What Are AI Hallucinations?

Domos Team - 4 min read - Updated June 2026

An AI hallucination is when an AI model produces an answer that sounds confident and fluent but is actually false, fabricated, or unsupported. The model is not lying - it is filling a gap with a plausible-sounding guess, because that is what language models do when they lack a grounded answer.

In a property management context, that could look like an agent confidently quoting a resident the wrong lot rent, inventing a pet policy, or citing a lease clause that does not exist.

AI hallucinations in one line

It is when AI states something untrue with total confidence - the failure mode that matters most when answers carry legal or financial weight.

Why hallucinations happen

Hallucinations are a known property of how large language models work. They predict likely text from patterns rather than retrieving verified facts, and standard training tends to reward a confident guess over admitting uncertainty. As Google Cloud explains, a major cause is a lack of grounding - when a model has no connection to real, authoritative information, it can generate output that is plausible but factually wrong, and can even fabricate details that were never there.

That distinction is the whole game. A model left to free-generate answers about your community is guessing. A model that can only answer from your actual data is reporting.

Why this matters for operators

In most consumer chatbots, a hallucination is an annoyance. In property management, it is a liability. A wrong answer about money owed, a notice deadline, or a lease term is not just a bad experience - in a fair-housing-sensitive, heavily regulated industry, an inconsistent or invented answer can create real compliance exposure.

This is where a lot of generic AI tools quietly fall short. A bolt-on chatbot trained on the open internet will happily answer a question about your community it has no real basis to answer. It sounds great in a demo and invents a fee in production.

How a well-built system avoids it

The fix is not a better-sounding bot - it is architecture that constrains what the AI can say:

• Grounded in your data. Answers come from your PMS records, your policies, and the hundreds of parameters configured for your operation - not from the model's general training. It reports your lot rent, not a guess at it.

• Bounded scope. The system is configured to handle the topics it should and to not free-form answers outside them.

• Escalation on uncertainty. When the AI is not confident, it hands off to a human instead of inventing an answer - roughly 20% of interactions at Domos go to a person for exactly this reason.

That combination is why "where do the answers come from?" is the single best question to ask any AI vendor. Grounded-and-escalating beats fluent-and-guessing every time.

Bottom line: grounded, not guessing

Hallucination is a risk you manage by design, not a feature you switch off. The right standard is not an AI that always answers - it is an AI that answers from your data and knows when to get a human.

Frequently asked questions

Can AI hallucinations be eliminated completely? Not entirely - it is inherent to how language models work. But they can be sharply reduced by grounding the AI in your real data, constraining what it is allowed to answer, and escalating to a human when it is unsure.

Why are hallucinations risky in property management? A confident wrong answer about rent owed, a lease clause, or a fee can create a fair-housing or compliance problem, not just a bad experience. Accuracy is not optional in regulated communications.

How do I handle an AI tool hallucinates? Ask vendors how they manage hallucinations. If they claim they don't have any, they might be hiding something. A good vendor has a system in place to catch and handle hallucinations.

Want an AI that answers from your data, not a guess?

Want to learn more? Check our glossary