Somebody reads about your work at nine on a Tuesday evening. They are moved. They open your website intending to give, and they have one question first: how much of this actually reaches the people you help?
If they can find the answer, a good number of them give. If they cannot, most of them close the tab, and the feeling that brought them there does not survive until morning.
Intent is perishable in the charity sector in a way it is not in most others, because nothing external forces the decision. Nobody needs to donate this week. There is no deadline, no broken boiler, no expiring MOT. The only thing carrying the action forward is a feeling, and feelings do not wait for office hours.
Three audiences, one front door
The awkward part of a charity website is that it serves three groups with almost nothing in common, all arriving through the same page.
Supporters want to know where money goes, how to give, how to change or cancel a regular gift, whether their employer will match it, and how to leave a legacy.
Volunteers want to know what is involved, how much time it takes, whether they need a DBS check, and whether their skills are useful.
People who need help want to know whether they qualify, what happens when they get in touch, and whether it is confidential.
A bot can route between those three within a couple of exchanges, which a static homepage cannot. That routing alone is worth the setup, because at the moment each group is wading through content written for the other two.
The question that decides a donation
"How much goes to the cause" is asked constantly and answered badly across the whole sector, usually with either silence or a defensive paragraph about how overheads are also the cause.
Publish the number. Publish what the rest funds and why. Let a bot deliver it plainly and without apology when someone asks. Donors are far more forgiving of honest overhead figures than of evasiveness, and evasiveness is what most of them currently encounter.
The same applies to the second question, which is what the money actually does. A specific answer beats an abstract one every time. Someone who learns what a given amount funds is making a concrete decision rather than an open-ended one.
Regular giving is where the value sits
One-off donations are useful. Regular gifts are what let a charity plan.
A bot should be able to explain how a regular gift works, how to change the amount, how to pause it, and how to cancel. That last one causes anxiety in fundraising teams, and the anxiety is misplaced. Making cancellation hard produces chargebacks, complaints, and a supporter who will never return. Making it easy produces some cancellations you would have had anyway and a lot of goodwill, some of which comes back later.
Where a donor asks to cancel, the bot's job is to explain the process honestly and route to a person, who can offer a reduction or a pause. That conversation retains more income than obstruction ever does.
Volunteers need expectations, not enthusiasm
Volunteer recruitment fails most often because of a mismatch discovered late. Somebody signs up imagining one thing, arrives to find another, and leaves after two sessions having consumed a lot of coordinator time.
The fix is boring and effective: answer the practical questions before anyone commits. How many hours. Which days. What the work actually involves on an average shift. Whether training is provided. Whether a background check is needed and how long it takes. What the minimum commitment is.
A bot that walks through those honestly filters out mismatches at no cost. The people who continue after reading the realistic version are the ones who stay.
The boundary that matters most
If your organisation supports people in difficulty, this section is more important than everything above it.
A chatbot must never attempt to support someone in crisis. Not triage, not assessment, not a sympathetic conversation while they wait. Any message that touches on self-harm, abuse, immediate danger, homelessness tonight, or a child at risk should produce one thing: the relevant emergency number and your trained human contact, immediately, with no intermediate questions.
Write that as an explicit instruction and test it deliberately before going live. Test it with the phrasings people actually use, which are rarely clinical and often oblique. A model that responds to "I don't know what to do, I've got nowhere to go tonight" with a cheerful FAQ answer is a serious failure.
The bot's legitimate role for this audience is narrower: explaining what services exist, who they are for, what happens when someone makes contact, and whether it is confidential. That is genuinely useful, because uncertainty about the process is a real barrier to asking for help. It stops there.
Small teams, disproportionate return
Charity operations tend to be thin, and the same two or three people answer everything. A bot that absorbs the routine layer of supporter and volunteer questions gives those people back hours that go directly into the work.
Point it at your about pages, impact reporting, giving pages, volunteer information, and service descriptions. Upload your annual report, which usually contains the specifics that answer the hardest supporter questions.
Then write the escalation rules first, before anything else, and test them properly.
Judging whether it helped
Look at donations completed after a chat interaction against your baseline completion rate, and at volunteer enquiries that survive to a first session.
But the number worth watching most closely is how quickly someone seeking help reaches a human being. If that time went down, the tool has done something more valuable than anything it did for fundraising.