Are AI tools recommending your charity?
Picture someone who has just received a difficult diagnosis, or who has watched a humanitarian crisis unfold on the news. They want help, or they want to help, so they pick up their phone.
Ten years ago they would have typed a question into Google and scanned a page of results. More and more, they now ask ChatGPT or Claude and get a single answer back. Even if they are still going to Google, they are increasingly likely to rely on the AI summary at the top of the page.
How AI summaries can impact charity recommendations
AI summary answers carry a lot of weight, and many people won't look any further. AI assistants may not be the biggest driver of charitable decisions yet, but they are an increasingly influential factor in which charities people consider – and their importance is only set to grow. It’s also probably fair to say that most organisations don’t know how they come across in them.
There are knock-on effects as well. Stories from across the marketing world suggest paid search is delivering less than it used to. When people get what they need from the AI summary at the top of the page, fewer of them click through, so the same budget buys less traffic. If paid search has been a reliable channel for you, keep a close eye on it.
nfpResearch’s next step to help charities understand AI tools
We wanted to give charities a proper picture of how they're represented across the AI tools the public actually uses, so nfpResearch designed a benchmark. We wrote around 60 prompts covering the kinds of questions donors, beneficiaries and people looking for information might ask. We also added some deliberately tricky ones designed to draw out criticism.
We sent each prompt to five assistants (Claude, ChatGPT, Gemini, Grok, and Perplexity) eight times to see how consistent the answers were. That gave us more than 4,000 responses, all of which we coded.
Results from our AI prompt benchmark pilot
We piloted the approach with UK overseas aid charities and found an array of valuable – and, at times, surprising – results.
The big names get mentioned a lot, but they don't get recommended as often. Save the Children and Oxfam come up constantly. Yet the charity recommended most often was the Against Malaria Foundation, which most people outside the sector have probably never heard of. Its focus on measurable cost-effectiveness matches the effective altruist thinking that seems to have shaped many of these models.
Results also vary widely between assistants. CARE International does well on Gemini, Grok and Perplexity and poorly on ChatGPT and Claude. Oxfam shows the opposite pattern. Criticism of charities came up more often than we expected, sometimes with no source given at all.
The sourcing findings were the most sobering. When an assistant did say where its information came from, it was rarely the charity's own website. Wikipedia and GiveWell dominated, and only 24% of citations pointed to charities' own content.
What charities can do to show up in AI tools
Charities can influence all of this. Three things stand out:
Pay attention to the third-party sources AI relies on, such as Wikipedia and GiveWell, and make sure what they say about you is accurate.
Publish evidence in a form AI can easily cite: clear figures, plain statements of impact and dated reports, as well as your more persuasive content. As Google said earlier this year: “good SEO is good GEO”.
Make sure up-to-date information about your charity is easier to find than older material.
How to benchmark charity AI results
We're offering our benchmarking process as a bespoke audit, at £3,000 + VAT for the first five clients.
It covers how often you come up compared with charities you choose to benchmark against, a scorecard showing how often you're mentioned and how often you're recommended, and a prioritised list of what to fix.
You can watch the full webinar recording to see the data in more detail. If you'd like to talk about running this for your charity or your part of the sector, email Cian Murphy (cian.murphy@nfpresearch.com) or Bijal Rama (bijal.rama@nfpresearch.com).