AI Sales Role-Play for Solo Real Estate Agents: Free Tools Compared

A solo agent can practise objections and scripts for free with the voice modes in ChatGPT, Claude and Gemini by giving each a seller or buyer persona. Dedicated tools such as SalesGo (built for agents), Hyperbound, Exec and Yoodli add scoring, but their free tiers are small, from one role-play to a handful of sessions.

Free means different things here. The assistants meter by usage that resets, so they can supply the volume of practice, while the dedicated tools give a fixed number of sessions, so they work as benchmarks. Four 15-minute sessions a week for eight weeks is 480 minutes. Yoodli's free Starter plan gives five lifetime sessions, Exec's free plan five a year and SalesGo one first role-play: eleven sessions. At ten minutes each (my assumption, because none of those pages states a length) that is about 110 minutes, under a quarter of the target.

Follow me on Instagram@sagnikteaches

What each free tier gives a solo agent, checked 1 November 2026

Limits change often, so read this table as a snapshot and confirm on the vendor's page before you build a routine on it. Where a page did not say, the table says so.

Connect on LinkedInSagnik Bhattacharya

Two patterns show up. The general assistants have no fixed session count, only limits that reset, which suits daily practice. The dedicated tools bring scoring and ready-made scenarios, but their free allowances read as trials, and only SalesGo's scenarios are written for agents. Hyperbound's 45 prebuilt role-plays are the largest free allowance among the dedicated tools, and they are written for B2B sales teams, which matters further down.

Subscribe on YouTube@sagnikteaches

One persona prompt that works in every tool

Every tool below plays a prospect better when you do the work of describing one. Keep a single persona prompt in a note and paste it into whichever assistant you are using. The tool changes; the prompt does not.

Role-play a real estate prospect so I can practise out loud.
Stay in character until I say "stop".

SCENARIO: [expired listing call / owner selling privately / first buyer
consultation / seller who wants to wait until spring]
THE PERSON: [invented first name], [two lines: how long listed, asking
price, what they are annoyed about]
DIFFICULTY: [1 friendly / 2 sceptical / 3 hostile]

RULES
- Speak in short sentences, 25 words at most per turn.
- Raise one objection at a time. Do not accept my first answer; push back
  at least twice before softening.
- If I give a long speech, interrupt or say you are losing interest.
- Do not agree to a meeting unless I ask for it clearly and give a reason
  that matters to you.
- Do not invent market statistics. If you need a number, use one I gave you.
- When I say "stop", leave the role and give me: (1) what I did well,
  (2) the moment I lost control, (3) a better line for that moment,
  (4) a score out of 10 against the scorecard below.

SCORECARD: [paste your five criteria]

Three habits make the prompt earn its keep. Change the difficulty weekly, because a friendly prospect teaches you little after the first session. Keep the 25-word limit, since a persona that makes speeches lets you off the hook. And keep the line banning invented statistics: a role-played seller who announces that homes on the street sell 20% over asking is making it up, and you do not want that number in your head when a real seller asks.

ChatGPT voice: rehearse the expired-listing call

ChatGPT's free plan includes voice with limited access, and OpenAI's help page says limits may change. When you reach the limit a voice conversation can end, and you can carry on in text or start voice again. That makes it the best default for a first session and a poor choice for a long one, which is why the weekly routine later rotates tools.

Check what is kept. According to the same page, a voice conversation adds a transcript to your chat history, the audio clips are stored with the transcript and retained for 30 days, and clips are not used for training unless you choose to share them. Free users can opt in through the model-training switch under Data controls, so leave it off for practice and use invented names regardless.

One warning about where to build your persona. Skip role-play GPTs in OpenAI's store: OpenAI's retirement FAQ schedules custom GPTs to stop running on 11 December 2026, and a role-play GPT someone else built will not carry over to your account. A Project with your persona prompt as its instructions is the durable version.

Here is an illustrative exchange with a persona called Dana, whose listing expired last week after 94 days at $525,000 and who is annoyed that there was one open house. Difficulty 2.

Dana (AI): The listing expired last week. The last agent did nothing, so why would you be different?

You: That sounds frustrating. What did you expect them to do that they didn't?

Dana: One open house in three months. That was it.

You: One open house in three months is thin. I'd show you my marketing plan for the first 30 days, and I have a lot of experience with homes like yours.

Dana: Everyone has a lot of experience. Just send me an email.

What I would fix. The fourth turn pitched before asking whether the open house was the only gap, and "a lot of experience" is a claim with no number behind it. The email brush-off then went unchallenged, so the call ended with no next step. Better: "Was the open house the only gap, or were there others?", then "I'll send it, but a plan on its own won't help. Could we spend ten minutes tomorrow at 5:30 going through it against your house?" The assistant's debrief may catch the first problem and miss the second, so read the transcript for a missing next step yourself. My tutorial on expired listing scripts with AI has the call, text and letter wording to rehearse.

Claude: an objection ladder for the price-reduction call

Claude's voice mode is a beta feature on every plan, Free included. Voice conversations count towards your normal usage limits, and transcripts are saved in the chat history. On the free plan I would use Claude mostly for text, where Projects help: Free allows up to five, so make one called Practice and put in your fee page, your 30-day marketing plan and your three favourite scripts. Tell Claude to play prospects who have read them, and the persona argues with your actual promises instead of generic ones. If you are choosing between assistants for wider agent work, my comparison of Claude and ChatGPT for real estate agents covers it.

The format that suits text is a ladder: five objections that escalate, one at a time, with a verdict after each of your replies. Take a seller whose home has been listed for 30 days with 11 showings and no offers.

1. "It has been 30 days. What exactly are you doing?"
2. "The house down the road sold for more than your price."
3. "Can we just drop the price by $10,000 and see?"
4. "My partner thinks we should switch agents."
5. "If it doesn't sell this month we're pulling the listing."

Ask Claude to give one rung, wait for your reply, then answer with one word (held, dodged or conceded), one sentence of reasoning and the next rung. An illustrative verdict on rung 3: "Conceded. You offered a price drop before asking what the 11 showings told you. Try: 'Before we move the price, what has buyer feedback said?'" By rung 5 you have rehearsed the whole conversation, and the five verdicts show which rung you reliably lose.

Gemini Live: rehearse the open-house sign-in on a walk

Google's plans page lists Gemini Live on the free plan. Its general usage note says limits refresh every five hours until a weekly limit is reached; that is the app-wide footnote, not a figure for voice alone. Google's Live page also says Live cannot use Gems, so paste or speak the persona as your first message. For drafting scripts before you rehearse them, keep the persona as a skill, Google's replacement for Gems: its help page says Gems on personal accounts turn read-only on 17 November 2026, are converted to skills automatically and disappear on 17 December 2026. Live's page does not say whether it can call skills, so keep the spoken-persona habit either way.

Voice on a walk suits a short, repeatable skill, and the open-house sign-in is one. Set the persona as a visitor who says, "We're just looking, and please don't contact us." Your job is to ask for the sign-in without pressure by tying it to something the visitor wants. An illustrative good line: "No pressure at all. If you pop your details on the sheet, I'll text you the floor plan and the last three sale prices on this street while you walk round." The usual miss is pleading for the sign-in before offering anything, so ask the persona to say "why would I?" until you have named a reason. Only promise follow-up you will really do, and remember that consent rules for texts and calls apply; my tutorial on open house sign-in follow-up covers the next hour.

SalesGo: use the one free role-play as a benchmark

SalesGo is the one tool in this list built for agents. Its site describes role-play training with more than 25 real estate scenarios, including FSBO (for sale by owner), expired listings, buyer consultations and listing presentations, plus an AI coach you can reach by text or voice. The first role-play is free and needs no card, after which it costs $49 a month or $399 a year. The page states no trial length and no free credits beyond that first session, so treat it as a single test.

Spend that session on your hardest scenario, say the owner who wants to sell privately and save the fee. Write down your own score before you see theirs. Then run the same scenario in ChatGPT voice with the persona prompt. If the assistant plays the owner about as hard and the two scores agree within a point or two, you have no reason to pay. If SalesGo caught things your scorecard misses, copy those criteria into the scorecard and keep the free tools. Presenting to sellers is a separate skill, and my tutorial on the AI listing presentation deck covers the material you will be practising on.

Hyperbound and Exec: borrow the structure, not the scenarios

Hyperbound sells AI role-plays with customisable scorecards, and its pricing page confirms a free tier with 45 prebuilt role-plays and example scorecards. Paid plans are priced per user and quoted on request. The industries its pricing page names are B2B software, transport and logistics, sales-training agencies, and banking and financial advisers, so expect business-to-business call scenarios, not listing appointments.

Exec describes its role-plays as voice-based, with instant feedback on tone, technique and approach and a custom-scenarios feature. The free plan allows five sessions per user per year. Starter costs $20 per seat a month billed annually, with ten sessions per user a month. A solo agent can use the five free sessions as a yearly check-up: two before the busy season, two in the middle of it and one spare.

The value of both for an agent is translation. The structure of a good answer survives the change of industry even when the product does not.

Yoodli: five lifetime sessions, spent on baselines

Yoodli's Starter plan is the individual entry tier and includes five lifetime sessions. Pro is $8 a month billed annually with up to ten role-plays a week, and Advanced is $20 a month billed annually with unlimited use. Its example feedback is scored on criteria such as discovery questions, objection handling and clear next steps, which map neatly onto a listing appointment. The pricing page counts any practice session over 30 seconds as a role-play, so a false start or a dropped connection can cost one of the five. The site shows a video-call style screen but does not spell out whether you speak your answers, so check before you spend a session.

With five sessions in total, use them like tests, not practice. Run your listing-appointment opening as session one and note the scores. Make one change, say asking a discovery question before the pitch, and repeat the scenario in a month as session two. Three sessions then remain for the following quarters, and the difference between one and two shows whether your practice in the free assistants is working.

Where BrokerBot and MaverickRE fit

Both appear in search results for this topic, and neither is a free option for a solo agent. BrokerBot targets brokerages: its home page lists practising sales scripts and objection handlers with coaching guidance, and says the assistant is trained on what the brokerage knows. Its pricing page lists a Team plan at $499 a month for two to ten agents and a Brokerage plan at $999 a month, with a demo but no free plan or trial. That page never mentions role-play, so ask exactly what a plan includes. MaverickRE sells call grading and an AI sales coach to real estate teams. It lists the coach at $199 per user a month for teams under ten agents, with a free demo as the only no-cost offer; its site says agents role-play buyer and seller scenarios by phone and get a graded report, and its call-grading tier needs Follow Up Boss or BoldTrail as the CRM. The sums are blunt: $199 a month is $2,388 a year for one agent (199 × 12), and $499 a month is $5,988 a year (499 × 12) for BrokerBot's Team plan if role-play is part of it, against nothing for the rotation below. A paid coach has to change what you do on real calls to earn that. If you later join a team that already pays for one, use it; until then the free tools above do the same job with less setup.

Eight objections to drill, and what a good answer does

Follow Up Boss publishes a free written guide to five real estate role-play scenarios, including first-time buyers, open house sign-ins and the other-agent objection, which is a good place to find situations. This bank adds the persona line to give the AI and the shape of a good reply.

Three filled-in replies show the shape. They are illustrative, so replace the specifics with what you actually do.

Fee: "That's fair, and I'd want to know what I was paying for too. What did the other agent include besides the lower fee? Here is what mine covers: professional photographs, a buyer feedback report every Friday, and a price review at day 14. If it helps, I'll put my fee and exactly what it covers on one page so you can compare the two side by side."

Waiting: "Waiting makes sense if something changes in spring. What are you hoping will be different? If it's the garden, we can plan the photographs for the first good week and list the next day. If it's the price, I'll show you what has sold since January and what you'd net now compared with later. Can I send that by Thursday?"

Online estimate: "Those estimates are averages worked out by a formula, and nobody has walked through your house. I'd rather compare three homes that have actually sold near yours, adjusted for the kitchen and the extra bedroom. If the numbers support a higher price, I'll tell you so."

Scoring yourself without trusting the flattery

An AI that plays a prospect is also an eager marker, and eager markers hand out sevens. Put the scoring in a rubric you control, with five criteria worth two points each.

Then test the scorer before you trust it. Read it a deliberately weak answer: "Trust me, I'll definitely get you your price." A sound rubric gives that three out of ten or fewer and flags the promise. If it gives seven, tighten the rubric wording and try again. An illustrative result for the Dana call above: 5 out of 10, with two points lost for a missing next step, one for the vague proof point and two for naming no concern. Keep the rubric fixed for a month so that the scores form a trend instead of a mood.

Fair-housing traps are worth rehearsing

Role-play is a safe place to practise the questions that tempt agents into steering. Give the persona a buyer who asks, "Is this a good area for families like ours?" and, when you answer, presses with "But what kind of people live here?" The check is whether you redirect every time: ask what matters to them (commute, space, budget, outdoor room), offer objective facts such as distances and published sources they can verify, and decline to describe the people.

Ask the assistant to mark only that: did the agent describe residents, hint at demographics or answer a question the buyer was not entitled to have answered by an agent? Treat the verdict as practice, not compliance advice, because your broker's or association's fair-housing guidance is the rule. My tutorial on whether ChatGPT can write listings that pass fair housing covers the written side of the same risk.

A weekly rotation and a practice log you can check

Rotating across three free allowances means one limit never ends your week. Four sessions of 15 minutes give the 60 minutes a week used in the sum at the start.

After each session write one line in a log, and after each real call write one more: the objection you hesitated on. An illustrative week of entries:

Check that it is working at the four-week mark. Count how often real calls hit the same three objections and whether you hesitated less than before. If one objection keeps appearing in the log and in real calls, build a drill for it and nothing else for a week. Pay for a tool only after you have hit the same free limit twice and can say which paid feature would have prevented it.

Questions solo agents ask about free AI role-play

Is ChatGPT's free voice mode enough for role-play practice?

For a first session, yes. Free voice has limited access, and a conversation can end at the limit, after which you can continue in text or start voice again. Give it a persona prompt, keep sessions to ten or fifteen minutes, and rotate with Claude and Gemini so one limit never stops a week of practice.

Do I need a dedicated real estate role-play tool?

Not to begin. A dedicated tool adds ready-made scenarios and scoring, which saves set-up time, but the free allowances are small: one role-play at SalesGo, five lifetime sessions on Yoodli's Starter plan. Use a free session as a benchmark, then build the same scenarios in a general assistant and compare.

Can I rehearse fair-housing situations with AI?

Yes, and it is a good use. Give the AI a buyer who keeps asking what kind of people live in an area, and check that you redirect each time to objective facts and the buyer's own criteria. The rehearsal helps with wording, but your broker's or association's fair-housing guidance is the rule, not the assistant.

Is it safe to use real client details in a free AI role-play?

Avoid it. Use first names you invent and generic addresses. Consumer chat plans carry a switch that lets conversations improve the models, and ChatGPT keeps voice audio with the transcript for 30 days, so treat practice sessions as non-confidential and keep real client details out.

How do I stop the AI being too easy on me?

Tell it to push back at least twice before softening, to keep replies short and to score against a fixed rubric. Then test the scorer by reading it a deliberately weak answer. If that still earns seven out of ten, tighten the rubric before you trust any score it gives you.

Further reads

• How to Write Sales Call Scripts With AI That Don't Sound Scripted — Scripts that survive being said out loud.
• How to Turn Sales Call Recordings Into Coaching Notes With AI — Turn real calls into the next practice session.
• Real Estate Wire Fraud and Voice-Clone Calls: A Call-Back Script — A call-back script worth rehearsing before you need it.
• ChatGPT Training for Real Estate Agents: A One-Week Team Plan — A one-week plan if you start to hire.
• AI Lead Generation for Real Estate Agents: No Bought Lists Needed — Fill the pipeline you are practising for.
• AI Real Estate Cold Calling: Scripts, Lists and Do-Not-Call Rules — How a real estate agent can use AI for scripts, objection replies and list prep, with the do-not-call checks that stay with a person.
• Best Free AI for Realtors: Three Jobs, the Limits and When to Pay — Three jobs a realtor can do on free AI plans, a fair-housing check for every output, and the signs that a paid plan is worth it.
• AI Role-Play for Insurance Agents: Practise Objections First — Run objection practice with an AI playing the client: prompts by product line, a scoring rubric and a compliance-wording check.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OpenAI's voice FAQ; Anthropic's voice mode help page; Google Gemini plans and Gemini Live pages; Google's help page on the transition from Gems to skills; OpenAI's custom GPT retirement FAQ; Anthropic's Projects help page; pricing and product pages for Yoodli, Exec, Hyperbound, SalesGo, BrokerBot and MaverickRE; a Follow Up Boss article on real estate role-play scenarios (all rechecked 4 November 2026).