Agents remain responsible for AI-generated work

Artificial intelligence helps Realtors® work more efficiently, from summarizing reports to preparing marketing material. But one thing it can’t do is assume responsibility for the results.

“AI does not have a real estate license,” said Kevin Hawkins, a partner with WAV Group. “It is your license.”

In an interview with Florida Realtors® Live host Tracy Velt during the 2026 Florida Realtors Convention & Trade Expo, Hawkins said agents should remain the “human in the middle” whenever they use AI. That means choosing an appropriate tool, protecting customer information, checking for mistakes and never accepting the first answer without review.

WATCH 5 AI MISTAKES THAT COULD COST YOU A CLIENT

The first decision comes before an agent even types a prompt.

Hawkins said there are three levels of AI products: free tools, paid consumer products and business-grade versions. Each offers different levels of performance and protection, he said.

“If the product is free, you’re the product and also your data,” Hawkins said.

An agent who uploads a purchase and sale agreement to a free AI tool may unknowingly place personally identifiable information at risk, he said. The tool also may provide an inaccurate summary or add information that wasn’t in the original document.

Hawkins gave a similar warning about inspection reports. A free tool may pull information from other reports and include features that aren’t part of the property being evaluated, leaving the agent or customer to sort out the errors.

Business-grade products offer stronger protections and generally default to not using business information to train the AI, Hawkins said. However, he cautioned that any system can be hacked. Decisions about what to upload should reflect an agent’s business practices, comfort level and guidance from an attorney.

Selecting the right product is only part of the job. Hawkins said AI improves through iteration, correction and input from the person using it.

“Make sure that you don’t ever accept the first answer of AI,” he said. “This idea of cut, paste and pray when you’re using AI and get the first answer, that’s not where AI works.”

Agents also can train paid AI tools to better understand their voice and brand. Hawkins suggested providing examples of emails, newsletters, customer letters, market reports and website content. Agents can identify words they use, words they avoid and examples of work they like or dislike.

AI still can make unexpected mistakes, particularly when a tool is being updated, he said. Hawkins described finding dozens of errors in a report after months of accurate results. He ultimately completed the report manually.

He also uses multiple AI products to check one another, comparing their work with the original document. The same review is needed for images, which may alter faces or other important details.

Hawkins offered several tips for using AI more safely:

• Know which product you’re using. Determine whether the tool is free, consumer-grade or business-grade and learn how it handles information.
• Ask about its protections. Hawkins suggested asking the chatbot how its data protections compare with those of a business-grade product. Terms of service also can be reviewed for provisions that protect the company rather than the user.
• Be careful with sensitive information. Consider the privacy and security risks before uploading contracts, inspection reports or documents containing personally identifiable information.
• Don’t accept the first response. Check the answer, correct errors and provide more context until the result is accurate and appropriate.
• Train the tool. Give it examples of past work and explain which words, phrases and formatting reflect the agent’s voice.
• Cross-check important work. Compare the AI’s response with the original material and consider using a second tool to identify errors the first one missed.
• Watch for money and urgency. Stop and independently verify any unexpected message or call involving funds, changed instructions or a deadline.

That final warning has become more important as AI makes fraudulent communications increasingly convincing. Hawkins said a company presenting at a recent AI conference demonstrated technology that could clone a person’s voice using only five seconds of audio.

“If it involves money and urgency, stop,” Hawkins said. “Hang up. I’ll call you back on the line that I have for you.”

He advised calling the title company, escrow agent or other party using contact information already known to be accurate, not a phone number or link provided in the incoming message.

“You see money and you see urgency, you stop and you take a breath,” Hawkins said. “You don’t call anything on the information you’ve been given. You call the numbers that you know.”

© 2026 Florida Realtors®