Start with people who already know you or have already asked about a property. Use AI to write personal check-ins to your sphere and past clients, publish local answers buyers and sellers search for, and run a website chat that captures enquiries after hours. Each lead arrives with consent you can prove, so no bought or scraped list is needed.
AI makes sending easy, and the rules do not care who wrote the message. Marketing texts sent through an automated tool can need prior written consent, commercial email needs an opt-out you honour within 10 business days, and a list of strangers carries neither. So the plan below sorts every contact by the consent you hold before it picks a channel, and only then asks the assistant to write anything.
Which channel each kind of contact can receive
The channel decides the rule, and your records decide which channels a contact qualifies for. The summary below comes from the published telemarketing rules (the TCPA) and the commercial-email rules (CAN-SPAM), checked on 4 November 2026. Treat it as a briefing for your brokerage's compliance contact, not legal advice; your brokerage may set stricter limits than the rules do.
| Channel | What must be true first | Build into the routine |
|---|---|---|
| Commercial email | The rules set conditions rather than asking permission first: accurate sender details, an honest subject line, a clear ad label where the message is an ad, your postal address and a working opt-out. Business-to-business email gets no exemption. | Honour every opt-out within 10 business days (do it the same day). Penalties run up to $53,088 per email. |
| Texts sent by a tool or AI | Prior express written consent for automated marketing texts: a signed agreement (an electronic signature counts) that lists the number, says automated marketing texts are authorised and says consent is not a condition of buying or selling. Whether a given tool counts as automated dialling equipment can be disputed, so the safe routine is to collect this consent for every marketing text. | Honour STOP and any other reasonable opt-out within 10 business days. Keep the consent record. |
| Calls you dial yourself | Either the number is not on the do-not-call registry, or you have an established business relationship: up to 18 months after a transaction, up to 3 months after an enquiry. | Check the registry at least every 31 days, keep your own do-not-call list and honour it for 5 years, and call only between 8 a.m. and 9 p.m. in the contact's time. |
| AI voice calls | The same written consent as prerecorded calls, because AI-generated voices count as artificial voices (the regulator ruled so in February 2024). | Read the voice-agent follow-up tutorial before switching one on. |
| Replies inside your website chat | Answering a visitor inside the chat needs no extra consent. A later automated text or AI call does, while a live call you dial yourself falls under the 3-month enquiry window above. | Put the consent wording beside the phone field. |
A few details trip agents up. A relationship ends the moment someone asks you to stop calling, whatever the windows say. The registry rules include a defence for calls to family members, friends and acquaintances, but a person who handed you a business card at an open house two years ago is not an acquaintance in any safe reading. And a past client who once said "text me" in conversation has not signed anything. Written consent is a document, so the fix is a consent line on paper or in a form, never memory.
Your CRM therefore needs three fields for every contact: email opt-out status, text consent (with the date and the exact wording agreed) and a do-not-call flag. If you are still choosing a system, make those fields part of the test; the CRM comparison for agents lists the contenders.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowA solo agent's 240 contacts, sorted into channels
Take an illustrative solo agent who exports 240 contacts from a phone, an old spreadsheet and a CRM. Two columns turn the pile into something usable: the date of the last transaction or enquiry, and a consent column recording what each person has signed. An assistant can tidy phone formats and merge duplicates (the older tutorial on cleaning messy data covers that step), but the consent column is filled in by the agent from paperwork. The model never guesses it.
| Group | Contacts | Email with opt-out | Live call | Automated text |
|---|---|---|---|---|
| Closed with the agent in the last 18 months | 46 | Yes | Yes, inside the 18-month window | 29 who ticked the consent line at closing |
| Closed more than 18 months ago | 31 | Yes | Registry check first | 9 with a signed consent |
| Enquired but never transacted | 58 | Yes | The 14 who enquired in the last 3 months; the other 44 only if clear of the registry | 6 who ticked a form |
| Friends, family and neighbours | 105 | Yes | Personal calls | None until they sign |
| Total | 240 | 240 | About 60 inside the windows | 44 |
Read the table as three lists. All 240 people can receive email that carries the footer. About 60 (46 recent clients and 14 recent enquirers) fall inside the relationship windows for a live call. Only 44 can receive an automated text today. The first quarter therefore has one structural goal besides the touches themselves, which is to grow that 44: a consent line goes into the closing paperwork, every web form and the open-house sign-in sheet, and each new signature is logged with its date.
The quarter is one touch per contact, 20 a week for 12 weeks, matched to the group. Recent clients get a personal email with the sold prices for their own street. The 14 recent enquirers get a phone call. Older clients get a handwritten note that the assistant drafts and the agent copies out by hand. The neighbours get a short email inviting questions about the market. Weekly time is about 30 minutes of drafting with the assistant and an hour or so of sending and calling, and every reply is logged with a one-line note that feeds the next quarter's check-ins.
Triggers make the touches feel timely. Keep three date fields in the sheet (the closing anniversary, a birthday if the person offered it, and a follow-up date from your notes) and sort by them each Monday. A spreadsheet formula finds who is due; the assistant only writes the message. Asking the model itself to work out who is due is the step where it miscounts.
Here is the sum for the quarter, using invented figures to show what to track rather than what to expect: 240 touches, 31 replies, 9 conversations about moving or introducing someone, 3 listing appointments and 1 signed listing. Tracking matters because each number has a source. If all 3 appointments came from the group with recent enquiries, the next quarter's effort goes there, not into a fresh purchase of names.
Check-ins that sound like you: a prompt, its output and the edits
The usual failure with AI check-ins is sameness: twenty messages that open "Hope you're well!" and are recognised as templates at once. The cure is to give the assistant one real fact per contact from your own notes and forbid it from adding any. It does not need phone numbers or email addresses to write a note, so paste a first name and the note only; your CRM merges the details when you send.
You write short check-in messages that a real estate agent sends to people she already knows.
Rules:
- Use only the facts in the contact note and the local fact I give you. Never invent details about the person, their home or the market.
- Email: 30 to 60 words. Text: under 300 characters, ending with "Reply STOP to opt out."
- No exclamation marks, no sales pitch, no "just checking in".
- One specific reference from the note, one useful offer, one easy way to reply.
- Mark the text version CONSENT REQUIRED. I send it only to contacts with a tick in the text-consent column.
Contact note: [first name], bought a 3-bed in March last year, redoing the kitchen, has a dog.
Local fact I can use: [for example, the sold prices on her street from April to September, taken from my own MLS report]A typical first draft looks like this (illustrative). It ignores the prompt's rules against invented facts and sales pitches:
Hi [first name], I was thinking about your kitchen project today, because two homes on your street sold this month and both had updated kitchens. Homes like yours are in strong demand right now. If you'd like a free valuation, just reply and let me know!
Three things have to go. "Two homes on your street sold this month" is a fact the agent never supplied; the assistant invented it to sound specific. "Strong demand right now" is a market claim with no source behind it. And the free valuation with its exclamation mark turns a relationship message into a pitch, which is the opposite of what the note was for. The edited version keeps the kitchen, offers the real data and asks one question:
Hi [first name], how did the kitchen turn out? I'm sending my past clients the sold prices for their street from April to September, with nothing attached to it. Want yours? [your name]
The matching text, for contacts who ticked the consent line, reads: "Hi [first name], it's [your name]. How did the kitchen turn out? I can send the latest sold prices for your street if useful. Reply STOP to opt out." That is 29 words and carries the opt-out. If your tool adds the STOP line automatically, leave it in the template anyway, because the tool can be changed and the habit should not.
Local answers that bring in enquiries
The questions clients ask before they call are the best raw material for content, because they are real and specific to your market. Pull 10 to 15 from the last month: emails, direct messages, open-house conversations, anything you answered twice. For each, record a two- or three-minute voice memo answering it as you would to a client, then give the assistant the transcript, your verified local facts and the format you want: the answer in the first 50 words, three or four headings and a closing offer. Putting the answer first helps a reader, and an AI search tool, get it without scrolling.
| Question from a client | What you supply | What you check afterwards |
|---|---|---|
| How much cash do I need to buy? | A closing-cost range from your lender partner, with its date | The range is current and sourced; the page promises no mortgage rate |
| Is now a good time to list? | Last quarter's sold count, median days on market and price range from your MLS report | Every figure matches the report; forecasts are deleted |
| What should I do before listing? | Your own pre-listing checklist, in order | Contractor claims and local requirements are accurate |
| What are the schools like on Maple Court? | Distances and links to the schools' own pages | No quality judgements and no steering language |
Never let the assistant supply a statistic. Give it the figure and the date range, tell it to quote them unchanged, and delete any number that is not in your report. First drafts often contain a line such as "prices in the area rose 6.2% last year" with no source. The fix is to replace it with the median sold price for three-bedroom homes in the third quarter, copied from the MLS report and with the period named, or to delete the sentence.
Assistants also drift into steering language on neighbourhood pages. Fair-housing law makes it unlawful to publish any advertisement or statement about a home that indicates a preference or limitation based on race, colour, religion, sex, disability, familial status or national origin. A draft that says "Maple Court is a quiet, family-friendly pocket, popular with young professionals and close to the best schools" describes people, not property. The fix describes the place: "Maple Court is a cul-de-sac of 24 homes, 0.4 miles from the nearest primary school and 0.9 miles from the high street (distances from a mapping tool, checked on [date])." The neighbourhood pages tutorial goes deeper on avoiding steering.
End each page with one capture: "Get the sold prices for your street", a short form and the consent sentence from the next section. A reader who wants that page has asked for it, which is consent you can point to.
A website chat that captures enquiries at 11 p.m.
A chat widget earns its place by doing three small jobs outside office hours: answering the five questions you get most from text you have approved, asking three qualifying questions, and recording consent beside the phone number. It must never give a price opinion, describe who lives in an area or promise a result. The longer treatment of chat design is in the chatbot best-practice tutorial; this section covers consent and handover only.
Visitor (10:42 p.m.): How much is a 3-bed worth on Maple Court?
Chat: I can't give a price opinion because that needs a person who knows the street. I can book a 15-minute valuation call with [agent] or send the sold prices for the last six months. Which would you like?
Visitor: Send the sold prices.
Chat: Happy to. Are you thinking of buying, selling or just curious?
Visitor: Selling in the spring.
Chat: Thanks. What's the best email for the sold prices? If you'd also like [agent] to call or text you tomorrow, add your number and tick the box.
[name] [email] [phone] [ ] consent boxThat transcript is illustrative, and the behaviour to copy is in the first reply: the bot declined the price question and offered two real next steps. Test for it by asking your own bot what a house is worth; a bot that answers with a figure needs its instructions tightened. Four more checks catch most problems:
- The consent box is unticked by default, and a submission without a tick gets an email reply, never an automated text.
- The lead reaches your CRM with the transcript, the consent wording and a timestamp.
- The bot tells the visitor when to expect a reply ("[agent] will reply by 9 a.m."), and you keep that promise.
- Once a month you chat at night as a stranger would, and fix the weakest answer.
The consent sentence beside the phone field can read as follows. Ask your compliance contact to approve your own version, and save the exact wording with the date and time each person agreed.
[ ] I agree that [agent name] of [brokerage] may call and text me at the number above about my enquiry, including with automated technology. I understand consent is not a condition of buying or selling a property, and I can reply STOP to texts at any time.Open-house sign-ins and referral partners
Print a consent line at the top of the sign-in sheet: "By signing, I agree that [agent] may email, call and text me, including with automated technology, about this home and future listings. Consent is not a condition of buying." After the open house, photograph the sheet and ask the assistant to transcribe it into a table. Handwriting is where it slips. On an illustrative sheet of 23 names it read 21 rows cleanly and misread digits in two phone numbers, so the agent checks every number against the photo before anything is sent, and deletes the photo from the chat once the table is saved.
The follow-up goes out within a day and mentions the home the person saw. Give the assistant the two fields the sheet already holds (how soon they hope to move, and whether they work with an agent) and ask for a four-sentence email that asks what the home was missing. That question produces replies because it asks for an opinion rather than a commitment.
Referral partners work differently. A financial adviser who meets clients six to twelve months before they buy is worth more than a long list of strangers. The assistant can draft a one-page note on what you do for a partner's clients and a quarterly market note built from your sold-prices report, with one paragraph of commentary only you can write. Partner email is still commercial email, so it carries the same footer, and your note should ask the adviser to check with a client before passing their details to you.
What "no scraped lists" rules out, plus four vendor questions
Scraping means collecting names, emails or phone numbers from listing portals, social profiles or public records with software. Avoid it for four reasons. The people have no relationship with you and have consented to nothing. The platforms' own terms usually forbid it; LinkedIn's user agreement, for example, bans using crawlers, plugins or bots to scrape or copy profiles and other data. The data is often out of date, so you contact the wrong person. And an unrequested message to a stranger is exactly what gets reported as spam.
Services that promise "AI-found leads" often resell data like this, so put four questions to the vendor before paying:
- Where and when did each contact give consent, and can you show me the form and the date?
- Does the wording name my business or only "partners"? The written agreement has to authorise the seller to deliver the calls and texts, so a box that names another company may not cover you.
- How do stop requests reach me, and how fast?
- What happens to my account and my costs if the data proves non-compliant?
Check your MLS's data-use rules before using any export for outreach, and if you still want to call cold, the cold-calling tutorial covers scripts and do-not-call rules.
The weekly routine and the numbers worth recording
| Task | When | Time |
|---|---|---|
| Pick 20 contacts due a touch and draft the messages with the assistant | Monday | 30 minutes |
| Send emails, make calls, write notes | Tuesday to Thursday | 60 to 75 minutes |
| Log replies and update the consent and do-not-call fields | Friday | 15 minutes |
| Publish one local answer page (voice memo, draft, fact check) | Monthly | 45 to 60 minutes |
| Read chat transcripts and fix one answer | Monthly | 20 minutes |
Record six fields for every contact so the numbers mean something: source, consent status with date and wording, last touch, whether they replied, appointment date and outcome. After a year you can answer the only question that matters, which is which source produced signed listings and what each one cost in time.
The money side is small. ChatGPT Business is $20 per seat a month billed annually ($25 billed monthly) with a two-seat minimum, and Claude Team is $20 or $25 on the same terms. Individual plans cost less: Claude Pro is $17 a month billed annually or $20 monthly, and OpenAI lists the ChatGPT Plus price on its pricing page. A solo agent usually starts on an individual plan and keeps phone numbers and email addresses out of the prompts; a team should use a business plan, which does not train on your content by default. For the larger picture of follow-up automation, the tutorial on automating buyer and vendor follow-up shows where a CRM takes over from the assistant.
Questions agents ask before switching on AI follow-up
Can I text past clients without asking again?
Not safely. Marketing texts sent through a texting tool or by AI can count as automated telemarketing, which needs prior express written consent even from people you have worked with, and whether your tool counts is not a point to argue later. The 18-month and 3-month relationship windows cover live calls, not automated texts. Add a consent line to closing paperwork, text only people who signed it, and honour STOP within 10 business days.
Is it safe to paste my client list into ChatGPT or Claude?
Give the assistant first names and short notes, not phone numbers, email addresses or home addresses, and let your CRM merge those at send time. For a team, ChatGPT Business and Claude Team do not train on your content by default. On a free account, check the data settings and your brokerage's policy before pasting anything about a client.
Can an AI chatbot call or text a lead from my web form?
Only if the form carried consent wording that names you, mentions automated calls and texts, and says consent is not a condition of buying or selling. AI-generated voices count as artificial voices under the telemarketing rules, so calls need the same written consent as prerecorded ones. Store the consent record with the lead and route STOP replies back to your CRM.
What if I have no sphere of influence yet?
Start with the sources that do not need one: local answer pages, website chat and open-house sign-ins that carry a consent line. Add one habit: after every conversation, log the person with a source, a consent field and a note. Within a year that spreadsheet becomes your sphere, and every row has a source you can name.
Further reads
- AI Marketing for Real Estate Agents: A Four-Week Content Plan — A four-week content calendar to feed the local answers described here.
- AI Real Estate Social Media: A Month of Posts From Your Listings — Turn your listings and sold prices into a month of posts.
- Leasing Chatbot for Real Estate: A Fair Housing Checklist Before You Pay — Fair-housing checks for any chat that talks to prospects.
- Best ChatGPT Plan for Realtors: Free, Plus or Business? — Which assistant plan suits a solo agent or a small team.
- Follow Up Boss Pricing: What a Small Team Really Pays — What a small team really pays for a CRM that stores consent.
- ChatGPT for Realtors: Brokerage Rules for Emails, Disclosures, Follow-Ups — Brokerage rules for AI-written emails, disclosures and follow-ups.
- Claude or ChatGPT for Realtors: Which Assistant Fits Your Day? — Five realtor tasks to run on both assistants, what each plan costs, how each reads a 90-page disclosure pack, and the privacy settings that matter.
- 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.
- AI Automations for Real Estate Wholesalers: Sourcing to Contracts — A step-by-step map of what to automate in a wholesaling business, where a person must stay in the loop, and the consent and disclosure rules.
- AI Sales Role-Play for Solo Real Estate Agents: Free Tools Compared — Which free AI tools can play a seller or buyer, what each free tier allows, and an objection bank and weekly routine to practise with.
- ChatGPT for Estate Agents: Rules for Vendor Updates and Feedback — For the branch or brokerage owner: the rules, templates and fact checks that let agents use ChatGPT for vendor updates and viewing feedback safely.
- Open House Sign-In Sheet With AI: Follow-Up Within the Hour — A printable sign-in sheet, consent wording, buyer-type message templates and a 60-minute plan for personal follow-up after every open house.
- AI Policy for a Real Estate Brokerage: What Agents May and May Not Do — Turn a broker AI policy into rules an agent can follow on a Tuesday: do and don't lists, sample clauses and a sign-off sheet.
- Can Loan Officers Use AI Videos for Market Updates? Rules to Follow — Disclosure rules for AI avatar videos written before the first render, with a script template, caption checklist and review gate for loan officers.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: the published telemarketing rules (TCPA implementing rules), the CAN-SPAM compliance guide for business, the do-not-call FAQ for telemarketers and sellers, the 2024 ruling on AI-generated voices, the fair-housing advertising provision, LinkedIn's user agreement, and the ChatGPT Business and Claude pricing pages. Checked 4 November 2026.