Can I see donations that came from ChatGPT or Perplexity?

Sometimes. If the visitor clicked a link and their browser passed on where they came from, your analytics can show the referring site. Many visits from AI cannot be traced this way.

Do I need consent to measure this?

That depends on the law where you and your donors are, and on the tools you use. Where consent is required, measure only visitors who agreed, and say clearly what you record.

How big is AI traffic to charity sites?

Small in the one study we found that measured it: 2% of total volume across 24 nonprofit websites from January to October 2025, up about 1,000% on the year before.

What else can I ask donors?

A single optional question on the donation form, such as how they first heard of you, with AI assistant as one of the choices. It captures what analytics miss.

Measuring AI referrals and donations on a charity website

A charity can see some visits that come from AI assistants, but not all of them, and it can link some of those visits to donations. The measurable part is visits where the visitor clicked a link in an assistant and their browser reported where they came from. The rest, including people who read an answer and typed your name later, shows up as direct or search traffic, or does not show up at all.

Citewake has no charity data yet, and this guide does not claim any. It describes what analytics can and cannot tell you, and quotes published figures with sources. For how recommendation itself is checked, see how to check whether an AI recommends your charity.

What do published figures say about AI traffic to charities?

Little, and from small samples. Blue State's report on brand discovery, as summarised by Candid, looked at 24 nonprofit websites from January to October 2025. AI-driven traffic made up 2% of total volume, about 1,000% more than a year earlier. Those visitors stayed more than 70% longer than other visitors, and their average donation rate was about one-eighth of the average across all traffic sources. Among those who gave, the average gift was $250 against $210 for donors from non-AI search sources (Candid, December 2025).

Treat those as a snapshot of 24 sites in one period. Your numbers will differ, and with small charities they will often be too small to read.

What can your analytics attribute?

Look in your reports for visits whose source is an AI assistant's website or app. These are visits where the visitor clicked a link in an answer and the browser passed the referring site on. What you can see depends on your analytics tool, the assistant, the app the person used and their browser settings, so check your own reports instead of assuming.

You can also see whether those visitors went on to a donation page and whether they completed a gift. For a small charity the counts may be a handful a month. Report counts, not percentages, when the numbers are that low.

What can you not attribute?

Several common cases leave no trace.

A name typed later. A donor reads an answer, remembers your name and searches for you. It appears as search or direct traffic.
Missing referrer information. Some clicks arrive with no source, and tools file them as direct.
Answers read, never clicked. Candid notes that donors may research charities through AI tools and never reach a charity's website, and suggests new measures: how often the organisation is cited in AI-generated answers, and where AI-referred traffic originates (Candid, September 2026). That article is a consultant's opinion piece, not a study.
A decision made elsewhere. A donor may give through a platform or by bank transfer, where your site never sees them.

So a low AI figure in your reports is a minimum. The true figure is higher by an amount you cannot see.

How do you measure without breaking donors' trust?

Use consented analytics only. That means you record visitor behaviour only where the person has agreed, or where your tool and your law allow measurement without personal data, and you say so in your privacy notice. The rules differ by country, so take advice that fits yours.

In practice:

Choose an analytics tool that does not need to follow individuals around the web.
Record the source of a visit and the page reached. You do not need a person's identity to count visits.
Do not try to identify individual donors from assistant traffic.
Keep what you collect to what you will use.

What else can tell you more?

Ask. Add one optional question to your donation form or thank-you page: "How did you first hear about us?" with options that include "an AI assistant such as ChatGPT". A donor's own answer catches what the analytics miss, including the name typed later. It is consented by design, because the donor chose to tell you.

Combine this with the repeated question test in the check guide. If the assistants start naming you and donors start saying they heard from an assistant, that is a more convincing pair of signals than either alone.

How should you report it internally?

Keep it modest. Say how many visits and gifts you could attribute, say that this is a minimum, and give the date range. Do not set a target based on a number you cannot fully measure. See AI visibility metrics that actually mean something and what an AI citation is worth.

What will Citewake add?

Citewake measures what assistants recommend. What happens on a charity's website is outside its scope. Its charity programme starts in February 2027, and its method is on the methodology page. It will not tell you what your own analytics show.