Virtual Assistant or AI for a Real Estate Agent: What Goes Where

Start with an AI assistant for work that is words you will read before they leave: listing copy, follow-up emails, social posts and call summaries. Hire a virtual assistant for work that needs someone to act: chasing signatures, tracking deadlines, scheduling showings and updating the CRM. Most agents end up with a VA who uses AI.

The cost gap explains the order. An AI seat costs about $20 a month, while a VA at 20 hours a month costs several hundred dollars, so the cheap tool should prove itself first and show you what is left. Keep three tasks yours whatever you buy: pricing advice, negotiation and reading contract terms.

Follow me on Instagram@sagnikteaches

Sorting the week: which tasks go to AI, which to a person

The test for each task is what the output is. If it is a draft you will read before it goes out, AI is a fast first pass. If it needs someone to phone, wait, enter data or notice a missing signature, a person has to do it. Many tasks need both, with AI drafting and a person sending or checking. This table sorts 18 common tasks that way.

Connect on LinkedInSagnik Bhattacharya

The pattern in the right-hand column matters more than the sorting. Almost every AI row ends with you checking a fact, and almost every VA row ends with you confirming a date, a sum or a promise. Neither helper removes your accountability; each removes a different kind of typing. The tutorial on AI assistant against virtual assistant compares the two for any business, and the transaction coordinator tutorial goes deeper into the paperwork rows.

Subscribe on YouTube@sagnikteaches

Where AI drafts go wrong in a property business

AI is confident about things it was never told. In property copy the damage is concrete: an invented feature misleads a buyer, and a phrase about who a home is "perfect for" can breach advertising rules on discrimination that apply where you work. Give the assistant a facts sheet and forbid anything outside it. This prompt does that:

Write a 90-word listing description from the facts below. Use only these facts: add no features, no neighbourhood claims and no words about what kind of buyer the home suits. If a detail is missing, write [confirm] instead of guessing.

Facts: 3 bedrooms; 2 bathrooms; 1,450 square feet; kitchen refitted in 2019 with quartz worktops; south-facing garden; single garage; gas central heating; open-plan living and dining room.

Here is a typical reply, illustrative but realistic, and what to fix:

Three-bedroom, two-bathroom home of 1,450 square feet with an open-plan living and dining room. The kitchen was refitted in 2019 and has quartz worktops. Outside are a south-facing garden and a single garage. Gas central heating throughout. [confirm: boiler age]

Two things need attention. "Throughout" wasn't in the facts, because you never said every room has a radiator, so strike it. And the [confirm] flag is correct behaviour; resolve it or delete it before publishing. Now picture the same assistant with no facts sheet, asked simply to "write a listing for a family home". Unconstrained drafts often fill the gaps with lines such as "a stunning, newly renovated kitchen with granite worktops, in a quiet and safe neighbourhood perfect for young families". Set that against the facts: a 2019 refit called new, granite where the worktops are quartz, a safety claim nobody can verify, and a phrase about the buyer. The facts sheet turned four errors into one. For the wider choice of assistant for listings and follow-ups, see Claude against ChatGPT for real estate agents.

Where a VA goes wrong, and the handover that prevents it

People fail differently. A VA rarely invents things; they miss things, because a task wasn't written down or a date had two possible meanings. The commonest failure is an unclear deadline: the VA diarises the closing date when the date that mattered was the inspection deadline a week earlier. The cure is a written checklist for each recurring event, with the agent confirming dates from the signed contract first. Here is a filled-in one for an accepted offer.

Offer accepted: VA checklist (agent confirms dates from the signed contract first)
Day 0: agent sends the signed contract and key dates in one message; VA enters each date in the shared calendar with two reminders
Day 1: VA emails the other agent, the lender contact and the legal representative to confirm roles and deadlines
Day 2: VA books the inspection and sends access details to the client
Every 3 days: VA sends the agent one line per open item: done, waiting, or at risk
VA never: interprets contract terms, gives a date to a client without agent sign-off, or edits a contract

The "never" line is the most useful one. It tells the VA where the boundary is, so they stop and ask instead of improvising. The same handover habit works for the open-house follow-up; the sign-in sheet tutorial shows a version for it.

A solo agent's month, staffed four ways

The example agent works alone with around four deals in progress. Admin takes 66 hours a month: 22 on writing (listings, posts, follow-ups, newsletter), 6 on call notes, 12 on CRM and data entry, 16 on transaction coordination and 10 on scheduling. I assume AI takes writing from 22 hours to 9 and call notes from 6 to 2.5, counting your review, and leaves the other three groups alone. A VA at $20 an hour takes the data, coordination and scheduling work: 38 hours, and I add 4 hours a month of your time for briefing and checking the VA. These are assumptions to replace with your own log, not measurements.

The last row assumes the VA's own use of AI trims coordination from 16 hours to 13 and scheduling from 10 to 8, so the VA works 33 hours, not 38. Read the table in two passes. The AI-only row is almost free per hour, but it frees only 16.5 hours because the coordination and data work still need a person. The VA rows free far more hours at a cost a little above the VA's own hourly rate, because your supervision time counts against them. Adding AI for your own writing costs $20 and frees another 16.5 hours, so the cash per hour freed drops from $22.35 to $15.45, and letting the VA use AI trims it to $13.47. That is why starting with AI is rational: these are the cheapest hours you can free, and they show how much work is left for a person.

The VA rate matters. At $35 an hour, an illustrative rate for someone with transaction experience, the fourth row costs $1,350 and the fifth $1,175, or about $26.73 and $23.27 per hour freed. That can still be worth paying if the freed hours go to clients, but it is no longer a rounding error, so test whether those hours get used on clients and not absorbed by more admin. The published rate guides are wide: goLance's October 2026 guide shows $20 to $35 for mid-level assistants and $35 to $55 for senior ones, while a managed-assistant agency's July 2026 guide shows $14 to $22 mid-level and $25 to $60 for specialists. Treat both as ranges set by sellers.

For the AI column, ChatGPT Plus is $20 a month and Claude Pro is $20 monthly or $17 billed annually. Business plans such as ChatGPT Business and Claude Team list at $25 per seat monthly, or $20 on annual billing, and each needs two seats, so a solo agent pays about $40 to $50 for a business plan. The choice between plans is covered in the tutorial on choosing a ChatGPT plan as an agent, so I won't repeat it here.

A typical week once both are working

It helps to see how the handoffs fit together, because most friction comes from not knowing who has the ball. This is one workable rhythm for the agent in the example.

• Monday morning. You send the VA three lines: this week's priorities, any new accepted offers with their dates, and anything not to touch. The VA updates the shared calendar and the CRM before lunch.
• Tuesday. After weekend open houses, the VA has typed the sign-in sheets into the CRM. The AI drafts a follow-up for each visitor from your template and the notes. You read them in one sitting, correct two, and the VA sends the batch.
• Wednesday. New listing day. You paste the facts sheet into the assistant for the description and captions, check every claim against the listing, and hand the final text to the VA to post and schedule across channels.
• Thursday. The VA's three-day status line arrives for each open transaction: done, waiting or at risk. You deal only with the "at risk" ones. The AI turns your call notes from the week into CRM summaries, and the VA checks that each landed on the right contact.
• Friday. A 15-minute call. The VA lists what was late, what was unclear and what they'd like to take on next. You fix the checklist that caused the confusion before the weekend, so it doesn't recur.

Notice what you do each day: a short brief on Monday, a read-and-approve pass on Tuesday and Wednesday, an exceptions check on Thursday and a checklist fix on Friday. The supervision is small but real: the Monday brief, the Thursday exceptions check and the Friday call come to about 35 minutes a week, and with questions in between I count 4 hours a month in the table above. The read-and-approve passes on Tuesday and Wednesday are already inside the 11.5 hours of AI-assisted writing and call notes.

What goes into a chat window, and what never does

Client data is where agents get nervous, and the rules are simple enough to write on one page. Start with the tool's own terms. OpenAI says it doesn't train its models on data from ChatGPT Business, Enterprise or its API by default. Anthropic says the same of its commercial products, and adds that feedback you choose to send, such as a thumbs-down report, can be used. On personal plans of both, whether chats are used for training is a privacy setting you have to find and check yourself: OpenAI's help centre says "Improve the model for everyone" is on by default for ChatGPT Free, Plus and Pro, and Claude's personal plans have a Model Improvement switch under Privacy Settings. So a business plan is the safer home for client material, and a personal free account is the wrong place for it.

Then decide what a prompt may contain. A workable rule has two lists.

• Fine in a business-plan chat: a property's public listing facts, a client's first name and general preferences, anonymised notes such as "Buyer A wants three bedrooms and a garden", and your own template wording.
• Never in any chat tool: identity documents, bank or card details, signed contracts with financial terms, loan pre-approval letters, anything about a client's health or finances beyond what the task needs. If a task needs a figure, paste the figure, not the document.

Use placeholders by habit: [Buyer A], [Seller B], [Address]. It costs seconds, it makes the output reusable, and it means a slip exposes less. Check what your brokerage requires too, because its policy may be stricter than any vendor's terms. For a VA, the VA's side of this is covered in the confidentiality checklist for virtual assistants using AI.

Terms to settle before a VA sees a client file

A VA touches more client data than an AI chat does: contact details, contract dates, sometimes documents. Put the basics in writing before day one. This wording is a starting point to adapt, not legal advice, so have a lawyer review the final version.

Confidentiality: all client information is confidential and used only for the tasks assigned.
Tools: client data goes only into tools the agent provides or approves. No personal free AI accounts, no personal cloud storage, no screenshots kept on personal devices.
Access: individual logins only; two-step verification on; passwords held in the agent's password manager; access removed on the last day.
AI use: the assistant may use AI tools for drafting only inside the agent's approved workspace, and must not paste identity documents, financial documents or signed contracts into any AI tool.
Return: on exit, the assistant deletes or returns all client data within 7 days and confirms in writing.
Incidents: any suspected leak or lost device is reported to the agent the same day.

Two of those lines do most of the protecting. "Individual logins only" gives you a record of who did what and lets you cut access cleanly. "Approved workspace" means you pay for the AI seat that applies the privacy defaults, instead of hoping the VA's own account has the right switch set. Review the terms again whenever you add a tool.

Choosing the first ten VA hours

A new VA's first hours should go on work that is repetitive, low-risk and governed by clear rules, because that is where mistakes cost least and a checklist works best. Score each candidate task on hours a month, risk if it goes wrong (1 low to 3 high) and how clear the rules are (1 vague to 3 exact). Hand over the high-hours, low-risk, clear-rules tasks first. An illustrative scoring for one agent:

The first three add up to 13 hours a month of work, so a 10-hour trial covers most of it and the rest can wait for the second month. The enquiry replies are the biggest block of hours but the worst fit at the start: a slow or wrong reply to a hot lead costs more than the VA saves. Write the templates first, then move the task across once the VA has handled the low-risk work well for a month.

A 30-day trial, week by week

You don't need to decide now. A month of structured testing costs about $20 for the AI seat and 10 paid hours for a VA, and gives you real numbers in place of my assumptions.

1. Days 1 to 3: log the week. Write down every admin task and its minutes for three ordinary days. Sort them with the table above into writing, data, coordination, scheduling and calls. Multiply up to a month and compare with the 66-hour example.
2. Days 4 to 14: AI first, on five tasks. Pick listing copy, follow-up emails, call summaries, social posts and the market update. Build a one-page facts sheet and a style note, run each task through the assistant, and log minutes with review plus every correction you made.
3. Days 15 to 24: a VA on three tasks. Hand over transaction reminders, showing scheduling and CRM clean-up, using the checklist and the written boundaries. Pay for 10 hours. Log turnaround, questions asked and errors found.
4. Days 25 to 30: combine and decide. Let the VA use your approved AI workspace for reminder drafts, then compute cash per hour freed for each route.

Fill in a scorecard like this at the end. The figures below are illustrative.

The "keep going if" column is my rule of thumb, so adjust the thresholds to your own business. What matters is that you set them before the trial, because a month of effort makes almost any result look acceptable afterwards. If the VA misses the thresholds but the AI passes, keep the AI and stay on your own admin until the handover notes are better. If the AI misses them, tighten the facts sheet before blaming the tool.

When the month ends, write down the three tasks you will never hand over, the five you will always hand to AI first, and the three that go to the VA, then review the list quarterly. For a wider rollout across several agents, the first-month plan for an agency, listed below, covers roles and tools.

Questions agents ask before hiring help

Can my VA put client details into their own AI account?

Only if your agreement says so. Require that client data goes only into tools you provide or approve, preferably a business plan that doesn't train on content by default, and ban personal free accounts for client files. Write that into the agreement and check it again at the 30-day review.

Is an AI voice tool better than a VA for calls?

They do different jobs. A VA can confirm showings and call people back using your approved wording. AI answering tools raise questions about call consent, disclosure and handover that deserve their own review. Start by listing which calls are routine and which need you in person, then match each to the cheapest safe option.

Do I need to tell clients I use a VA or AI?

Check your brokerage's policy and any rules on advertising and data use that apply where you work. As a habit, state in your engagement terms that assistants and software tools help with admin and that you review everything yourself. Never let an AI-written message pass as a personal reply to something sensitive.

How many hours should a first VA engagement be?

Ten hours a month for the first 30 days, on three tasks you've written down, is enough to learn whether the person is reliable. Add hours or a retainer only after you've compared turnaround, errors and cost per hour freed. A big first commitment locks you in before you know the task list.

Further reads

• AI Marketing for Real Estate Agents: A Four-Week Content Plan — A four-week content plan once AI is doing the drafting.
• Claude or ChatGPT for Realtors: Which Assistant Fits Your Day? — Choosing the assistant you will use first.
• AI for a Real Estate Agency: Roles, Tools and a First-Month Plan — Roles, tools and a first-month plan for an agency.
• Best AI CRM for Real Estate Agents: Follow Up Boss, Lofty, BoldTrail — The CRM your VA and AI will both work in.
• How to Automate Buyer and Vendor Follow-Up With AI — Follow-up automation for buyers and vendors.
• AI Receptionist for Realtors: Cost, Call Routing and Showing Cover — Call cover for showings, if calls are your bottleneck.
• AI Receptionist for Estate Agents: Cost, Call Flow and Handover — Cost bands for AI versus human phone cover, plus a call flow for valuations, viewings and applicants and clear handover rules.
• AI Virtual Assistant Services: How a VA Business Can Package Them — Three package tiers, the pricing sums, scope limits and disclosure wording for a VA business selling AI-assisted work.
• AI Voice Agent for Real Estate Follow-Up: Setup, Scripts, TCPA — Consent wording, a pre-dial gate, call windows, a sample script and per-minute costs for AI follow-up calls to real estate leads.
• Follow Up Boss AI: Which Plan Adds Smart Summaries and Suggested Tasks? — All three Follow Up Boss plans list smart summaries and suggested tasks, but the AI leans on calling data. Plan-by-plan features and when Pro beats Grow.
• Follow Up Boss Pricing: What a Small Team Really Pays — A six-person team on Follow Up Boss pays about $648 a month on Grow with calling or $499 on Pro. The full price ladder, annual savings and side costs.
• 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.
• AI Tools for Real Estate Brokers: From Transactions to Agent Support — Five tool types for a small brokerage, each with published costs and watch-outs, a worked cost for 12 agents and a 90-day adoption plan.
• Best AI for Real Estate Marketing: Ads, Email and Listing Video — Pick an AI tool by job for real estate marketing: ad copy, email nurture sequences and listing video, with fair housing and disclosure checks.
• Best AI for Real Estate Wholesalers: Underwrite a Deal Before You Call — Use AI to screen a wholesale deal before the seller call: comps, ARV and repair sanity checks, a maximum offer sum, and the checks you still do by hand.
• ChatGPT Prompts for Virtual Assistants: Inbox, Scheduling, Reports — Eight ChatGPT prompts for inbox triage, scheduling and client reports, each with a sample output and a client-approval step.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: ChatGPT Business and Claude pricing pages, OpenAI's business data page and Anthropic's privacy centre article on model training (all read 1 November 2026); published virtual assistant rate guides from goLance (October 2026) and Stealth Agents (July 2026), used only as ranges.