AI training for a real estate team — what to cover in the first 90 days

The first 90 days of AI training for a real estate team should do one thing: get agents from "I tried ChatGPT once" to running at least one complete AI-assisted workflow without help. That's it. Not automation of the whole business. One reliable workflow, built on actual prompt fluency, so the team has something real to compound from.

Why most team AI rollouts stall in week two

The pattern I see across real estate teams is almost always the same. A team leader attends a conference, gets excited, buys three tool subscriptions, runs a two-hour offsite demo, and then... nothing changes. Agents open the tool, write a vague prompt, get mediocre output, and go back to doing things the way they did before.

The failure isn't motivation. It's sequencing.

Teams try to automate before they can prompt. They try to build a system before any agent has a repeatable workflow. They introduce GoHighLevel and Zapier before anyone can reliably get a useful email draft out of Claude. This is like teaching someone to drive stick in rush-hour traffic before they've done a parking lot.

The 90-day framework below fixes the sequencing. Each phase unlocks the next. Skipping phases doesn't speed things up — it usually sets you back four weeks.

Week 1: pick the stack, not all the stacks

The first week has one job: decide what the team will use, and commit to that decision for 90 days.

The working stack I'd start any team on is short on purpose.

• One writing and thinking tool: Claude or ChatGPT. Either works; pick one and standardize. I lean toward Claude for drafting client communication because the default voice is cleaner, but ChatGPT's browsing and integration ecosystem is real if your team needs live market data. See the Anthropic pricing page and OpenAI's pricing page and get everyone on the same paid tier.
• One transcription tool: Otter.ai is the easiest on-ramp. Agents record a showing debrief, get a transcript, use that transcript as AI input. That's the seed of several workflows.
• The CRM they already use. Not a new one. The existing one.

That's the stack. No image generators, no automation platforms, no extra subscriptions until week seven at the earliest. Tool overload is the number one adoption killer.

Weeks 2 and 3: voice first, prompts second

This is the phase most team leaders want to skip because it feels slow. It isn't.

Before agents can write a useful prompt, they need to know what "useful" output looks like for their specific team. That means building a team voice document: a one-to-two page reference that captures how the team communicates with clients, what words it avoids, what tone it holds, and what a good email or listing description actually sounds like for this brokerage and this market.

The fastest way to build this: have each agent bring three client emails they're proud of. Pull the patterns. Turn them into a reference doc. That doc becomes the first line of every prompt the team writes.

From there, the prompting curriculum for weeks two and three is simple.

• What a weak prompt looks like versus a strong one. ("Write a follow-up email" versus "Write a follow-up email for a buyer couple who saw a two-bedroom condo last Saturday. They have two kids and mentioned needing more storage. They haven't responded yet. Use our voice doc. Three sentences max.")
• Giving context, persona, and constraints in every prompt.
• Iterating on output rather than starting over. Most agents throw away a draft and start again when one edit would fix it.

Two to three hours of live practice in weeks two and three, with actual agent emails and listing descriptions as the raw material, is worth more than a ten-module video course.

Weeks 4 to 6: one workflow, dialed all the way through

By week four, agents should be able to get decent output on the first or second prompt attempt. Now the work is building one complete workflow that runs start to finish without friction.

The workflow doesn't need to be ambitious. It needs to be complete. A simple one that works well for most teams: showing debrief to follow-up email.

The steps look roughly like this. Agent records a two-minute voice debrief after a showing (the property address, buyer reactions, objections raised, next steps). Otter transcribes it. Agent pastes the transcript into Claude with a prompt built from the team voice doc. Claude drafts a follow-up email. Agent reviews, edits if needed, sends from their actual email client or CRM.

That workflow saves real time. It also teaches agents to trust the output, because they're working with their own words, not hallucinated content.

Pick one workflow for the whole team. Run it for three weeks. Fix what breaks. Don't introduce a second workflow until the first one runs without anyone needing help.

Weeks 7 to 12: compound workflows and light automation

This is where the work from the first six weeks pays off. Agents who can prompt reliably and run one workflow end-to-end can start connecting workflows.

A compound workflow is just two or more steps where the output of one feeds the next. Examples for a real estate team:

• The showing debrief workflow above, extended so the Claude summary also populates a CRM note and generates a task reminder.
• A listing launch workflow: agent inputs property details and three unique selling points, Claude drafts the listing description, the social caption, and the email to the buyer list, all from one prompt session.
• A market update workflow: agent pastes a data pull from the local board, Claude synthesizes the key points, outputs a client-ready paragraph for the newsletter.

None of these require Zapier yet. Most of them can run manually, with an agent copying output from Claude into the next tool. That's fine. Manual compound workflows build the muscle. Automation comes after the manual version is proven.

If the team has a technically capable person and a genuine volume problem (the manual steps are happening dozens of times per week), then Zapier or GoHighLevel automation is worth exploring in weeks ten through twelve. Not before. Automating a broken workflow makes a bigger mess faster.

What I'd do if I were running this rollout

I'd resist the pressure to show progress in week one. The questions from agents in week one are almost always "when do we get to the cool stuff." The answer is: this is the cool stuff. A team that can prompt fluently and run one reliable workflow in 90 days is ahead of most brokerages in this market.

I'd also keep the accountability structure simple. A fifteen-minute team check-in at the end of each week, where one agent shares what they built or fixed, is enough. Public practice accelerates adoption faster than a course with completion certificates.

The teams that stall tend to have one of two problems: the team leader is excited but the agents aren't required to participate, or the training is theoretical and agents never touch a real prompt on a real workflow. Fix either of those and the 90-day framework holds.

FAQ

What should a real estate team cover in AI training? The first 90 days should move in phases: picking one shared tool stack (weeks 1–2), building prompt fluency and a shared team voice (weeks 2–4), dialing one complete workflow end-to-end (weeks 4–6), then compounding by connecting workflows into systems (weeks 7–12). Most teams fail because they skip straight to the system before any agent can reliably run a prompt.

Which AI tools should a real estate team start with? Start with one writing tool (Claude or ChatGPT), one meeting or transcription tool (Otter.ai or similar), and the CRM your team already uses. Avoid adding GoHighLevel, Zapier automations, or image-generation tools in the first month. Tool overload kills adoption faster than anything else.

How long does it take for real estate agents to get fluent with AI prompting? In my experience, most agents hit basic prompt fluency — meaning they get useful output on the first or second try — within three to four weeks of regular practice. The gap is usually not intelligence; it's lack of a structured prompting framework to start from.

What is a compound workflow in real estate AI training? A compound workflow is two or more individual automations connected so the output of one feeds the next. An example: a showing debrief recorded in Otter feeds a Claude summary, which populates a CRM follow-up task, which triggers a nurture email draft. Each step is simple; the compound effect is meaningful time savings.

Should every agent on a team use the same AI tool? Not necessarily the same tool forever, but yes for the first 90 days. Shared tooling means shared prompts, shared wins, and shared troubleshooting. Once the team has a baseline, individual agents can branch into specialized tools for their own workflow needs.

What is the biggest mistake real estate teams make with AI training? Trying to automate everything before anyone can do anything reliably. The pattern I see is a team leader demos five tools at an offsite, agents try each one once, nothing sticks. Narrow first. One tool, one workflow, one win — then build.

---

Emma Pace — strategic marketing consultant, AI coach for realtors, keynote speaker. Realtor at Monstera Real Estate. Builds AI-operated marketing systems at emmapace.ca.