The Broke Agent had a LeBron James trend jack ready to go, timed to the news about LeBron’s free agency. He used AI for the images and the first draft of the script.
He told Dave Hutch how it went on a recent podcast.
“The post is completely flopping. I used all of my own inputs, but it’s completely flopping.”
The idea was good and the timing was right. The problem was viewers could hear AI in the cadence of the script, and they scrolled past.
Agents use AI every day for listing copy, video scripts, follow-up and research. Some are saving hours a week with it. Others are posting content their audience tunes out.
Which is pretty much the opposite of what you’re aiming for.
BAM contributors and podcast guests have spent the past several months sharing how they use AI in their businesses, learning as they go through trial and error.
Here are the 10 mistakes that came up again and again, along with what working agents do about each one.
Context Mistakes That Leave AI Guessing
AI can only work with what you give it. The first mistake happens before an agent opens a chat, and the second happens in the prompt itself.
1. Re-Explaining Yourself Every Time You Open a Chat
Every new chat starts from zero unless you give AI a way to remember who you are.
The Broke Agent fixed this with a voice brain dump. He talks for 10 to 15 minutes using voice mode or a recording app and covers:
- His professional background
- His current role
- His team and tools
- Projects he’s working on
- His career stage
- His target clients
- The markets he works in
- His business goals
He pastes the transcript into Claude and asks it to add the key details to its memory. He refreshes it once a month.
He described the difference it made.
“Now when I open Claude, it already has the context for everything I’m working on. No re-explaining. No starting from scratch. It just knows.”
For tasks you repeat, Haley Ingram of Coffee & Contracts saves her instructions as a Skill, which is a saved set of instructions Claude follows each time you use it. She builds one by going back and forth with Claude on a task until the output is right, then telling Claude to save it.
Ingram explained how a Skill works for a weekly newsletter on The Broke Agent’s podcast.
“A Skill is essentially like a standard operating procedure or like guidelines that you are giving Claude that anytime that you mention this specific task, it’ll run that Skill automatically.
“So, for example, if you have an email newsletter that you send out every week, instead of every time you go into Claude saying, ‘Hey, I want to write my email newsletter, here’s the details, here’s what we’re doing.’ You just say, ‘I need to write the email newsletter.’”
She recommends deleting or archiving Skills you stop using, so old instructions don’t clutter your setup.
Robert Cedeno runs a 25-agent team in Southern California and gave Meta’s AI assistant, Muse, standing instructions for his home tour voiceovers. He walked The Broke Agent through how he set it up.
“Now that Meta came out with Muse… all I do is I set up the prompt, like, ‘Hey, this is how I want the voiceovers. Anytime I give you an address, I need you to create three different voiceovers. One, I need to start with something like, this is what 850,000 gets you in this area.’ So it’s giving me three different versions on Muse. Which is, if you’re not using it, get on it now. All I have to do now is give it the address. It gets all the information from Zillow and creates the three voiceovers for each property.”
2. Giving AI Too Little to Work With
Krys Benyamein uses ChatGPT to write his listing descriptions. He opened his post on the process with a warning for other agents.
“One of the biggest mistakes I see agents make with AI is assuming the prompt is the most important part of the process. It isn’t.”
He builds a full property file before he asks AI to write anything. It pulls from seven sources:
- A recorded walkthrough with the seller
- His own walkthrough after the seller leaves
- Tax assessor and public records
- Previous listing data
- Property photos
- Screenshots from the 360 tour
- Neighborhood and lifestyle details
Benyamein summed up what changed his results.
“AI didn’t make my listing descriptions better. Feeding AI richer, more detailed information made my listing descriptions better.”
A vague ask gets a vague result, and a narrow ask gets you something you can use. Dave Hutch showed how narrow he goes when he set up a daily MLS search with Claude. He shared the exact instruction he gave it.
“I had to find condos under $500,000 in my farm area that have had a price drop in the last three days.”
Review Mistakes That Let Bad Output Slip Through
AI makes mistakes, and with the agent’s name on the post, their credibility is on the line. These three mistakes come down to skipping the check before setting it live.
3. Publishing Data Without Checking It
AI tools can pull numbers that are out of date or details that are wrong. A bad stat in a market update or a wrong square footage in a listing description is on you.
Katie Day built her Houston Unlocked Instagram account to 46,500 followers by covering local news and new developments. She has Claude scan Houston city council websites and send her summaries of what’s been filed.
Day explained why she reads the original agendas herself.
“I have it searching the Houston city council websites for planning agendas and plat filings and different stuff like that and giving me kind of a summary of the things that are happening because there’s so many different things that are being filed and talked about and most of them aren’t newsworthy, but I want it to kind of comb through them and also remind me when the meetings are so that I can go look at the physical agendas themselves just cuz, again, AI might miss something that’s actually a big deal to the people of Houston.”
If an AI tool pulls property details from a third-party site, check them against the MLS or public records before they go into a description or a voiceover.
4. Treating the First Output as Final
Dave Hutch built a Claude Skill that tests his content ideas before he films them. It runs each idea past a board of five AI advisors:
- Two marketer seats modeled on Alex Hormozi and Sabri Suby
- A Champion who pushes the idea toward its best version
- A Skeptic who tries to kill it
- A Chair who makes the final call
Hutch explained how the board reaches a decision.
“The fifth seat is just the Chair and they decide the outcome. So this prompt and what this does is every single time I bring an idea to this board of advisors, they argue back and forth, they tell me whether the idea is keep, reshape or kill.”
On the podcast, he ran an idea called “best English food in Toronto” through it. The Chair called for a reshape. The new version featured three of the city’s most English neighborhoods, each paired with a pub and a home price, tied to the World Cup.
Ingram described how she thinks about Claude in her own business.
“The best way to think about it is more like a partner and someone that you can use to go find the information, analyze the data, make shortcuts for you to make things easier, create dashboards for you that create views better, help you think through things and think through it through different angles.
“So really looking at it as like it gets you most of the way there and then you finish it with your own touch.”
5. Trusting Automations Without Testing Them
A 60-second demo reel can make automation look effortless. Real setups, though, can break or miss crucial details. Content that shares an inaccurate or outdated stat makes you look either careless or uninformed. Either way, your audience learns (quickly) to tune you out.
BAM’s breakdown of Hutch’s MLS setup noted that Claude’s Chrome extension doesn’t always run as smoothly as a quick demo suggests, and results with CRMs and other tools can be mixed.
Plus, some MLSs don’t allow the use of AI tools to scrape for listing information, so that’s something you’ll want to check before setting up an automation involving the MLS.
The steps BAM laid out for setups like Day’s include a test run. Before you trust a scheduled task, run it for a couple of cycles and compare what it sends you to the real source.
Once a task runs on its own, keep checking what comes back.
Voice Mistakes That Make Agents Sound Like Everyone Else
Clients hire people. These three mistakes strip the person out of your content and your conversations.
6. Posting Social Content With Nothing of YOU in It
In May, The Broke Agent wrote a warning for agents who type in a prompt and post whatever comes out. He uses AI all the time himself.
He described what fully AI-written content looks like to people scrolling past it.
“You can REALLY tell in video scripts because they’re completely lifeless. It’s turning people into content robots who take a breaking hyperlocal news story, throw it into ChatGPT, and spit it back out with zero opinion or personality of their own.”
Several agents in BAM’s coverage draw a line on what they hand to AI:
- Hutch writes his own video scripts and uses AI for hook ideas
- Day picks her own stories and edits AI drafts into her own voice
- Benyamein builds his videos from his own experiences, like being a dad and lessons he’s learned in sales
Real estate marketing coach Will Draper recommends what he calls the boring content strategy. These are plain talking head videos that answer questions clients ask all the time, posted on a consistent schedule.
The Broke Agent closed his warning with a test to run before you post.
“Moral of the story: think before you post. Is this actually adding anything? Is this something only YOU would say? If not, that’s where the danger is.”
7. Using Robotic Scripts in Client Conversations
AI can write a call script in seconds. Generic AI scripts tend to run long and explain more than they need to, and clients can hear it.
Former FBI negotiator Chris Voss keeps his scripts short. On the Playmakers Podcast, RealScout CEO Andrew Flachner threw common objections at him in a live role play, and Voss answered in real time.
When a seller said another agent would charge less, Voss’s answer was one sentence:
“Sounds like you’re reluctant about paying for results.”
For a prospect who’s gone quiet, he suggests this text:
“Have you given up on working with me?”
It gets a clear answer about where the prospect stands.
Voss explained the idea behind his approach:
“The commodity in all negotiations is information.”
BAMx’s AI Script Advisor is built to avoid generic output. It’s trained every week on the frameworks and scripts the BAMx coaches’ teams use today.
Scripts get sharper when you say them out loud to real people. Byron Lazine, Lisa Chinatti, and Tom Toole host the live BAMx Role Play Mastermind, where members bring real scenarios from their businesses to the hot seat. The September scripts started as suggestions during the live calls, and the team developed them further with the BAMx Script Advisor.
In a post on scripts from a recent Mastermind, Byron gave a delivery tip for a neighborhood pitch.
“Watch your delivery. Keep your pitch low and steady, because a high-pitched voice tells people you’re nervous.”
8. Letting AI Reply to Clients With No Review
AI assistants can answer DMs and draft emails for you. The agents getting results with them keep a person in the loop.
Cedeno uses Muse for a lot of what a personal assistant would handle. He explained where he draws the line.
“Muse has been scheduling all my calls. It’s been going and responding to DMs. I still have to approve them when they respond, but it’s basically like my personal assistant.”
One of the tasks Ingram listed for Claude’s Cowork feature is drafting emails into your inbox. The drafts wait there for you to review and send.
Chinatti makes the case for a personal touch with open house leads. She suggests a handwritten note on heavy paper, which puts you ahead of agents who send a few texts and drop leads into automated email campaigns.
Time Mistakes That Keep Agents Stuck on Busywork
Some agents use AI for writing and nothing else. This mistake costs them hours every week.
9. Not Using AI for Repetitive Tasks
Cedeno used to spend more than an hour writing 30 home tour voiceovers. With his AI setup, he now gets them done in 6 to 7 minutes.
Agents in BAM’s coverage are handing off a range of repeat work:
- Hutch runs scheduled MLS searches for price drops, new listings and CRM leads nobody has touched in more than 48 hours
- Ingram has Claude build a morning dashboard, time-block her calendar and draft emails
- Phil Stringer uses a team of AI agents that turned one prompt into the full follow-up plan for a live training
- Day gets a scheduled summary of Houston city council filings
Cedeno’s advice is to start with one task you repeat every week and give AI clear instructions. Then review what comes back.
AI can help you plan your week, too. Katie Lucie used it to check whether her business plan fit inside her calendar.
After logging her time for two weeks, she broke each of her lead pillars into the tasks it takes. She put the full task list and her existing time blocks into AI and asked it to find:
- Commitments she hadn’t set aside time for
- Recurring tasks she’d left out
- Places where she’d planned more output than her calendar could support
Lucie described what the exercise did for her.
“That exercise was incredibly clarifying. It’s very easy to write ‘social media’ on a business plan. It’s much harder, and much more useful, to prove your calendar can produce the six pieces of content you just promised yourself you were going to create every week.”
Her post lays out the full two-week time audit.
Visibility Mistakes That Keep AI From Recommending You
Buyers and sellers in your market are asking AI who to call for help with buying or selling a home. This last mistake all but guarantees your name won’t show up in the answer.
10. Ignoring How AI Recommends Agents
Ask ChatGPT, Claude, Perplexity or Gemini for the best agent in town, and you’ll get names.
Draper broke down how those tools decide which agents to recommend with a five-part framework:
- Foundation
- Authority
- Relevance
- Proof
- Media
Draper said none of it depends on gaming the system.
“It’s not about tricks and hacks. It’s about reputation and clarity.”
Foundation comes first. It starts with the same name and brokerage on every profile online, plus one positioning statement that names who you help. Draper recommends writing it in three lengths:
- One sentence for your elevator pitch
- 750 characters to match the Google Business Profile limit, reused across directories
- 1,500 characters for the about page on your website
Phil Stringer, a former team COO turned AI strategist, adds website content built around the questions people ask AI. His framework includes:
- “Best choice” articles titled with questions like “Who is the best Realtor for first-time homebuyers in Phoenix?”
- “Reasons to choose” articles for each of your core services
Stringer was clear that this raises your odds of a recommendation without guaranteeing one. He suggests revisiting the content every six to 12 months.
To see where you stand, ask ChatGPT whether someone should hire you to sell their home, using your own name.
Then ask it who the best real estate agent is in your neighborhood. And why.
To get real clarity on whether AI would recommend you to a buyer or homeowners in your market, take a page out of Draper’s book and change the question you’re asking:
“I think people should stop asking ‘How [do] I rank in ChatGPT?’… You want to ask what would ChatGPT need to know about me to actually recommend me?”
Pick One Mistake to Fix This Week
You don’t need to fix all 10 at once. Start with the one closest to your daily work, whether that’s saving one set of instructions or running the AI check on your own name.
AI can handle the first draft. It’s great for that. But the voice and judgment that get clients to pick up the phone come from you. The real person behind the content. The person AI cannot replace.
