as models update, so these results describe October 2026. We hand-checked 20 randomly selected answers the grader counted as knowledgeable for OpenAI’s smallest nonprofits. Only 7 contained details that couldn’t be guessed from the organization’s name and location, which suggests OpenAI’s true rate for small nonprofits is well below the 53% shown. The same grader scored every size band, so the rise in knowledge with size holds. Our $20M+ panel group is also relatively small for the panel-named measure (comprising only 77 nonprofits).
Cite or share this study
Citation: Pixel Lighthouse, “2026 Nonprofit AI Visibility Study,” October 2026, https://pixellighthouse.com/nonprofit-ai-visibility-study/.
Media contact: Austin Hattox, Principal, Pixel Lighthouse
Email: austin@pixellighthouse.com
Frequently asked questions
Does AI favor large nonprofits?
In our study, yes. When asked to recommend local nonprofits, AI named local organizations with $20 million or more in annual revenue at 3.7 times their share. Local nonprofits under $1 million got about half of their share.
How accurate is AI about small nonprofits?
It depends on search capabilities. With web search enabled and when asked about them by name, AI identified small nonprofits correctly 98% of the time or more. Without it, Claude said it didn’t know most of them, and Perplexity contradicted the organization’s own website 17% of the time.
Why does AI recommend larger nonprofits?
This study doesn’t prove a cause. Larger nonprofits likely appear in more news coverage, partner sites, and directories, which gives AI more to learn from and find.
Does turning on web search fix it?
Not in our data. The recommendation answers used web search, and the skew toward large nonprofits still held. Search helps AI describe a small nonprofit accurately once it’s named, but it didn’t prevent the skew in which nonprofits got named.
What can a small nonprofit do?
State on your homepage what you do, who you serve, and where you’re located. Keep your directory profiles updated and current. Then ask an AI for recommendations in your city and cause area, and see whether your organization is named.
How was the study done?
We analyzed 12,000 answers from three AI platforms, plus 1,534 homepage reviews, collected in October 2026. See the methodology above.