AI for real estate agents: 7 jobs to hand over in 2026

Real estate agents get the most from AI by handing it the work around the deal: replying to every enquiry within minutes, booking viewings into the diary, answering the phone during a showing, drafting listing descriptions for approval, following up past clients and valuations, and triaging tenant repairs. Most agents already use AI, but mostly to write. In NAR's 2026 technology survey, 75% of agents who use AI use it for listing descriptions. The bigger prize is speed: in a secret-shopper study of US brokerages, nearly half of online inquiries got no reply at all. Below are the seven jobs worth handing over, what stays with you, the fair housing and consent rules to respect, and a sample five-agent office with the hours it gets back.

How are real estate agents using AI in 2026?

More often than a year ago, and mostly for words. NAR's 2026 Technology Report, released on 22 September, found that nearly half of Realtors use AI daily or weekly:

Chart

How often Realtors use AI (US, 2026)

• Use AI daily23%
• Use AI weekly25%
• Experimenting now and then31%
• Curious, but haven’t tried it9%
• Not using AI, and no plans to12%

Show the data as a table

How often	Share of Realtors
Use AI daily	23%
Use AI weekly	25%
Experimenting now and then	31%
Curious, but haven’t tried it	9%
Not using AI, and no plans to	12%

The tools are general assistants. Chicago Agent's summary of the same survey puts ChatGPT at 91% of AI users, up from 58% a year earlier, with Gemini at 42% and Copilot at 27%. Saving time is the main reason agents adopt any new technology (81%). Looking ahead, the same summary reports that 48% are interested in AI for lead generation and follow-up over the next year.

That is the gap. Writing a listing description faster saves half an hour. Answering a buyer at 8:47pm, before another agent does, can decide the whole deal.

How fast should a real estate agent respond to a lead?

Within minutes, whatever the time. Two numbers explain why:

• Most buyers only talk to one agent. In NAR's 2024 Profile of Home Buyers and Sellers, 67% of first-time buyers and 77% of repeat buyers interviewed only one agent before choosing.
• Many enquiries are never answered. Mike DelPrete's 2024 secret-shopper study made more than 100 online inquiries on median-priced homes across over 25 US brokerages' own websites, all during weekday business hours. Nearly half (47%) got no response. The median reply took 39 minutes, and the average 8 hours and 17 minutes.

So the agent who replies first, with something useful, often wins the buyer by default. Here is the same enquiry handled two ways:

TUESDAY · ONE ENQUIRY

BY HAND

1. 20:47Website enquiry: “Is 14 Alder Lane still available? Could we see it this weekend?”
2. 20:47Nobody sees it. The agent is at a valuation, then home.
3. 08:40Agent replies: “Thanks for your interest! When would suit you?”
4. 11:15Buyer: “Thanks, we’ve booked to see it with another agent on Saturday.”

TUESDAY · THE SAME ENQUIRY

WITH AN AI ASSISTANT

1. 20:47Website enquiry: “Is 14 Alder Lane still available? Could we see it this weekend?”
2. 20:48Reply in the office’s own voice: yes, it’s available. Two questions: are you pre-approved, and do you need to sell first?
3. 20:52Buyer: pre-approved, nothing to sell, hoping to move by spring.
4. 20:53Offers three Saturday times from the listing agent’s calendar. The buyer picks 10:30.
5. 20:54Sends the written buyer agreement to sign before the tour, and a reminder for Friday.
6. 20:54Lead, answers and booking saved in the CRM. The agent gets a three-line summary.
7. 08:40The agent rings to say hello, with a confirmed viewing already in the diary.

Nothing in the AI column is clever. It answers, asks the two questions an agent would ask, offers real times from the diary, and takes care of the paperwork NAR now requires before a tour. The agent arrives in the morning to a booked viewing instead of a cold lead.

7 jobs to hand to AI in a real estate business

These are the jobs where AI saves the most time or wins the most business, what you keep, and where we cover each one.

Start with job 1. It touches money directly, it is easy to measure (reply time and booked viewings, before and after), and it fixes the leak the secret shoppers found.

What to keep for yourself

Three things stay human, and clients notice when they don't:

1. Pricing and negotiation. AI can pull comparables, but the advice on what to list at, and what to accept, is the job clients pay you for.
2. Final approval of anything public or signed. Listing copy, ads and any document a client signs. You are responsible for it, whoever drafted it.
3. The hard conversations. A sale falling through, a probate sale, a tenant in arrears. These need a person from the first minute.

Is there a ChatGPT for real estate agents?

Not one, but three kinds of tool cover it:

• A general assistant (ChatGPT, Claude or Gemini) on a business plan for writing, research and emails. Our comparison of ChatGPT, Claude, Gemini, Grok and Muse sets out what each plan includes and which ones train on your data.
• The AI already in your software. Many real estate CRMs and portal tools now include AI features for drafting messages and suggesting follow-ups. Switch on what you already pay for before buying anything new.
• A private assistant on your own documents, which answers from your listings, tenancy agreements and office procedures, with sources. Here is how that works, and why it keeps client data out of public chatbots.

Some brokerages go further. In September NAR profiled brokers who used Claude to build saved workflows for MLS descriptions, their own dashboards, and lead and CRM tools. NAR's own caution in that piece is the right one: have an AI policy, and keep a person reviewing the output.

Is it legal for realtors to use AI?

Yes. The rules apply to what the AI writes and does, exactly as if you had done it yourself. Four matter most. None of this is legal advice; check your state's rules and your broker's policy.

Fair housing. The Fair Housing Act makes it unlawful to publish a listing or advertisement that indicates a preference or limitation based on race, colour, religion, sex, disability, familial status or national origin. HUD's 2024 guidance on AI and digital advertising was withdrawn in a Federal Register notice on 6 April 2026, but the same notice says conduct that breaks the Act is still enforced. AI drafts can drift into describing the buyer ("ideal for a young family") rather than the home, so the safe rule is: describe the property, not the people you imagine living in it, and have a person approve every listing.

Calls and texts. The FCC ruled in February 2024 that an AI-generated voice counts as an "artificial" voice under the Telephone Consumer Protection Act. Outbound AI-voiced calls need the person's prior express consent, and prior express written consent for marketing. Replying to someone who has just contacted you is different from calling a list, but take advice before any outbound calling or texting campaign.

Buyer agreements. Since 17 August 2024, under NAR's settlement practice changes, an MLS participant working with a buyer must have a written agreement before touring a home, including live virtual tours. An AI that books viewings should send that agreement as part of the booking, as in the example above.

In the UK, estate and letting agents must include material information in listings, as set out in National Trading Standards' guidance. Part A covers price or rent, council tax and tenure. Part B covers what applies to every property, such as type, construction, utilities and parking. Part C covers issues such as flood risk, where they apply. AI drafts must not leave these out. Marketing texts and emails also need consent under PECR, and applicant and tenant data falls under UK GDPR.

Will AI replace real estate agents?

No, and the data points the other way. In NAR's 2025 Profile of Home Buyers and Sellers, a record 91% of sellers used an agent, 88% of buyers did, and sales by owner fell to 5%, an all-time low. More than half of agents (55%) say AI has helped their business. AI is taking the replies, scheduling and paperwork. People still choose a person for the biggest purchase of their lives.

A sample five-agent office: what it gets back

Here is how we would estimate the value for a fictional office with five agents, 60 online enquiries a week (about 240 a month) and 8 new listings a month. The minutes per task are assumptions to replace with your own.

About 39 hours a month is roughly a working week, every month. The bigger number may be the enquiries themselves. If this office looked like DelPrete's sample, about 113 of those 240 enquiries a month would get no reply at all. With an assistant, every one gets an answer within minutes, at any hour.

One more job belongs on every agent's list: protecting the closing. Fake wiring instructions sent just before closing are a form of business email compromise, which cost US victims who reported it to the FBI $3.05 billion in 2025. Our case file of real deepfake attacks on businesses shows how those scams arrive, and deepfake and AI scam protection covers the rules that stop them.

What we would do

For most offices we start in this order, and measure each step before the next:

1. Enquiries first. Connect your website, portal and email enquiries to an assistant that replies within minutes, qualifies, and books into the right agent's calendar, with the buyer agreement step built in.
2. Then the phone, so calls during showings and after hours are answered and booked, not lost to voicemail. Our guide to the cost of missed calls shows how to put a number on what you're missing now.
3. Then the writing: listing drafts with a fair housing and material information check, and follow-up drafts you approve with one tap.
4. For property managers, maintenance triage within each landlord's limits.

Everything connects to the CRM, calendar and portals you already use. If you're not sure which job to start with, AI consulting begins with that question, and you can ask for a written quote once you know.