Autonomous Real Estate Workflows: Zillow Data & AI Agents
Combine a property data feed, an LLM writing agent, and a synthetic video platform inside a single Make.com scenario — and a new listing becomes a finished, video-narrated marketing campaign in under 60 seconds with no human involvement. The data-sourcing step is the part most guides get vague about — here’s the actual 2026 landscape, not just “use RapidAPI.”
Quick answer: what this pipeline does
- Data extraction — Make.com pulls raw JSON (price, beds, baths, square footage) from an MLS feed or a licensed third-party property data provider when a new listing goes live. There’s no self-serve public Zillow API anymore — more on the real options below.
- LLM copywriting — The JSON is passed to an OpenAI prompt constrained to write a 60-second property script using only the provided fields, with fair housing compliant language built into the instructions.
- Synthetic video output — The script goes to HeyGen’s API, which renders an AI avatar reading it aloud and returns a downloadable MP4.
- Distribution — Make.com pushes the text and video to GoHighLevel, Google Drive, or Slack for final human review before anything publishes.
The manual listing problem
Most brokerages are paying people to copy and paste. A new property clears MLS approval, and then a junior agent or virtual assistant manually transfers the data fields into a listing description, records or edits a video walkthrough, and schedules social posts one by one. That sequence takes 45–90 minutes per listing and scales with headcount, not with volume.
The no-code AI agent approach replaces each of those steps with an API call. The only meaningful cost is the platform subscriptions — not hourly labour.
The API-to-video pipeline architecture
Make.com acts as the orchestration layer that ties all three phases together. Each phase runs sequentially within a single scenario — no separate workflows or manual handoffs required.
Phase 1 — Data
Property data trigger
A Make.com webhook fires when a new property status appears. The HTTP module retrieves the raw JSON payload: price, square footage, bedroom count, bathrooms, and neighbourhood name — sourced from your MLS feed or a licensed third-party provider (see below).
Phase 2 — AI
LLM copywriting agent
Make.com passes the JSON to the OpenAI module. A strict system prompt forces the model to write a 60-second property script using only the fields provided — hallucinated features are blocked by the prompt constraints, and fair housing language rules are built in.
Phase 3 — Video
Synthetic video generation
The finished script text is mapped into the input_text field of the HeyGen API. HeyGen renders a photorealistic AI avatar reading the script and fires a webhook back to Make.com with the completed MP4 URL.
Phase 4 — Distribution
CRM and social delivery
Make.com pushes the text and MP4 file to GoHighLevel, Google Drive, or a Slack channel. The listing is ready for a human to review and publish — treat this as a required checkpoint, not an optional one (see the compliance section below).
The real 2026 Zillow data-access landscape
A lot of guides on this topic — including an earlier version of this one — wave at “use RapidAPI” and move on. That’s directionally right but skips the part that actually determines your cost and your data quality. Zillow’s original public API was shut down in 2021 and never replaced with a self-serve equivalent. What exists now splits into genuinely different categories, and which one you need depends on what data you’re after.
| Path | Who it’s for | Cost | Zestimate included? |
|---|---|---|---|
| Bridge Interactive | MLS-affiliated brokerages, approved tech partners | ~$500+/month, weeks of application review | Generally no — it’s Zillow’s proprietary valuation, not standardized MLS data |
| Third-party wrapper API (RapidAPI listings, similar marketplaces) | Indie developers, agencies, SaaS builders without MLS credentials | Free tier available on some providers; production use typically $0.002–$0.01 per call or $99–$299/month | Yes, on most providers — extracted as a sub-resource |
| Public-records aggregators (ATTOM, Estated, etc.) | Teams that need beds/baths/sqft/ownership but not Zillow-specific fields | $95–$299/month starting, or pay-per-call | No — sourced from county assessor records, not Zillow |
A caveat most comparison posts skip
Zillow’s Terms of Use prohibit automated access — scraping or bots — without express written permission. Bridge Interactive is the sanctioned path; third-party wrapper APIs are operating in a legal grey area, extracting data from Zillow’s public surface rather than a licensed feed. That doesn’t mean they’re unusable — plenty of production tools run on them — but it’s the trade-off you’re accepting for the lower cost and faster setup, and it’s worth knowing rather than discovering later.
If you’re building this for your own listings and already have MLS access, feeding the pipeline directly from your MLS’s RESO Web API (the same standard Bridge is built on) sidesteps the whole question — you’re not sourcing Zillow’s data at all, just your own listings.
For most solo agents and small teams without MLS-vendor status, a third-party wrapper is the realistic starting point: instant signup, a free or low-cost tier to prototype the pipeline, and the Zestimate field the original version of this guide’s script generation assumed was available. If your use case is strictly property characteristics — no Zestimate needed — ATTOM or Estated are cleaner, more clearly licensed sources for that narrower need.
→ Start wiring the Make.com scenarioWhat each tool does in the stack
Data layer
MLS feed / third-party property API
Provides live property data as structured JSON. See the data-access table above for which path fits your situation — MLS-affiliated, third-party wrapper, or public-records aggregator.
AI layer
OpenAI via Make.com
Converts raw listing data into a structured property script. The system prompt is the critical control layer — without constraints, the model invents features and can drift into non-compliant language.
Video layer
HeyGen API
Accepts a plain-text script and returns an MP4 with an AI avatar. Connects via Make.com’s HTTP module using the HeyGen v2 API.
Distribution layer
GoHighLevel CRM
Receives the finished text and video assets for human review, then powers automated follow-up sequences triggered by new listing contacts from social distribution.
The LLM prompt template
The quality of the entire pipeline depends on this prompt. A loose instruction produces hallucinated school districts, fictional amenities, and scripts that do not match the actual property — or worse, language that creates fair housing exposure. Every field the model is allowed to use must be stated explicitly, and the constraints must be absolute, not suggestions.
Why the word limit matters
Most synthetic video platforms render at approximately 130–140 words per minute. A 150-word cap keeps the output inside a 60-second slot — the maximum for most short-form social placements. Exceeding this causes the video to be cut off or the platform to reject the render job.
Fair housing compliance for AI-generated copy
This is the section most automation guides skip entirely
The Fair Housing Act prohibits advertising language that indicates a preference, limitation, or discrimination based on protected characteristics. That obligation doesn’t change because an LLM wrote the copy instead of a person — liability generally sits with the listing agent and brokerage regardless of who or what drafted the wording, and “the AI wrote it” is not a defense.
The prompt constraint above (describe the property, not the occupant) handles the most common failure mode: models trained on general real estate copy sometimes default to phrases like “perfect for young families” or “walking distance to [specific religious institution]” because that language is common in the training data, not because anyone instructed it to. Explicitly forbidding occupant-type language in the system prompt materially reduces this risk, but it doesn’t eliminate the need for a human set of eyes.
Treat the human review step in Phase 4 as mandatory, not optional — even a well-constrained prompt can produce an edge case, and the review checkpoint is what catches it before it reaches a listing platform or ad account. If you’re operating in a market with additional state or local fair housing rules beyond the federal baseline, build those specific restrictions into the prompt’s constraints section too rather than relying on the general instruction alone.
Triggering synthetic video generation
Once Make.com has the script text from the OpenAI module, it posts to the HeyGen API using a standard HTTP POST module. The request body maps the LLM output into the input_text field alongside your chosen avatar ID and voice ID from your HeyGen account.
HeyGen processes the render asynchronously — the job may take 30–120 seconds depending on video length. Make.com handles this by polling the job status endpoint, or by setting HeyGen to fire a callback webhook when the render completes. The callback carries the MP4 download URL, which Make.com then routes to your CRM or cloud storage.
Build the automation foundation first
Before connecting video generation APIs, make sure your Make.com webhook routing is solid. The guide below covers the foundational scenario structure this pipeline builds on.
Master the Make.com automation foundation →Frequently asked questions
What are autonomous real estate workflows?
Automated pipelines that connect property data to an AI writing agent and a synthetic video platform. A new listing enters the pipeline and a written script plus a finished MP4 exit — with no manual steps in between, though a human review checkpoint before publication is strongly recommended.
Can Make.com connect to the Zillow API?
There’s no self-serve public Zillow API in 2026 — it was discontinued in 2021. The two real paths are Bridge Interactive (Zillow’s official RESO Web API program, MLS-affiliation required, ~$500+/month) or a third-party property data API that Make.com can call through its standard HTTP module. Third-party wrappers typically expose the Zestimate; Bridge generally doesn’t, since that’s Zillow’s proprietary valuation rather than standardized MLS data.
How do you automate real estate listing videos?
Make.com sends the LLM-generated property script to a synthetic video API such as HeyGen or Synthesia via an HTTP POST. The platform renders an AI avatar reading the script and returns a downloadable MP4. The whole render typically completes in under two minutes.
What LLM prompt structure works best for listing scripts?
Define a role (real estate copywriter operating under fair housing rules), pass the raw JSON as the only permitted data source, enforce a hard constraint against inventing features not present in the input, explicitly forbid describing the type of occupant or referencing protected characteristics, set a 150-word maximum, and specify plain spoken text output only — no markdown or stage directions.
Is AI-generated real estate marketing copy legally risky?
It carries the same fair housing obligations as copy a human writes. Liability generally falls on the listing agent and brokerage regardless of whether a human or an AI drafted the wording. A well-constrained prompt reduces the risk significantly, but a human should review AI-generated copy before publication rather than treating the pipeline as fully unsupervised — this isn’t legal advice, and if you’re uncertain about your specific market’s requirements, that’s a conversation for your broker or counsel.
Does this pipeline work without a developer?
Yes. Make.com, the OpenAI module, and HeyGen are all no-code or low-code tools. The HTTP module handles the API calls. The most technical step is reading the JSON payload from your property data source and mapping field names correctly — no custom code is required.
Affiliate disclosure: CreatorOpsMatrix is an independent technical research publication. Links to Make.com, HeyGen, and GoHighLevel on this page are affiliate links. If you sign up through them, we earn a commission at no extra cost to you. This funds continued publication of free automation guides.
Operator responsibility: The code samples, JSON structures, and routing configurations described here are provided for educational purposes only and are not legal advice. Fair housing requirements vary by jurisdiction — confirm your specific obligations with your broker or legal counsel. You are responsible for testing, reviewing, and maintaining any automation you deploy in a production environment.