Screen 500 Properties in 30 Minutes: A PropGPT Buy-Box Walkthrough

If you have ever tried to screen a market by scrolling through 500 listings on a consumer portal, you already know the pattern. You open a tab. You check the price history. You open another tab for the county assessor. You guess at owner equity. You note it in a spreadsheet. Then you do it 499 more times.

That workflow is not broken because investors are slow. It is broken because the data you need to make a triage decision lives in three or four separate systems, and each tab switch is a small tax on attention. The point of a buy-box is to specify exactly what you want once, then push the filtering down to a tool that can hold all the relevant signals in one place.

This walkthrough shows how to take a written buy-box and turn it into a ranked shortlist in PropGPT using three prompts. The goal is not to replace your underwriting. It is to get from a wide candidate pool to the 20 or 30 addresses that deserve real attention, quickly.

Start with a buy-box, not a search bar

Before you touch the chat, write your buy-box down in five dimensions. This is the same discipline acquisitions teams use internally, and it is what makes a prompt specific enough to return usable output.

• Market: one metro, a set of ZIP codes, or a polygon. Be specific. "Atlanta" is not a market; "ZIPs 30314, 30315, 30318" is.
• Price: a band, not a ceiling. A band forces the tool to exclude obvious mispricings on both ends.
• Beds and baths: minimums are usually enough. Most investors care about 3/2 or better.
• Condition: a proxy, since no database labels this directly. Year built, days on market, and price cuts are reasonable stand-ins.
• Equity: the signal most portals bury. You want owners with meaningful equity, because that is who can actually transact.

A complete buy-box looks like this:

Single-family, 3+ bed / 2+ bath, $180k to $320k, ZIPs 30314/30315/30318, built before 1990, at least one price cut or 60+ days on market, owner equity above 40 percent.

That sentence is the input to everything that follows.

The three-prompt workflow

The workflow has three stages: pull, enrich, rank. Each stage is a single prompt. You do not need to engineer clever phrasings. You need to be precise about what you want back.

Stage 1: pull the candidate grid

The first prompt takes the full buy-box and asks PropGPT to return a structured grid of candidates. In practice the chat returns a 48-row property grid with photos, list price, beds/baths, square footage, days on market, and year built. For a three-ZIP area with a reasonable price band, 40 to 80 rows is typical. If you get 200, tighten the band. If you get 5, loosen the condition proxies.

This is already faster than a portal because the grid is inline. You are not clicking into each listing to read the stats.

Stage 2: enrich with owner and equity data

This is where the workflow separates from Zillow. Consumer portals do not show you the owner of record or estimated mortgage balance. PropGPT can append them to the grid you already have. You ask it to add owner name, owner-occupied flag, last sale date, last sale price, estimated current value, and estimated equity percentage.

The output usually reveals that a third of your initial list is owner-occupied with low equity, which means they cannot sell without bringing money to closing. Another chunk is recent buyers with thin equity. What you want is the tail: long-tenured owners, absentee in many cases, with 40 percent or more equity. That is your shortlist.

Stage 3: rank by composite score

Ranking is where judgment enters. You tell PropGPT which signals matter most and let it produce a weighted score. A reasonable default weighting:

• Equity percentage: 40 percent
• Days on market: 20 percent
• Price cuts: 15 percent
• Owner tenure: 15 percent
• Price relative to ZIP median: 10 percent

The chat returns your rows sorted by composite score with the top 20 at the top. You export, and you have a call list.

Why this is faster than a portal

The time savings come from three places, not from raw search speed.

First, you skip the tab tax. Owner records and equity estimates are inline with the listing grid rather than one search away at the county assessor.

Second, you define the filter once. A buy-box written in plain language is reusable. The same paragraph that worked for Atlanta works for Birmingham with two edits.

Third, the ranking is explicit. Most portal users eyeball a sort by price or days on market and hope for the best. A composite score makes the weighting visible, which also makes it easy to argue with and adjust.

The 30-minute figure is not a marketing number. It assumes you already know your buy-box. If you are still deciding whether you want 3-beds or 4-beds, that is a separate exercise, and no tool will save you from it.

What to watch for

Two failure modes show up often.

The first is over-filtering on condition proxies. Year built and days on market are noisy signals. A 1985 home that sat for 90 days might be priced 20 percent too high rather than distressed. Use these as sorting aids, not hard cutoffs.

The second is treating equity as destiny. High equity means an owner can sell. It does not mean they will. You still need to call, mail, or door-knock. The shortlist is a starting point for outreach, not a list of motivated sellers.

How to apply this in PropGPT

Three prompts, copy-paste ready. Replace the bracketed fields with your own buy-box.

Prompt 1 — pull the candidate grid:

Show me all active single-family listings in ZIPs [30314, 30315, 30318] between $[180000] and $[320000], 3+ bed, 2+ bath, built before [1990]. Include list price, beds, baths, sqft, year built, days on market, and any price cuts. Return as a grid with photos.

Prompt 2 — enrich with owner and equity data:

For the grid above, add the following columns: owner of record, owner-occupied flag, last sale date, last sale price, estimated current value, and estimated equity percent. Flag any owner with tenure over 10 years and equity above 40 percent.

Prompt 3 — rank by composite score:

Rank these properties by a composite score weighted as follows: equity 40%, days on market 20%, price cuts 15%, owner tenure 15%, price-to-ZIP-median 10%. Show the top 20 with score breakdowns and export-ready columns.

Run the three prompts in order. If the first grid is too wide, tighten the price band or add a max days-on-market. If the shortlist looks thin, relax the equity threshold to 30 percent before you change anything else.

That is the entire workflow. The skill is not in the prompts. It is in writing a buy-box specific enough that the prompts have something real to act on.