Skip to content
← Back to Blog

If Your Chatbot Says Dues Are Deductible, You Said It.

A tribunal asked why customers should have to double-check information found in one part of a website against another part of the same website. For an association with a dues page and a bot, that is the whole risk.

The pitch is good, and the pitch is not wrong about the problem. Your staff answers the same forty questions all year. Dues amounts, renewal dates, who qualifies for which category, whether the early rate still applies, what the member discount is on the certification. A bot could answer those and your team could do something else.

The question worth asking before you buy one is what happens the first time it answers wrong, and whether anyone at your organization would find out.

Where this sits. This is the third of five posts on AI and association websites. The legal material below is from the decisions themselves and the measurement is from named academic and public-service research. We have flagged where the evidence is thinner than the headlines suggest.

The Case Everyone Cites, and What It Actually Held

Moffatt v. Air Canada was decided by the British Columbia Civil Resolution Tribunal on February 14, 2024. The airline's chatbot told a passenger he could apply for a bereavement rate retroactively within ninety days of issue. The airline's own policy page said the opposite. The tribunal awarded 650.88 dollars in damages plus interest and fees, a total of 812.02 Canadian dollars.

The amount is not the point. The money is trivial. Two passages are not.

First. On the argument that the company was not responsible for its own bot, the tribunal wrote that Air Canada suggests the chatbot is a separate legal entity responsible for its own actions, and called that a remarkable submission. That disposed of the idea that a bot is a third party you host.

Second, and this is the one for associations. On the argument that the correct information was elsewhere on the site, the tribunal asked why the page titled Bereavement travel was inherently more trustworthy than the chatbot, and why customers should have to double-check information found in one part of a website against another part of the same website. For an association with a dues page, a benefits page, a bylaws PDF and a bot, that sentence is the whole risk in one line.

Being Straight About the Legal Picture

Moffatt gets cited constantly as though it started a wave. As of this writing it has not, at least not here.

No US decision has followed it. We looked for a United States court decision since Moffatt holding an organization bound by what its own site chatbot told someone, and did not find one. Everything circulating is commentary recycling the Canadian case. Anyone telling you there is American case law on this is overstating it.

The cases that moved are European. The substantive successors are German. On May 12, 2026 the Oberlandesgericht Hamm held that the responses of a chatbot deployed by a company constitute commercial acts of that company, rejecting the argument that the operator escapes responsibility because the system acts on its own, and rejecting a disclaimer defense on the reasoning that a large share of consumers place particular trust in the correctness of a computer-generated answer. The case was brought by a consumer protection association rather than an individual, and appeal to the Federal Court of Justice was permitted.

What that adds up to. Two courts on two continents were asked whether a disclaimer and a correct page elsewhere get you out of what your bot said. Both said no. That is not American precedent, and it is a clear enough direction that we would not build a plan around the hope that the next court disagrees.

How Often Is It Actually Wrong

Vendors answer this with the word grounded, meaning the bot answers from your content rather than from a model's memory. That helps and it does not solve it, and there are real numbers.

  • Retrieval reduces errors rather than removing them. Stanford researchers evaluated purpose-built legal research tools from LexisNexis and Thomson Reuters, systems engineered by large vendors over clean proprietary corpora, and found they hallucinate between 17 and 33 percent of the time. Your website content is messier than Westlaw.
  • And wrong source content wins. A NeurIPS study found models adopt incorrect retrieved content over their own correct knowledge more than 60 percent of the time, and that the effect is strongest when the wrong content is only slightly wrong. An obviously absurd page gets rejected. A dues page that is one year out of date gets believed.
  • On sourcing and citation specifically. A European Broadcasting Union and BBC study of 2,709 responses across four assistants found 45 percent had at least one significant issue and 20 percent had significant accuracy problems. A Columbia Tow Center study of 1,600 queries found incorrect answers ranging from 37 percent for one tool to 94 percent for another, and noted the tools presented wrong answers with confidence, rarely hedging.

The honest caveat. These measure different systems than an association support bot, and we would not claim otherwise. The transferable finding is the second one, because it describes exactly the condition every association website is in: pages that are mostly right, partly stale, and occasionally contradict each other.

Why Associations Are a Harder Case Than Retail

A shop's bot getting a return policy wrong creates an unhappy customer. Several association answers sit on top of actual obligations.

Dues deductibility. Contributions to a 501(c)(6) are not deductible as charitable contributions, though they may be deductible as an ordinary and necessary business expense. Separately, organizations with lobbying expenditures must notify members of the nondeductible portion of dues at the time dues are assessed or paid, and failure to give that notice means a proxy tax at the highest corporate rate. A bot that cheerfully answers yes when a member asks whether dues are deductible is contradicting a disclosure the statute requires you to make correctly, at the exact moment the statute cares about.

Auto-renewing dues. State automatic renewal laws set out how terms must be presented and how cancellation must be offered. A bot that misstates how to cancel an auto-renewing membership is misstating a legally specified term. Note for accuracy that the federal click-to-cancel rule was vacated in full by the Eighth Circuit in July 2025, so this is state law, not federal.

And disclosure is starting to be legislated. Utah requires a prominent, unprompted disclosure when a generative AI system interacts with someone in connection with regulated services, including financial, legal, medical and mental health advice, and a disclosure on request in ordinary consumer transactions. Certifying bodies and professional societies whose bots answer credential questions should have a lawyer read that one.

What Has to Be True Before You Turn One On

We are not against these. We are against turning one on without the five things underneath it:

  • A content corpus somebody has reconciled. If the bot answers from your pages and your pages disagree with each other, the bot will confidently pick one. Reconciling dues, eligibility and deadlines across the site is the actual project, and it improves the site whether or not you ever ship a bot.
  • A written list of what it must refuse. Dues deductibility, eligibility determinations, bylaws interpretation, anything touching a legal or financial obligation. Refusing and handing off is a feature, not a failure.
  • Disclosure that a person actually sees. Not a paragraph in the terms. Visible at the start of the interaction, in plain words, because both courts that looked at this were unimpressed by disclaimers placed where nobody reads them.
  • A named person reviewing transcripts. Somebody reads a sample of real conversations on a schedule and is accountable for what they find. A bot nobody audits is a publication nobody edits.
  • A clear answer to who is responsible when it is wrong. If the answer is nobody, you have bought a liability with a monthly fee.

Send Us the Forty Questions

Before any of this becomes a purchase, write down the forty questions your staff actually answers and mark each one: safe for a bot, needs a person, or touches an obligation. Most associations find the third column is smaller than they feared and more serious than they expected, and that exercise alone tells you whether you need a bot or a better dues page.

Send us that list and the pages the answers currently live on, and we will come back with which questions your existing content can already answer correctly, which ones your own pages contradict each other on, and which ones we would not let a bot near at all. If the contradictions turn out to be the real problem, we will tell you that and it will be cheaper than what you were about to buy.

83 Creative

We're a web development studio that works exclusively with trade associations, professional societies, and membership organizations.

← Previous Article The European Accessibility Act Already Applies to US Companies That Sell Online to Europe