Every lead answered in seconds, including the ones at midnight

No single one wins, because the category covers four different jobs: lead response, content writing, valuation and transaction admin. Pick by the job that is costing you money. If enquiries sit unanswered for hours, fix lead response first and ignore everything else until it works.

For most individual agents, a general assistant such as ChatGPT, Claude or Gemini for writing, plus the automation already inside your CRM. A custom agent only pays off when volume genuinely outruns a person, which usually means a team or a brokerage rather than a solo agent.

Several products wrap a general model in a real estate interface, and agents use ChatGPT itself for drafting. The gap is not the writing. A chat window cannot see your CRM, your calendar or your MLS feed, so it cannot act. That connection is what makes something an agent.

Four clusters, in rough order of adoption. Writing: listing copy, emails, social posts. Answering: website chat and instant replies to portal leads. Admin: summarising calls, drafting follow-ups, cleaning the database. Analysis: comparable properties and market questions. The first three are settled practice. The fourth still needs checking.

Yes, and the booking rules matter more than the connection. Before launch you set how far ahead it may book, the buffer between showings, which agents cover which areas, and what happens when nobody is free. It handles the confirmation, the reminder and the reschedule too.

It can draft them from the property record, and a person must read every one before publication. Listing copy is advertising, and advertising is where fair housing language rules bite hardest. Draft by machine, approve by human, every time, with no exception for a busy week.

Usually it changes what they do rather than removing them. The agent takes first response, basic qualification and the long nurture, which is the grinding part. Your people take the conversations needing judgement, market knowledge and a relationship. Teams that clear the ISA desk on launch day tend to regret it.

An informal coaching habit rather than an official rule, used a few different ways. The common version is a follow-up cadence: contact a new lead three times in three days, then three times in three weeks, then three times in three months. Agents intend to run that. Software actually does.

Follow Up Boss, Lofty, Sierra Interactive, kvCORE and BoldTrail, Real Geeks, Wise Agent, plus general platforms such as HubSpot and Salesforce. Where a documented API exists we use it. Where it does not, we say so during scoping rather than discovering it three weeks in.

Yes, through the proper door. Your MLS display rules decide what may be shown, to whom, and how fresh it has to be, and we read those rules before writing the integration. We do not scrape portals, which breaks quietly and usually violates a licence.

Anything that produces a record: your own website forms, Zillow and Realtor.com enquiries, Homes.com, social lead forms, open house sign-ins and referral submissions. The important part is that every source is tagged accurately, because you cannot judge a lead source you cannot separate.

It may not steer. The Fair Housing Act makes it unlawful to publish any statement about the sale or rental of a dwelling indicating a preference or limitation based on a protected characteristic, and an automated message is a statement. So the agent never describes an area by who lives there, and never answers demographic, crime or school questions.

Three layers, none of them a prompt. A classifier catches protected-characteristic questions before generation. The tools themselves cannot return demographic data. Every outbound message is logged so your broker can audit what was said. Flagged requests get one neutral reply and an immediate handoff to a licensed person.

No, and it refuses clearly rather than hedging. Pricing advice, contract terms, repair negotiation, disclosure questions, tax and mortgage advice all leave the agent immediately with a summary attached to the record. Those are licensed conversations, and several of them are legal or financial advice no automated system should give.

You are accountable, exactly as you would be for an assistant or a script, which is why the design points are boundaries, approvals and logs. No AI system is right every time. Good design makes mistakes visible, small and correctable rather than silent. Every run is replayable from the log.

It threatens the parts of the job that are typing and chasing, and little else. Nobody hires an agent because they are good at sending follow-up texts. The real risk is not being replaced by AI, it is competing against a team that answers in seconds while you answer in hours.

Not currently. Transactions still involve licensure, fiduciary duty, negotiation, disclosure and a large emotional purchase, and NAR itself frames the REALTOR® as the human in the loop for AI-assisted tasks. What is being replaced is the administrative middle of the job, which most agents are happy to lose.

For research, yes. AI is good at explaining process, comparing areas on published data and drafting questions. For the transaction itself you need someone licensed and accountable. AI carries no fiduciary duty, cannot hold earnest money, will not attend the inspection and cannot be sued for bad advice.

Commission rates are not set by law or by any association and never have been. They are negotiable between the parties, and since the 2024 practice changes buyers sign a written agreement with their agent before touring, while offers of compensation are no longer published in the MLS. An AI agent should never quote a rate.

A single queue with a CRM and a calendar connected is usually a matter of weeks, then a few more weeks in shadow mode before it acts alone. Adding MLS data and voice lengthens it. What moves the timeline most is access: API credentials, sample conversations, and one decision maker.

We will not put a figure on a page, because a number written without seeing your setup is aimed at an average rather than at you. Scope depends on how many queues and systems, whether those systems have clean APIs, whether MLS data is involved, and how much conversation design you need.

Yes. The repository, the connectors, the prompts, the evaluation sets and the cloud accounts are yours. There is no proprietary runtime you lose access to if you stop working with us, and nothing an ordinary engineering team could not maintain.