Skip to main content

Zillow & Realtor.com Scraping: A Pipeline That Lasts (2026)

· 9 min read
Yassine El Haddad
Software Developer & Automation Specialist

I build production AI agents, web scrapers, and automation pipelines. Most of what I publish here comes from the actual problems they run into: proxies that get banned, anti-bot stacks that fingerprint your client, RAG that drifts when the underlying data moves. Stack: Python, TypeScript, Go, FastAPI, LangChain, Crawlee, Playwright, deployed on AWS, GCP, and Cloudflare.

Scraping one Zillow page is easy to try and fails within seconds. Keeping a table of listings fresh every night, across both Zillow and Realtor.com, is a pipeline problem: bot protection, a search step and a detail step, deduplication, and storage that keeps history.

This guide builds that pipeline on Apify Actors that are live today, with real output and costs from test runs on 23 October 2026.

Quick Answer

Do not point a plain HTTP client at Zillow. It sits behind PerimeterX (HUMAN) and blocks scripted traffic on the first request. Use a two-step Actor pipeline instead: ZIP search → detail, chained by dataset ID, plus a Realtor.com Actor for a second source. Key every row on zpid, store dated snapshots, and run a small nightly delta. Five Zillow listings cost $0.025 through both steps on a paid plan.

The blocker: bot protection, verified​

We sent one request to a Zillow search page with a Python-style user agent. The response was HTTP 403 with an x-px-blocked: 1 header, and the page body loaded a px-captcha challenge. That is PerimeterX, now part of HUMAN Security. Realtor.com returned HTTP 403 to the same request, from CloudFront, without naming a bot-protection vendor.

So the question is not whether you get blocked but who handles it. You have two options:

ApproachWhat you maintainWhen it fits
Build your own scraperResidential proxies, browser fingerprints, session rotation, CAPTCHA handling, parser fixes every time the page changesYou need a field no Actor returns, or volumes where per-result pricing stops making sense
Use maintained ActorsInputs, scheduling, storageMost teams. The Actors used here take no proxy settings at all: blocking is the Actor's problem, not yours

If you do build your own, residential or ISP proxies are the minimum. Our datacenter vs residential vs ISP proxies comparison covers which to pick. The rest of this guide uses maintained Actors.

The three Actors​

StepActorInput that mattersPrice per result (Free / Starter plan)
Zillow searchZillow ZIP Code SearchzipCodes, sale/rent/sold flags, daysOnZillow, resultsLimit$0.0024 / $0.002
Zillow detailZillow Detail ScrapersearchResultsDatasetId, or startUrls / addresses$0.0036 / $0.003
Realtor.comRealtor Scrapersearch or startUrls, mode, maxItems, includeDetails$0.01 per search query, $0.001 per property detail

All three were live and maintained when we checked them on 23 October 2026. Prices are tiered and change, so check each Actor's pricing tab before you budget.

Why ZIP search and not a search URL. The general Zillow Search Scraper wants a full Zillow search URL. Our first test passed a plain area URL and the run failed straight away with "No searchQueryState in URL": the scraper needs the encoded map and filter state that Zillow adds to the URL once you search in a browser. ZIP codes avoid that. They are also the natural unit for partitioning a metro.

Step 1: search by ZIP code​

{
"zipCodes": ["78704"],
"forSaleByAgent": true,
"forSaleByOwner": false,
"forRent": false,
"sold": false,
"daysOnZillow": "7",
"resultsLimit": 5
}

daysOnZillow accepts "1", "7", "14", "30", "90" and longer windows. A nightly run with "1" returns only listings that appeared since the last run, which keeps the delta small.

The search output already carries zpid, listingPrice, livingArea, bedrooms, daysOnZillow, homeType and, for some listings, zestimate, rentZestimate and taxAssessedValue.

Step 2: chain the detail scraper​

Pass the search run's dataset ID straight into the detail Actor. No URL lists to copy:

{
"searchResultsDatasetId": "<dataset ID from step 1>",
"propertyStatus": "FOR_SALE"
}

On the Apify platform, an Actor webhook can start step 2 automatically when step 1 succeeds.

What the test run returned​

Real output from a test run on 23 October 2026, ZIP 78704, five for-sale listings. Addresses, agents and listing IDs are removed:

ListingTypeList pricezestimaterentZestimatedaysOnZillowtaxAssessedValueHOA
ASingle family$300,000nonenone6nonenone
BSingle family$699,000$677,600$3,0217$782,458none
CSingle family$1,075,000$1,077,600$5,0237$775,883none
DSingle family$1,295,000$1,258,900$6,4492none$600 a year
ESingle family$1,990,000nonenone7$1,872,660none

The search step cost $0.010 and the detail step $0.015, so $0.025 for five fully detailed listings on a Starter plan.

What five rows already teach about pipeline design:

  • Two of five listings had no Zestimate and no rent Zestimate. Treat these as nullable columns. Do not fill them with an average: a model that cannot tell "no estimate" from "estimate" will lie quietly.
  • The field names are not the ones older guides use. Price lives in listingPrice.amount, history in listingPriceHistory and listingTaxHistory, and annual tax in taxAnnualAmount. A loader built on price and priceHistory gets nulls without any error.
  • Listing A is an outlier. It is listed at about $120 per square foot, against $639 to $893 for the others in the same ZIP, and it has 22 price-history events. A pipeline should flag a row like that for a human, not feed it into an average.
  • The detail output contains people's names and phone numbers (agent, broker, attributionInfo). Drop those columns at load time unless you have a lawful reason to keep them.

Step 3: add Realtor.com as a second source​

{
"search": "Austin, TX 78704",
"mode": "BUY",
"maxItems": 3,
"includeDetails": true
}

Our test returned three listings. It charged one search query and three property details, $0.013 in total. Realtor.com adds fields Zillow does not return here, such as buildingPermitsHistory and daysOnMls, and both sources carry the MLS listing number.

HOA fees exist on both sites, in different shapes. That matters because a missed HOA fee skews any cost model:

SourceWhere the HOA fee isExample from the test runs
ZillowresoFacts.hoaFee (text) and monthlyHoaFee (number)"$600 annually" and 50
Realtor.comFree text in details, category "Homeowners Association""association fee: 425", "association fee frequency: monthly"

Parse the Realtor.com text into a number and a frequency before you join. Normalise both sides to a monthly figure.

Step 4: dedupe and join​

  • Within Zillow, key on zpid. It is stable across runs.
  • Within Realtor.com, key on its property id.
  • Across the two, join on the MLS listing number first. Zillow returns it in mls.id. Realtor.com returns it in history[].sourceListingId, together with the MLS name in sourceName. Only match numbers that come from the same MLS. Fall back to coordinates plus a normalised street address when there is no MLS number, for example for listings sold by owner.

Our two test samples did not include the same property, so this join was not tested end to end. Validate it on a ZIP where both sources return the same listing before you trust it.

Step 5: schedule and store​

  1. Nightly delta. Schedule the ZIP search with daysOnZillow: "1" (see Apify tasks and schedules), then the detail step through a webhook.
  2. Weekly refresh. New listings are not the only change. Re-run the detail step on your active zpid list once a week to catch price cuts and status changes. listingPriceHistory records every event with a date.
  3. Store snapshots, not overwrites. Keep an append-only table keyed on (zpid, scrape date) and a current-state table you upsert. Price-change alerts then become a comparison between the last two snapshots.
  4. Partition by ZIP. One run per ZIP code keeps each run small, makes failures cheap to retry, and lets you see which areas return nothing.

For a server-side cron alternative to Apify schedules, see cron jobs for web scraping on Linux.

To turn these listings into a per-property short-term vs long-term rental yield comparison, see real estate data scraping with Claude.

What it costs​

At Starter-plan prices, 1,000 Zillow listings through search and detail cost about $5.00 ($2.00 + $3.00). On the Free plan it is about $6.00. A nightly delta usually returns far fewer rows than a full pull, so the weekly refresh of active listings is normally the larger line.

Realtor.com pricing works differently: $0.01 per search query plus $0.001 per property detail. Many small queries cost more than a few large ones.

Before you rely on it​

  • Terms of service. Zillow's and Realtor.com's terms restrict automated collection. Read them, and our guide to scraping legality, before you build on this data. Keep the data internal and avoid personal fields.
  • Zestimates are estimates. Zillow's own figure, not an appraisal, and often missing, as our test showed.
  • Fields change. Pin your loader to the fields you use and alert when a column suddenly comes back empty, which is how the old price field fails.

For what to do with the data once you have it, see the real estate web scraping guide.

Frequently Asked Questions

Not for general use. Zillow retired its old public API, and data access today goes through partner programmes for approved companies. For most teams, scraping public listing pages with a maintained Actor is the practical route, within the sites' terms.

Not with the Actors in this guide: none of them takes proxy settings, because blocking is handled inside the Actor. You only need residential or ISP proxies if you build your own scraper, and Zillow's PerimeterX protection makes that a significant maintenance job.

The ZIP search Actor accepts priceMin and priceMax. It has no home-type filter, so filter on the homeType field after the search step, before you pay for detail calls on listings you will discard.

The Zillow Search Scraper needs a full search URL copied from your browser after you run a search on Zillow, which includes the encoded map and filter state. A plain area URL does not work. ZIP code search avoids the problem entirely.

Advertise on this page