How to run a real-estate pipeline with ChatGPT and a Google Sheet (Tuesday morning, 12 minutes)
You don't need a CRM. You need a prompt and a Tuesday.
The 12-deal agent at Compass Hendersonville isn't running their pipeline in kvCORE. They're running it in iPhone Favorites, a text thread with their cousin's friend, and a Google Sheet labeled "BUYERS - DO NOT DELETE." NAR's 2025 Member Profile says the median REALTOR did 10 sides last year for $58,100 gross. NAR's 2025 Technology Survey says CRMs only produce 23% of quality leads — the other 77% comes from sphere, referrals, and repeat business. So the median agent is paying $499 a month for software that owns a quarter of their pipeline. The pillar piece — the three-gate framework — covers when that math flips. This page is what you do until it does.
I'm going to show you how to run your real-estate pipeline in 12 minutes a week using ChatGPT Plus, your phone, and a Google Sheet.
What you need before you start
- ChatGPT Plus or Claude Pro — $20 a month, cancel anytime
- An iPhone or Android with your text threads intact
- A Google Sheet (one tab: active deals, columns for name, type, status, last contact, next action)
- Your sphere as a CSV — exported from wherever you've been keeping it
That's it. Total stack cost: $20 a month. Total setup time: 30 minutes the first Tuesday, 12 minutes every Tuesday after.
Step 1 — Export your sphere as a CSV
Wherever your contacts live now, get them out. The CSV is your portable system of record. Five minutes, one time.
- Follow Up Boss — People > select all > Actions > Export. CSV downloads with name, phone, email, tags, last contact.
- kvCORE / BoldTrail — Smart CRM > Contacts > filter > Export to CSV. If you're on a paid plan and considering canceling, do this before the contract ends.
- Top Producer — Contacts > Tools > Export Contacts > CSV format.
- Google Contacts only — contacts.google.com > Export > Google CSV. This works fine for sphere-driven agents who never adopted a real CRM.
- Just your iPhone — Settings > Contacts > Export, or run the contacts through a free CSV exporter app. Messy, but workable for under 200 names.
Open the CSV in Google Sheets. Add a column called "type" (buyer / seller / sphere / past client). Add a column called "last contact." Add one called "next action." That's your active-deals tab. The rest of the sheet is reference.
Step 2 — Open ChatGPT Plus and paste your Context Card
Open chat.openai.com. Start a new chat. Paste a short version of your Context Card so the model knows who it's working for. The full Context Card spec is in the brand context card, but the inline brief version below is enough for this workflow.
Paste this first, before anything else, in the same chat thread:
You are my real-estate operations partner. I am a Hendersonville, TN
agent at [BROKERAGE]. I do roughly 12 sides a year, sphere-and-referral
driven. Williamson County and Sumner County primarily — Old Hickory
Lake waterfront, Cool Springs, downtown Franklin.
Voice: short, warm, no exclamation marks, no "just checking in," no
emojis. I write like a person who knows the client, not a marketer.
Do not say: "circle back," "touch base," "synergy," "leverage,"
"unlock," anything that sounds like a SaaS pitch.
When I paste raw message threads, read them all — don't summarize.
When you draft texts, write in my voice, not yours.
Confirm by repeating my voice rules before we start.
The model will confirm. Now you're loaded. The Context Card is sticky for the whole chat — every prompt that follows inherits it.
Step 3 — Paste this exact prompt every Tuesday morning
This is the workflow. One prompt. Same prompt every week. Update the names and the top-of-mind details, paste the last 14 days of messages, hit enter.
You are my real-estate operations partner. I'm a Hendersonville, TN
agent. I have 4 active buyers, 2 active sellers.
Below is the raw text of every message thread from the last 14 days
across SMS, email, and DMs. Don't summarize — read it all.
Then give me:
1. A ranked call list for today. Who do I phone first, second, third,
and why. One sentence each.
2. Who's stalling. Anyone whose energy has dropped, whose replies are
getting shorter, who hasn't been touched in 7+ days.
3. The "weak signal" flag. Someone I keep mentioning in messages but
never actually call. The neglected one. Be honest.
4. Three follow-up texts ready to send — written in my voice (short,
warm, no exclamation marks, no "just checking in").
Top of mind: offer due Friday on the Wilson buyers. Maria's Old Hickory
listing is 15 days in. Andersons have been quiet.
[PASTE FULL THREADS HERE — names, dates, content, all of it]
To paste your threads on iPhone: open Messages, tap the conversation, press and hold any message, tap More, select all, copy. Repeat for each active thread. Paste them into the prompt under the bracket. Don't clean them up. Don't redact. The model needs the mess to do the work.
Eight minutes of paste time. Four minutes of reading the response. Twelve minutes total.
Step 4 — What you get back, and what to do with it
The response comes back in roughly 90 seconds. Here's what it looks like for the agent above, on a real Tuesday.
- Wilson buyers — offer due Friday. Call the listing agent at 9:30 to confirm the inspection contingency timing. Highest priority.
- Maria, Old Hickory seller — 15 days on market, no offers. The price-drop conversation is overdue. Call this morning, don't text it.
- Hampton buyers — cold for nine days. Either revive them with a real listing match or kill politely. Draft text below.
- The Andersons — weak-signal flag. You've mentioned them in three weeks of messages. You've never actually called. They're the riskiest neglected lead in your pipeline.
Then three drafted follow-up texts in your voice — the Hampton revival, a Maria check-in note before the call, and a soft re-open to the Andersons.
What you do next: open your Google Sheet. Update last-contact and next-action for every name on the call list. Make the four calls. Send the three drafts after a quick read. The whole cycle — paste, read, update, call — runs about 35 minutes including the calls themselves.
That's the entire system. Twelve minutes for the AI work, the rest is the part you were always going to do.
What this costs versus kvCORE
- This workflow: ChatGPT Plus, $20 a month. No setup fee. No 12-month contract. Cancel anytime. Total year-one cost: $240.
- kvCORE solo: $499 a month plus a $1,000 setup fee. Twelve-month contract. Total year-one cost: $6,988.
The kvCORE dashboard would have shown you what you logged. Manual data entry is the number-one driver of CRM abandonment, per MetaData Corp's 2025 analysis — week 3 onward, agents stop entering, and by week 6 the pipeline is back in the phone. ChatGPT reads what you actually said in your messages and surfaces what's missing. No data entry. No dashboard nobody opens.
The $6,748 you saved by not buying kvCORE is roughly 11% of the median REALTOR's gross commission income. Reinvest it in actual prospecting. A door-knocking campaign in Cool Springs or a postcard run on Old Hickory Lake will move more deals than the dashboard would have.
Why this works (the weak-signal frame)
A CRM is a database. You log a contact, you log a note, you log a status. The dashboard shows you what you logged — nothing more. If you didn't log it, the dashboard doesn't see it. That's the structure of a database, and that structure is exactly why CRM hygiene collapses by week 6 for most solo agents.
A foundation model is the opposite category of tool. You hand it raw text — messy, unstructured, names and dates and half-thoughts — and it reads what's there. Andrej Karpathy framed this as Software 3.0 on X in 2025: English is the new programming language, and the model is the runtime. You skip the data-entry step entirely. The Andersons get flagged not because you typed "neglected" into a CRM field — you didn't — but because the model read three weeks of messages where you mentioned them and noticed you never picked up the phone. The CRM tracks what you wrote down. The foundation model thinks alongside what you actually said. From the team that wrote What We Learned from a Year of Building with LLMs (Yan, Husain et al., O'Reilly 2024): "Don't buy SaaS for what an LLM can do." That's the line. If the value is the thinking, you don't need the wrapper.
This is Owned-Data AI in practice. You own the messy text already. The model reads it. No vendor sits between you and your own data.
When this DOESN'T work
This workflow has three specific failure modes. They're the same three gates from the pillar's three-gate framework — clear any one and you've earned a real CRM.
- You're closing 25+ sides a year. Above 25, the cognitive load stops fitting in a notes app and a weekly prompt. Things start slipping in the seams between deals. That's when database structure starts paying for itself. Below 25, the model handles it — and 75–80% of working REALTORS sit below this gate per the NAR 2025 Member Profile.
- A second human touches your pipeline — a salaried ISA, a transaction coordinator, a buyer's agent on your team. The CRM exists to coordinate humans, not transactions. If a TC needs to know what's happening with the Wilson buyers without you forwarding texts, you need shared visibility. ChatGPT in your personal account can't give them that.
- Paid leads exceed 30% of your pipeline. Zillow Flex, Realtor.com Connections, paid social funnels at scale. Above 30%, speed-to-lead becomes the constraint — and the Lead Response Management Study (Oldroyd, MIT — 15,000 leads, 100,000 call attempts) found a 21x drop in qualification odds when response time slipped from 5 minutes to 30. You can't run sub-5-minute response by hand. You need automation. That's a real CRM with lead scoring and routing.
Below all three? This workflow wins. Above any one? Read the pillar — the three-gate framework — and shop for a real CRM.
Bookmark and rerun
Save the prompt. Save the Context Card brief. Pin the Google Sheet on your phone. Run it every Tuesday morning between 9 and 9:30, before the first appointment. Twelve minutes, one cup of coffee, a ranked call list with 30-second openers ready to send.
When you've cleared the gates and you want the full operator stack — Context Cards tuned to your voice, the prompt library, predictive seller scoring workflows, the AI-Enhanced Realtor credential — that's what The Listing Machine operationalizes.
Until then, save the $400 a month. Make the four calls. Run the prompt next Tuesday.
HowTo schema (JSON-LD)
Sources
Primary data
Independent builder/operator creators
Vendor pricing
Last updated 2026-04-29.