AssociationAI / AI Literacy
AssociationAI team Published · Updated

Article

Beginner

What Generative AI Is (and Isn't) for Association Staff

Generative AI drafts and summarizes; it isn't a person, isn't always right, and shouldn't decide for members. What association staff need this week.

Generative AI Association staff Basics Governance Privacy

Hero illustration for: What Generative AI Is (and Isn't) for Association Staff

Your executive director said “start using AI.” A member asked whether the newsletter was written by a bot. Somebody on the team pasted a renewal list into a free chatbot last week and nobody wants to bring it up.

That is the real starting point for most association staff, and it is a better one than a lab demo. You need a shared definition of what these tools do, a few jobs they can help with right away, and the habits that keep members trusting you. This is that starter kit.

What it is

Generative AI produces new text, images, audio, or code by learning patterns from enormous amounts of existing material. The output resembles what the system trained on. It does not know facts the way a colleague knows them.

The chat tools you open at work run on large language models, which predict the next likely piece of language given what you typed and what they have already written. It helps to think in layers: AI is the field, models sit inside it, and the products you actually click on (chat apps, copilots inside email and documents) sit on top.

It is good for drafts, summaries, and thinking out loud, but it is not reliable on its own, and its range is narrow compared with human judgment.

What it is not

It is not a person. It finds patterns. It does not have your membership’s history in its head, and it does not care whether the renewal goes out.

It is not general intelligence. Today’s tools are narrow systems that handle a few kinds of task, not artificial general intelligence that could do the range of work a person does, and nobody has built the second kind. Treating a chatbot as a board substitute is a category error.

It is not always right. Models produce confident, fluent, wrong answers. Most people call them hallucinations. NIST’s Generative AI Profile calls the same thing confabulation. Either way, the model is predicting words, not looking anything up. Check anything that matters.

It is not the decision-maker. ASAE’s organizational AI policy is explicit that humans review before publishing, sending, or acting, and humans stay accountable. That includes hiring, member standing, and anything the association says publicly.

It is not a privacy shortcut. Member PII, donor notes, certification records, and confidential board material do not go into unapproved public tools. Independent Sector frames data protection as central to nonprofit trust, and they are right.

It is not exempt from ordinary truth rules. The FTC’s Operation AI Comply exists because companies kept overselling what their AI could do. There is no AI exemption from deception law. That applies to your member marketing and to any vendor pitch you repeat.

Five things to try this week

The small examples inside each item are made-up teaching cases, not reports of what any association did.

Membership. Ask an approved tool for plain-language answers to common dues and benefits questions, using only your public website copy as the source. Keep the AMS export on your own screen. For example, generate three FAQ answers about renewal timing from the public membership page, then verify every date, grace period, and fee yourself before anything goes into a template.

Events. Speaker bios and session abstracts eat hours. Paste speaker-approved text into an approved tool and ask for a shorter house-style version. Do not let it invent quotes or affiliations. One way to start is to request a 75-word cut of one approved bio, then confirm title, affiliation, and pronouns with the speaker before print.

Advocacy. Position outlines and bill summaries are tempting jobs for a model, and they are also where confabulation does the most damage. Use the model for research prep and outlines only, and have a subject-matter expert and legal reviewer sign off before anything becomes an official position or a member alert.

Communications. First drafts, alt text, and social calendars are natural fits. Prefer one good send over a week of machine noise. If your policy asks for disclosure when AI meaningfully shapes member-facing content, follow it. A simple test is to draft a webinar newsletter intro from confirmed event details only, delete anything invented, and check date, time, and registration link before the send.

Board. Brief directors once a year on what these tools can and cannot do. NIST’s AI Risk Management Framework gives you shared vocabulary (Govern, Map, Measure, Manage) so you do not have to invent jargon. There is a full tutorial on the one-page board brief elsewhere on this site.

A short checklist

Confirm which tools your association has approved and which are off-limits for member data. Pick one low-risk, mostly internal workflow for a two-week pilot. Write the prompt from public or already-approved material only. Generate a draft and assume it can be wrong. Check every date, name, title, fee, and link against a primary source. Edit for your association’s voice. Put a named person’s name on the send, not the model’s.

Two warnings

Fluent text feels trustworthy. It is not evidence. NIST specifically calls out over-reliance and the tendency to treat the system like a person. Stay skeptical when output touches members, money, or public positions.

Skip invented hour-savings and adoption percentages in the board deck. Vendors will offer numbers like that freely; a director who asks where the number came from deserves a real answer. If it cannot be traced to a source someone checked, leave it out.

Sources (6)