How to check whether ChatGPT, Claude, Gemini or Perplexity recommends your charity

To check whether an AI assistant recommends your charity, ask it the questions a donor would ask, in a clean session, several times each, and write down every answer. Do this for ChatGPT, Claude, Gemini and Perplexity separately. The most useful questions do not name your charity, because a donor who already knows your name is not the one you are trying to reach.

This is a quick check you can run in an afternoon. It is not research, and one afternoon cannot tell you how often you are recommended. Citewake's own charity research programme starts in February 2027 and will use a fixed question set and a published method. The research agenda and categories pages show what it covers.

Which questions should you ask?

Write them the way a donor would type them. Use three groups.

• Cause questions. "Which charities help [your cause]?" and "Where should I donate to help [your cause]?"
• Place questions. "Which charities help [your cause] in [your city or region]?" and "Local charities for [your cause] near [place]."
• Check questions. "Is [charity name] a good charity to donate to?" and "Is [charity name] legitimate?"

Cause and place questions show whether you are recommended at all. Check questions show what the assistant says once someone does ask about you. Candid used prompts of that second kind ("Should I support [name] nonprofit?") when it tested free AI tools on 11 nonprofits (Candid, August 2025).

Keep the list short, about ten questions, and keep the wording fixed so you can ask the same ones again next quarter.

Why test signed out and in a private window?

Your own account can change what you see. Past conversations, saved memory and settings can all colour an answer, and you have probably used these assistants to talk about your own charity. A private browser window, signed out, is closer to what a new visitor sees. Some assistants limit what signed-out users can do, so note any limit you meet.

Even a clean session does not stand in for every donor. Location, language and the device can matter. Our guide on how to test AI answers properly covers this in more detail.

How many times should you ask?

More than once. Answers change between runs for the same question, which is why one answer proves little (why AI answers change every time you ask). For a quick check, ask each question five times per assistant, in fresh chats. Count the runs in which your charity appears. "Named in two of five runs" is a more honest statement than "it recommended us" or "it ignored us".

What should you record?

A simple sheet is enough. For every run, write down:

• the date;
• the assistant and whether it was searching the web;
• the exact question;
• the full answer, saved as text or a screenshot;
• whether your charity was named, and whether it was recommended or only mentioned;
• which charities were named instead;
• any sources the assistant listed.

The difference between being mentioned and being recommended matters. See mentioned, recommended or cited.

How do you find out where an answer came from?

Ask. Candid suggests adding "Provide links to where I can find this information" to a prompt to see the sources an assistant is using, and turning on the deep research feature to see what automated research can surface about your organisation (same Candid article). Where the assistant shows citations itself, copy them.

Then open the cited pages. Often they are a register entry, a profile on an evaluator's site, a news story or your own website. Those pages are the ones you can act on. What AI assistants cite when they recommend a charity explains the common ones.

What do you do with what you find?

Read the answers for three things.

1. Are you present? If you are never named, look at who is, and at what those charities have that you do not. Do not copy their wording.
2. Is it accurate? Check your income, location, mission and legal status against your official register entry. Errors usually trace to an old page somewhere.
3. Is it consistent across assistants? Different assistants often disagree. See how to compare answers across ChatGPT, Claude, Gemini and Perplexity.

Repeat the same test in three months. A single test is a snapshot. A repeated test starts to show a trend, and it keeps you honest about change that may only be random variation.

If you want a quick first look at how assistants describe your site, the Citewake checker is free. Verified charities can also apply to the Citewake Foundation.