AI Tools for Solo Real Estate Agents: You Hold the License

This article used to end with the line "This stack has been tested in real solo agent workflows over the past year." It wasn't. That claim contradicted the article's own body, which already said the agent results we once published were invented and removed. It has been deleted rather than reworded — and it is the ninth unverified testing claim we have found on this site, this one hiding in a closing italic disclaimer nobody thought to audit.

So here is what this article is instead. The tool recommendations below are unchanged in substance, and every price was re-checked at the vendor's own page on 1 September 2026. What's new is the part no other roundup covers.

Every "AI tools for real estate agents" article is the same six tools. Ours was too. But a solo agent isn't a solopreneur who happens to sell houses — they are a licensed solopreneur, and that changes what "use AI to write your listings" actually means. When we went and read the statute, a state regulator's advisory, an MLS license and two state AI laws, we found that three of the recommendations in our own article collide with rules that bind the agent and not the tool.

The rule underneath all of this

The Fair Housing Act's advertising provision, 42 U.S.C. § 3604(c), makes it unlawful:

"To make, print, or publish, or cause to be made, printed, or published any notice, statement, or advertisement, with respect to the sale or rental of a dwelling that indicates any preference, limitation, or discrimination based on race, color, religion, sex, handicap, familial status, or national origin..."

Read the verbs. Make, print, publish. The statute never asks who wrote the sentence. There is no authorship exception and no tooling exception. And there is no intent requirement — the test is what a reasonable reader takes the ad to suggest, not what you meant.

The California Department of Real Estate put it in one sentence in its advisory on AI in real estate, dated 17 March 2026:

"The use of an AI tool will not excuse the violation."

The same advisory adds that "responsibility under current law rests with the licensee and their responsible broker, not the technology provider," and that using AI for licensed activity "may be equivalent to asking an unlicensed assistant to do licensed activity."

That is a state regulator saying, in terms, that the tool is not a defence. Hold that thought through the next three sections.

Collision 1: AI writes exactly the phrases you can't publish

AI listing copy is the highest-ROI use of these tools. It is also the one with a compliance problem built into it.

Language models learned to write listings from decades of listings — including all the ones written before anyone was careful. So the phrases they reach for first are the flagged ones:

None of these read as discriminatory to the person writing them. That's the problem — 24 CFR § 100.75 reaches "words, phrases, photographs, illustrations, symbols or forms," and the test is the reasonable reader's perception.

The fix is one line in your prompt, and it's the single most useful thing in this article:

Describe the property, not the person who should live in it.

That test resolves nearly every case. "Family-friendly" describes an occupant; "large fenced backyard" describes the property. "Walking distance to the park" describes the resident's legs; "three blocks from the park" describes the property. "Safe neighborhood" describes an inference; "cul-de-sac, no through traffic" describes the property.

Put it in the prompt itself, not in your head:

Write a 200-word MLS description from these walkthrough notes.

Hard rule: describe the property, never the type of person who should
live in it. No references to families, children, schools, professionals,
retirees, religion, ethnicity, or neighborhood "safety" or "character".
Use measured distances, not "walking distance". State only facts present
in my notes — do not infer or embellish square footage, features, or
neighborhood details.

Then list any phrase you used that could be read as describing an ideal
occupant, so I can check it.

That last instruction matters more than the rest. It makes the model mark its own risky phrases, which turns the review step from "read it again carefully" into "check these four."

Then read it anyway. You are the publisher. Federal civil penalties for a first offense run to tens of thousands of dollars — the exact figure is set by regulation and adjusted annually for inflation, and we could not open the current schedule at its official source, so we're not quoting a number. State license discipline is separate and, for a solo agent, worse.

There's a structural reason this lands harder on you than on an agent at a big brokerage. Nearly all the published guidance on AI in real estate is written for brokerages: approve a tool list, appoint a compliance lead, require human review before publication, train the agents. About three-quarters of NAR members work solo. If you're under a broker you're still bound by their policy; if you are the broker, nobody is writing one. Either way the review step lands on the same desk as the drafting step — which is precisely the arrangement the guidance is designed to prevent.

Collision 2: the chatbot is pointed at the steering question

We recommend Tidio for 24/7 lead capture, and we still do. But our own setup instructions used to say to train it on "pricing, neighborhoods, school districts."

Think about what that means. A bot answering "is this a good area for families?" or "is the neighborhood safe?" is answering the steering question — unprompted, at 2am, in writing, with no human in the loop and a permanent transcript. NAR's Standard of Practice 10-1 specifically prohibits volunteering information about neighborhood racial, religious or ethnic composition in residential transactions.

Scope the bot to what it can safely do:

• Safe: listing facts already public in the MLS, square footage, price, lot size, days on market, availability, booking a showing, capturing contact details, general financing basics
• Escalate to you, always: anything about neighborhood character, safety, schools, demographics, "what kind of people live here", plus offers, contract terms and anything that sounds like advice

Write the escalation as the default, not the exception. A bot that says "that's a good question for me to answer personally — what's the best number for you?" has captured the lead and stayed inside the line. We covered the general setup in our AI customer support guide; the scoping above is the real-estate-specific part.

Two more things about bots that no tools roundup mentions. Texas's Responsible AI Governance Act took effect 1 January 2026 and requires that businesses "disclose to consumers that they are interacting with an AI system before or at the time of interacting with the AI system," in language that is "clear, conspicuous and written in easily understandable language." It applies to anyone "offering a product or service used by Texas residents." California's SB 1001 gives a safe harbour on the same terms: disclose that it's a bot and you're not liable under that section. Colorado's Chatbot Safety Act adds a similar disclosure duty from 1 January 2027.

Turning on the "I'm an automated assistant" line is a thirty-second setting. Leaving it off is a state-law problem in at least two states.

Read the free tier carefully — we didn't, the first time. Tidio's free plan includes 50 Lyro AI conversations "one-off", in Tidio's own wording. Not 50 a month. It's a trial allowance. Starter is $29/month billed monthly, or $24.17/month billed annually ($290/year). Beyond the free 50, Lyro is a separate add-on starting at $32.50/month for 50 AI conversations — $0.65 each, bought as a tier rather than per conversation. We previously published a flat $0.50 rate, which is a floor Tidio quotes at volume, not what a solo agent pays.

Collision 3: "train it on your listings" may breach your MLS agreement

Our old instruction was to "train Tidio on your active listings so it answers questions about the colonial on Elm Street correctly." That sounds harmless. It may not be.

MLSs are actively tightening license terms around AI. Metro MLS in Milwaukee now states plainly, in language NAR published as a model:

"Metro MLS data may not be used for AI training or machine learning in any way without Metro MLS's express written permission."

NAR's IDX policy separately requires listing displays to remain "under the actual and apparent control of the licensed participant." The industry argument about whether Zillow's ChatGPT integration satisfies that is still running.

The distinction that matters is training versus querying. Asking a model a question about text you paste in, where nothing is retained, is a transient task. Uploading your MLS export into a tool's knowledge base so it "learns" your listings is the thing licenses are being rewritten to prohibit. A consultancy that follows this closely described the practical failure mode in August 2026: data leaves the closed system through an authenticated session with "no consent step. No license. No audit trail."

This is a patchwork mid-tightening, not a settled national rule. Some agreements address AI explicitly; many still don't. So the honest instruction is not "this is illegal" — it's: read your MLS's data license before you paste listing data into any tool, and if it's silent on AI, ask before assuming silence means yes. Your own listings are a much safer starting point than the full feed.

The photo rule California agents keep missing

If you use Canva's AI background remover, generative fill, or any virtual staging, and you operate in California: AB 723 has been operative since 1 January 2026. It adds § 10140.8 to the Business and Professions Code and requires that a digitally altered marketing image carry a disclosure that is "reasonably conspicuous and located on or adjacent to the image", plus "a link to a publicly accessible internet website, URL, or QR code that includes, and clearly identifies, the original, unaltered image."

It applies everywhere your marketing appears — your site, social, print flyers — not just the MLS. Adding furniture, removing power lines or greening a lawn counts. Lighting, cropping and colour correction don't. The bill makes violations crimes under the Real Estate Law; sources disagree on penalties and the text we could read named no figure, so we aren't quoting one.

Outside California, NAR's Article 12 duty to "present a true picture" and most MLS rules already require virtual staging to be labelled.

The stack itself

Unchanged in substance, prices re-verified 1 September 2026 at each vendor's own page.

We previously totalled this at "$42/month", which never reconciled with its own table. Run it honestly: everything free except Make is $9/month on Make's annual billing ($10.59 month to month). A realistic paid configuration — Tidio Starter, ChatGPT Plus, Canva Pro, Otter Pro and Make Core, each billed annually where that's offered — is about $74/month; billed monthly throughout, it's about $95. Neither number is $42, and the old one was carried forward rather than re-derived.

HubSpot Free gives you 1,000 contacts, 2 users and one deal pipeline. We previously said it has "no workflow automation at all"; HubSpot's pricing page shows one automated action on free, which isn't the Workflows tool but isn't nothing. The pipeline maps cleanly onto a transaction funnel, and email logging is automatic. We compared it properly in our HubSpot Free vs Notion CRM piece.

Real-estate-specific CRMs start higher than we used to imply. Follow Up Boss is $69/user/month, or $58 annually — but calling is a $39/user add-on, so a solo agent who wants a dialler is at $108/month. kvCORE and Sierra Interactive both gate pricing behind a demo; figures circulating for them come from comparison sites, not vendor pages, so treat them as reported rather than checked.

Otter is free for 300 monthly transcription minutes. Pro is $16.99/month monthly but $8.33/month annually — we previously quoted only the monthly figure.

Make Core — $9/month billed annually, $10.59 monthly — is what ties it together: Tidio lead into a HubSpot contact, showing transcript filed against the buyer, check-ins scheduled after closing. We covered the patterns in our client follow-up automation guide.

One headline that will mislead you this year

In January 2026 HUD proposed rescinding its disparate-impact regulations, with a supplemental proposal in August. If you saw that headline, you might reasonably have concluded fair housing exposure has eased.

It hasn't, for two reasons. First, the proposal is not final — it was a notice of proposed rulemaking with comments closing 13 February 2026. Second, and more importantly, disparate-impact liability doesn't live in HUD's regulation. The Supreme Court confirmed it exists under the Fair Housing Act in Inclusive Communities; rescinding a regulation doesn't repeal a statute or overrule a precedent.

And none of it touches § 3604(c), the advertising provision that actually governs your listing copy. That text is unchanged.

The advertising rules that bind you go wider than fair housing — the per-advertisement identification duties, California's new altered-image statute, and the consent rules on AI voice calling, which carry by far the largest penalties of anything here. We took that apart separately in AI real estate advertising rules: what's real in 2026, including the four state laws we could find no evidence of.

What we corrected in this article

In the spirit of the audit that produced it:

• "70% of an experienced agent's business comes from referrals." No NAR figure says that. The real medians for agents with 16+ years are 49% repeat business and 32% referrals from past clients — separate medians that can't be added. Two-thirds of sellers (66%) use an agent they were referred to or had worked with before.
• "Leads contacted within 5 minutes are 9x more likely to convert." Real study — the Lead Response Management research conducted at MIT and published via InsideSales.com, covering 15,000+ leads. But it's from 2007, and it's about B2B sales calls, not real estate. The mechanism is believable; the multiplier isn't a real-estate finding and we shouldn't have presented it as one.
• "~4 hours per week on admin" and the $42 total — never measured and never re-derived. Both gone.
• A virtual assistant at "$1,500-3,000 a month" — unsourced. Gone.

One more thing worth knowing, from NAR's own guidance: AI-generated work is generally not protectable by copyright, because US law requires a human author. Your AI-written listing description probably isn't yours in the way you'd assume — a competitor who copies it has no copyright problem. We went through that whole question in who owns AI-generated work.

For the general stack, see the 2026 solopreneur AI stack. And if you're weighing a marketing-specific writing tool instead of a general one, our Jasper review explains why we don't think it fits a one-person business.

---

Every price re-verified at the vendor's own pricing page on 1 September 2026. Legal material quoted from primary sources where they could be opened — the statute, the CFR, California's AB 723 bill text and the DRE advisory — and attributed to secondary sources where they could not. This is not legal advice; your state licensing rules and MLS agreement govern. Links marked "Paid link" earn us a commission if you sign up through them, at no extra cost to you; no other link here pays us — how we make money.