The agent who runs into a former client at the supermarket and learns they sold with someone else didn’t have a relationship problem. They had a follow-up problem. The client liked them. The client also didn’t hear from them for fourteen months and assumed they had moved on. That story repeats in every market and the cause is almost always the same: the agent has too many people in the database to keep up with manually, no system to prioritise who needs contact this week, and no sustainable way to write personal-sounding messages at the volume their database actually demands.
The NAR 2025 Technology Survey shows 21% of agents using CRMs with AI insights and 68% of agents using AI in some form. The two figures together suggest that most AI use among agents has not yet reached the relationship management work, which is precisely where the biggest sustained edge is sitting.
Pick a CRM with AI baked in, not a CRM you bolt AI onto
You can use ChatGPT or Claude to write outreach for an exported spreadsheet. For ongoing relationship management you want a CRM with AI insights built in: Lofty, Follow Up Boss, or kvCORE. These platforms ingest contacts, interactions, and website behaviour and surface who is heating up, who has gone cold, and who matched a specific listing. The AI sits inside the workflow rather than next to it. If you’re not on one yet, export to spreadsheet monthly and run the segmentation work in ChatGPT. The workflow below works either way.
The segmentation prompt that beats your gut feel
Most agents segment their database by tier (A, B, C) based on a vague memory of who is closer to a transaction. AI segmentation produces a more useful answer because it can hold dozens of variables at once. Export your CRM to a CSV with whatever fields you have (last contact date, last property purchased, source, area of interest, notes, response history) and run:
“Below is a CSV of [N] contacts in my real estate database. Without inventing data, segment them into the following groups based on what is in the file: (1) hot prospects, defined as anyone who has interacted in the last 30 days or is actively looking, (2) warm past clients within the last 24 months who should hear from me at least quarterly, (3) cold past clients beyond 24 months who deserve a value-led re-engagement attempt, (4) referral-source contacts who are not buyers themselves but have introduced me to others, (5) dormant contacts with no engagement in 12+ months and no clear path to a transaction, who I should consider archiving or running one final re-engagement on. For each segment, give me the count, the criteria you used, and three specific contacts who are good examples. If a contact’s information is too thin to classify, put them in a ‘needs more info’ bucket and tell me what’s missing. Data: [PASTE OR ATTACH CSV].”
The “needs more info” bucket is the most useful output of this prompt. It tells you which contacts have rotted into uselessness because nobody has ever logged a meaningful note. Spend a Friday afternoon adding context to those records, even if it’s two sentences from memory.
The personalisation prompt that doesn’t sound like a template
The reason most “personalised at scale” outreach gets ignored is that it actually isn’t personalised, the merge fields just hide it for thirty seconds until the recipient notices the email is identical to the one their neighbour got. The fix is to give AI specific context about the contact and ask for variation, not just a name swap.
“I am writing a check-in message to a past real estate client. Here is the context: I sold them [HOUSE TYPE] in [SUBURB] in [YEAR]. They were [PROFILE: e.g., a young family upsizing, a couple downsizing into a townhouse, an investor purchasing their second rental]. The thing I remember about them personally is [ONE OR TWO SPECIFIC DETAILS, e.g., they had a Labrador called Max, the wife was changing jobs, they were nervous first-time buyers]. The market in their area right now is [BRIEF MARKET CONTEXT]. Write three different short check-in messages, each under 80 words, that I could send by SMS or email. Each version should reference the specific personal detail naturally, mention the market context only if it serves them, and end with a low-pressure invitation to reply. None of the three should sound like a sales pitch and none should sound like a template. Vary the openers fully so they don’t feel like the same message rephrased.”
Run this once per past client when their anniversary or a relevant market event triggers a touch. It takes ninety seconds. The output beats anything you would have written cold under time pressure.
A specific failure mode to watch for: AI will hallucinate personal details if you don’t give it any. If you ask for a “personal” message without supplying the specifics, the model will invent a reference to “your beautiful daughter Emma” and “the renovation project we discussed last spring” because those phrases pattern-match to what personal messages look like. Always verify that every personal detail in the output matches a real one you supplied.
The annual review the high-performing agents send
Here is the move that separates an agent from a referral source. Once a year, send each past client a one-page personalised market review of their own home. Not a generic suburb report, an estimated value of the specific property they bought, with a comparison to what they paid, and the local market context. AI does the synthesis in two minutes once you supply the comps and market context.
“Generate a personalised home value letter for a past client. They purchased [ADDRESS] in [YEAR] for [PURCHASE PRICE]. Recent comparable sales: [PASTE 3-5 COMPS WITH SALE PRICES AND DATES]. Local market context: [SUBURB-LEVEL TRENDS]. Write a warm letter that opens with a personal acknowledgement, gives an estimated current value range with one line on how I arrived at it, shares two market insights, and closes with a no-pressure offer to dig deeper. Tone: helpful neighbour, not sales pitch. Under 350 words.”
The counterintuitive insight from agents who do this annually: the response rate is high not because recipients are about to sell, but because they forward the letter to friends. The letter becomes a conversation piece, which is what referral generation actually looks like in 2026.
Where the human still has to show up
AI is now reliably good at segmenting, drafting, and personalising at the scale a typical agent’s database demands. It is not good at noticing when a past client’s life has changed in a way that should change how you communicate with them. The new baby. The job loss. The death in the family. Those signals come from human conversations and Facebook scrolling, not from your CRM, and they are the ones that matter most. The AI workflow gets you to ninety percent. The remaining ten percent is the part that makes a client feel like they are not just on a list, and that part is still you.