The Future of Real Estate Search Is Spatial, Not Filtered
Filters and a photo carousel cannot answer whether the second bedroom fits a desk. The future of real estate search runs through spatial data, walkable scenes and AI agents that can query a room instead of guessing from a caption.
A buyer opens a listing at eleven at night. The filters are already set: three bedrooms, two baths, under $650,000, inside one school zone. The results load as a grid, then a photo carousel opens under each one: living room, kitchen, primary bedroom, backyard, repeat. None of it answers what the buyer actually wants to know. Whether the light in the room they'd use as an office is any good at four in the afternoon. Whether the second bedroom is deep enough for a desk and a bed, not one or the other. How far the kitchen really is from the front door, in the fifteen seconds after carrying in groceries. Closing that gap, more than any single new feature, is what the future of real estate search has to do.
Why the future of real estate search still looks like a spreadsheet
Search on almost every portal today runs on two separate things bolted together. A structured record, price, square footage, bed and bath count, ZIP code, pulled from the MLS feed and matched against whatever filters the buyer sets. And a stack of unstructured media, photos, sometimes a video, occasionally a Matterport-style dollhouse view, that a human has to look at and interpret for themselves. The database can answer "under $650,000." It cannot answer "afternoon sun in the primary bedroom" or "quiet street," because those were never columns. They were always going to be answered by a person scrolling photos and guessing, or by driving over and finding out.
That split shows up in the schema itself. The RealEstateListing type search engines actually use carries price, address, floor size and a text description. It has no field for room orientation, sightlines, or the distance between two points in the house, because none of that has ever been captured in a form a database could hold. A photo is evidence a room exists. It is not a measurement of it.
Buyers have adapted to that gap rather than had it closed for them. They read the photos for tone, message the agent for the actual answer, or go and look. Navigating a property online has meant navigating a folder of images that someone else chose the angle for, then filling in the missing thirty percent by asking a person. That is the status quo the next few years of proptech are working against, not a hypothetical one.
What actually changes
Three mechanisms are doing the work, and none of them is a slogan. Each is something specific that already exists in an early form and is getting more capable on a visible timeline, and together they are what the future of real estate search is actually built from, not a single new app or a redesigned homepage.
Spatial data becomes a field you can query, not a PDF you open
A floor plan today is usually a static image or a PDF, produced once, attached to a listing, and never touched again. It cannot be queried, because it isn't data in the database sense, it's a picture of data. The alternative is capturing the space as real geometry: a point cloud or a reconstructed scene where every surface has a position, so "does the primary bedroom fit a king bed and a dresser" becomes a question with a real answer instead of a guess from a photo taken at a flattering angle.
The plumbing for this is being rebuilt right now, if slowly. RESO's Data Dictionary 2.0, the shared schema most MLS feeds in the US are built on, passed in April 2024, and every affiliated MLS was required to certify against it by April 2025 (WAV Group Consulting, 2024). That deadline says nothing about geometry fields specifically. What it says is that the industry's data layer is actively being rewritten on a real, multi-year cadence, which is exactly the condition under which a new class of field, spatial rather than descriptive, gets added. Capture pipelines that take LiDAR scans in E57, LAS, LAZ or PLY and output a reconstructed scene are already producing that data. What's missing is a place for it to live upstream of the listing page, in the feed itself rather than as an embed a buyer has to know to click.
Until that lands at the schema level, a listing team does not have to wait. Publishing the walkable scene on the listing page, built from a 360 walkthrough shot on the same visit as the photos, gets the spatial data in front of the buyer today, even before it's a filterable column anywhere.
The photo carousel is replaced by a scene you walk
Zillow's own data from 2022 found that listings carrying a 3D Home tour were viewed 43% more than listings without one. That was measured against dollhouse-style tours built from stitched photos, a real improvement over six static images but still, underneath, a photograph pretending to be a space: accurate from the exact spots the camera stood, and a blur everywhere else.
A Gaussian splatting reconstruction is a different kind of object. Instead of a point cloud, which is accurate but reads as a sparse diagram to anyone who isn't a surveyor, it stores the same captured data as continuous photoreal surfaces, so a viewer walking through sees a room rather than a scatter of dots standing in for one. That is the actual mechanism behind "beyond photo listings": not a better camera angle, but a format the buyer can walk through from any position, at midnight, from another city, without an agent narrating it.
This is also where how buyers will search for homes in 3d starts to diverge from how they browse listings today. A carousel is browsed in the order the photographer chose. A scene is walked in the order the buyer chooses, which means the questions it answers are the buyer's questions, not the listing photographer's idea of the flattering ones.
AI agents read geometry, not adjectives
In September 2024, Zillow rebuilt its search to accept plain language instead of only filters, letting a buyer type "apartments near Denver Union Station" or "homes thirty minutes from Millennium Park" and get a ranked result (Zillow, 2024). In October 2025, Zillow went further and became the first real estate app available inside ChatGPT, returning listings complete with photos, maps and pricing directly in a conversation (Zillow, 2025).
Both of those are real, shipped, and both still answer from the same fields a human search bar always used: price, address, a text description, a caption written by an agent. Ask either one whether a specific corner of a specific room gets afternoon light, and there is nothing behind the answer but the adjectives in the listing copy, "bright," "sun-drenched," "east-facing," none of which is measured. Spatial search for property listings is the step after natural-language search: an agent that can actually query a reconstructed scene, not paraphrase a caption about one, and answer with a real distance or a real sightline instead of a word someone chose to sound appealing. That requires the geometry from the first mechanism to exist as data an API can reach, which is why the three mechanisms are one shift rather than three separate trends.
What brokerages and portals will actually push back on
"We already have 3D tours, isn't this solved?" Most of what's live today is a nicer viewer wrapped around the same unstructured media, not a queryable record. The tour is easier to look at. It still can't be searched, filtered, or read by an agent the way a price field can.
"Capturing every listing this way is too slow." Cloud processing for a reconstructed scene runs 10 to 120 minutes with no local GPU and no software install, from ordinary walkthrough video an agent already shoots on a phone or a drone for the photos and the tour. It is added time on a walkthrough that already happens, not a second site visit.
"MLS fields move slower than any of this." True, and worth planning around rather than waiting on. A listing page can carry the walkable scene and its underlying geometry now, independent of whether RESO adds a spatial field this year or in three. When the schema catches up, the capture work is already done.
"Who on the team has to learn something new?" Nobody has to learn a new capture method. The video an agent already films for a tour is the same input a reconstruction pipeline takes. The change is in what happens to that footage after it's recorded, not in how it's shot.
| Filters plus carousel | Spatial search | |
|---|---|---|
| What's queryable | Price, beds, baths, ZIP | The above, plus room-level geometry once captured |
| What "3D" means | A dollhouse of stitched photos | A continuous, walkable reconstructed scene |
| Who can answer "does it fit" | The buyer, guessing from photos | The scene itself, measured |
| Comparing two units or a renovation over time | Two browser tabs and memory | Two captures aligned and compared directly, as with Multi Splat |
| Opens on | Desktop, mostly | Phone, tablet, desktop or headset, no app |
A scene reconstructed from a walkthrough only holds what was in the room that day. Capture before furniture is swapped out for staging, or immediately after, so the geometry a buyer or an agent later queries still matches what's actually there.
What a listing team can do this quarter
None of the three mechanisms needs the MLS to move first. A brokerage or portal product team can start capturing the walkthrough video it already shoots as 3D input today, publish the resulting scene alongside the existing photos, and keep the underlying geometry rather than throwing it away once the viewer is live. That geometry is the asset that gets more useful later: as a field a portal's own search can filter on, as something an AI agent can query instead of paraphrase, and as a record that survives longer than a caption someone wrote to sound appealing. The filters and the carousel are not going away this year. What changes first is what sits behind them, and that part is buildable now, one walkthrough at a time.
FAQ
Does 3D content help a listing get found in search, not just viewed?
Not directly through today's MLS schema, since floor geometry isn't a field search engines index yet. It helps indirectly: a walkable scene keeps a visitor on the page longer and gives an AI search agent more to work with than a caption, both of which existing ranking signals reward.
Do buyers need to install anything to view a spatial listing?
No. A reconstructed scene built for the web opens in a phone, tablet or desktop browser, and on a VR headset where one's available, the same way an embedded video does today. There's nothing to download and no plugin to approve, so it works the same way for a buyer scrolling on a commute as for one at a desktop.
What happens to the photo carousel once a scene is live?
Nothing has to be removed. A walkable scene sits alongside the existing photos and video rather than replacing them, since some buyers will still want the curated highlight shots first before they decide whether a room is even worth exploring further, and a carousel is faster for that first pass than a scene is.
All trademarks belong to their respective owners. No affiliation, partnership or endorsement is implied.