Proxies for Real Estate Data Scraping (2026): Zillow, Redfin and MLS Portals Without Bans

Real estate is the most scrapable data category on the public web and the most aggressively defended. Every listing portal — Zillow, Redfin, Realtor.com, Rightmove, Idealista, Immobilienscout24 — publishes structured, high-value data (price, beds, baths, square footage, days on market, price history) and then spends real money making sure you cannot collect it at scale from one IP address.

This guide is the practical, tested version of "which proxies do I need for real estate data scraping". It covers the defences you will actually hit, the proxy type that matches each portal tier, realistic bandwidth math for a 100,000-listing pipeline, and the provider shortlist we would use ourselves. If you only want the ranked shortlist, jump to our best proxies for web scraping list; if you want to understand why those picks work here, keep reading.

Why real estate portals are harder than they look

Most people underestimate the problem because their first 500 requests succeed. A portal will happily serve you a few hundred pages from a datacenter IP before anything happens. The blocking is statistical, not instant.

Three defence layers do the work:

1. Edge reputation (Cloudflare, Akamai, Imperva). The first filter is the IP itself: ASN, whether it belongs to a hosting provider, historical abuse score, and how many distinct sessions have arrived from it recently. Datacenter ranges from major clouds are effectively pre-flagged on the biggest US portals.

2. Behavioural rate models. Portals track requests per IP per minute, but also the *shape* of the traffic: sequential listing IDs, zero image loads, no CSS fetches, perfectly even inter-request timing, and a request mix that no human produces. A residential IP that behaves like a crawler still gets blocked — just later.

3. Client fingerprinting. TLS/JA4 fingerprints, HTTP/2 frame ordering, header order, and (on the JS-rendered tiers) canvas, WebGL, and font enumeration. This is where a mismatched setup fails: a "US residential" IP paired with a headless Chrome fingerprint and an Accept-Language: en-GB header is a trivially detectable contradiction.

The practical consequence: proxy quality is necessary but not sufficient. You need the right IP type *and* a request profile that matches it. Our antidetect browser reviews cover the fingerprint half of that equation.

Proxy type by portal tier

Not every real estate target needs the same class of IP, and paying residential rates for a public records site is how scraping budgets get destroyed. We group targets into four tiers.

Two rules follow from this table.

Never start at tier 3. Fetch the sitemap and JSON-LD from a tier 1 IP first. Most portals expose Product/Residence structured data in the page head, and a surprising share of "scrape the listing" work is really "parse the JSON-LD you already downloaded cheaply".

Never use tier 4 for volume. Mobile proxies exist to solve trust problems, not throughput problems. A single 4G port doing 100 GB is both slow and financially absurd.

If you are not sure which tier a specific target belongs to, our proxy type selector walks the decision in about thirty seconds, and proxy types explained covers the underlying differences.

Bandwidth math: what a 100,000-listing crawl actually costs

This is the part every vendor page skips, and it is the number that decides your architecture.

Measured page weights (median, HTML + required XHR, images excluded, gzip on):

• Listing detail page, HTML only: 180–420 KB
• Listing detail page, JS-rendered with XHR payloads: 900 KB – 2.4 MB
• Search results page (40 listings): 240–700 KB
• Photo assets per listing (20 images at 1440px): 4–9 MB

Now run the pipeline three ways for 100,000 listings:

A. Naive headless browser, full page load, residential at $3/GB. 100,000 × 1.6 MB ≈ 160 GB ≈ $480 — and that is before retries. At a 78% first-pass success rate, real spend lands near $615.

B. HTML-only fetch with a proper HTTP client, residential at $3/GB. 100,000 × 300 KB ≈ 30 GB ≈ $90. Same data, because the listing facts are in the server-rendered markup and JSON-LD.

C. Tiered: sitemap + JSON-LD discovery on datacenter, detail pages on residential. Discovery: ~8 GB on datacenter (~$6). Detail: 30 GB residential (~$90). Total ~$96, with the discovery layer refreshable daily at negligible cost.

The gap between A and C is roughly 6×. Blocking images, fonts, and analytics beacons is the single highest-ROI change most real estate scrapers can make. If you must render JavaScript, render it once per page template to extract the API shape, then call that API directly.

Session strategy: rotating vs sticky

Real estate scraping has an unusual property — most of it is *stateless*. You are not logged in, there is no cart, and each listing page is independent. That argues for aggressive rotation.

But two workflows need sticky sessions:

• Paginated search results. Portals frequently bind result pagination to a session cookie. Rotating mid-pagination gives you page 1 four times.
• Geo-scoped searches. A search for "Austin, TX" set through a location picker persists in a cookie; a new IP in a different metro resets it.

Our working default:

• Detail pages: rotate per request, no cookie jar, country-level targeting only.
• Search pagination: sticky session for the length of one pagination run (typically 10–25 requests), then discard.
• Geo-sensitive queries: city-level targeting where the provider supports it, sticky for the query lifetime.

Set sticky durations to the *shortest* window that completes the job. Long sticky sessions accumulate behavioural history on a single IP, which is exactly what tier 3 rate models are built to catch. The mechanics of getting this right are covered in depth in how to rotate proxies without getting blocked.

Geo-targeting: city level matters more here than anywhere else

Real estate is the one vertical where sub-country targeting is genuinely load-bearing rather than a marketing bullet.

Portals localise heavily. Rightmove serves different module layouts to UK vs non-UK IPs. US portals inject regional MLS disclaimers, and some suppress exact addresses or price history for out-of-region requests. Idealista and Immobilienscout24 both adjust currency, language, and consent flows by IP geography.

Practical rules:

1. Match IP country to target country, always. A US IP scraping Rightmove gets a materially different DOM. 2. Match IP region to search region where you can. For US metro-level work, state-level targeting is usually enough; city-level helps on portals that surface "nearby" recommendations you want to capture. 3. Match the rest of your identity to the IP. Accept-Language, timezone, and locale must agree with the proxy's geography. This mismatch is the most common self-inflicted block we see. 4. Verify, don't trust. Providers advertise city targeting; the real question is how many *distinct* IPs exist in that city. Test with a sample of 200 requests and count uniques before you commit.

Provider shortlist for real estate pipelines

These are drawn from our own benchmark cycle. Each links to the full review with test methodology, pricing, and measured success rates.

[IPRoyal](/reviews/iproyal) — best value for high-volume detail crawling. Pay-as-you-go residential with no forced monthly commitment, which suits the burst pattern of real estate work (heavy initial backfill, light daily deltas). City-level targeting and unlimited concurrent sessions. The right pick when your bottleneck is per-GB cost rather than the hardest possible target.

[Oxylabs](/reviews/oxylabs) — best for tier 3 portals at enterprise scale. The highest success rates we measure on heavily defended US portals, plus a Web Unblocker-style product if you would rather buy the unblocking than build it. Priced accordingly; justified when a failed crawl costs more than the proxy bill.

[NodeMaven](/reviews/nodemaven) — best success rate per GB. Filters low-quality IPs before they reach you, which shows up directly as fewer retries. On a bandwidth-billed product, retries *are* the cost, so a higher per-GB price with a 15-point better success rate is frequently cheaper in practice.

[IPFly](/reviews/ipfly) — best mixed residential + ISP + datacenter stack. Useful precisely because real estate scraping is tiered: you can run discovery on their datacenter pool, detail pages on residential, and persistent regional identities on ISP IPs, all under one dashboard and one balance.

[Smartproxy / Decodo](/reviews/smartproxy) — best dashboard for small teams. Clean sub-user management, straightforward city targeting, and predictable pricing. A good first residential provider if nobody on the team has run a proxy pool before.

[Rayobyte](/reviews/rayobyte) — best datacenter tier for discovery. Cheap, high-throughput, and the correct tool for sitemaps, robots-permitted crawl paths, and public records portals where residential IPs are wasted money.

[Proxy001](/reviews/proxy001) — best budget entry for regional portals. Competitive per-GB rates on tier 2 targets where you do not need the top-end pool quality.

Full pricing and pool comparisons across all of them live on our proxy comparison table, and current discounts are tracked on the deals page.

Architecture that survives contact with production

A real estate pipeline that works in month six looks like this:

Discovery layer (datacenter, cheap, daily). Pull sitemaps, search-result pagination, and any public feed. Output: a queue of listing URLs with last-seen timestamps. This layer answers "what changed?" so the expensive layer only touches what it must.

Fetch layer (residential, rotating, HTML-only). Requests with images, fonts, media, and third-party beacons blocked. Retry policy: exponential backoff with a *new identity* on every retry, not the same IP. Cap at three attempts, then park the URL for the next cycle rather than hammering it.

Parse layer (no network). JSON-LD first, then CSS selectors as fallback, then a template-drift alarm. Portals change markup roughly quarterly; a silent parse failure that writes nulls for two weeks is worse than a hard block.

Render layer (rare, mobile or premium residential). Reserve headless browsers for the small percentage of pages that genuinely require JS. Our scraping tools directory covers the managed APIs worth considering before you self-host this layer.

Observability. Alarm on four metrics: success rate by target, bytes per successful listing, unique IPs used per hour, and parse-field fill rate. The last one catches template drift; the first three catch proxy problems before your bill does.

Delta crawling: the technique that halves everything

Real estate data ages in a specific pattern. Roughly 3–6% of active listings change state (price, status, photos) on any given day; the rest are static. Full re-crawls are almost always waste.

A delta strategy:

1. Crawl the search-results tier daily on cheap IPs. Results pages expose price and status for 40 listings at a time — that is one request per 40 listings instead of 40. 2. Diff against your store. Enqueue only changed listings for detail fetches. 3. Re-fetch unchanged listings on a slow rotation — every 14–30 days — to catch fields not exposed in search results.

Teams that switch from nightly full crawls to delta crawls typically cut proxy bandwidth by 80–92% with no loss in data freshness. It is a better investment than any provider switch.

Legal and ethical boundaries

This matters in real estate more than most verticals, because MLS data carries contractual restrictions that public web data does not.

• Public listing pages are not the MLS. Scraping a portal's public pages and redistributing MLS-sourced content are different acts with different exposure. IDX/MLS feeds are licensed; the licence terms bind you.
• Respect robots.txt and rate limits as a default posture, and document where and why you deviate.
• Personal data. Agent names, phone numbers, and photos are personal data under GDPR and several US state laws. Have a lawful basis, a retention limit, and a deletion path before you collect them.
• No credential-gated content without authorisation. Logging in with a scraped or shared account crosses from "web scraping" into territory covered by computer misuse statutes in most jurisdictions.
• Copyright in listing photos and descriptions usually sits with the broker or photographer. Extracting facts is safer than republishing prose and imagery.

Our compliance and scraping policy sets out the framework we apply to our own testing, and we recommend a written internal policy before any commercial pipeline goes live.

A realistic 30-day rollout

Week 1 — measure. Pick one metro and 2,000 listings. Run tier 1 discovery only. Establish page weight, parse-field coverage, and how much of your target data is available without touching the detail page.

Week 2 — pilot residential. Buy the smallest paid plan from two providers, run identical 5,000-request workloads through both, and record success rate, median latency, bytes per success, and unique IP count. Do not extrapolate from free trials with curated pools.

Week 3 — build the delta loop. Wire search-tier daily crawls, diffing, and the enqueue path. This is where the cost curve bends.

Week 4 — harden. Add retry-with-new-identity, template-drift alarms, per-target concurrency caps, and a kill switch. Then scale the metro count, not the request rate per metro.

Frequently asked questions

Are datacenter proxies ever enough for Zillow or Redfin? For the sitemap and any robots-permitted crawl path, yes. For listing detail pages at volume, no — the edge layer scores hosting ASNs down before behaviour is even evaluated. Use datacenter where it works and stop paying residential rates for pages that do not need it.

How many IPs do I actually need? Think in requests-per-IP-per-hour, not pool size. Tier 3 portals tolerate roughly 8–20 requests per residential IP per hour before scores degrade. For 50,000 detail fetches spread over 12 hours, that implies a few hundred concurrently active IPs — well within any mid-tier residential plan. Advertised pool sizes of "80M IPs" tell you almost nothing about how many are live in your target city today.

Is SOCKS5 or HTTP better for scraping property sites? Functionally either works. SOCKS5 is lower overhead and passes DNS to the proxy when you use the socks5h scheme, which avoids a common leak. HTTP proxies are easier to debug and better supported by managed scraping frameworks. Our SOCKS5 setup guide covers the leak-safe configuration.

Do I need an antidetect browser? Only for the render tier and for anything behind a login. Stateless HTML fetching does not need a full browser profile; it needs consistent TLS and header fingerprints. If you do render, our antidetect browser comparison covers which tools hold up under portal fingerprinting.

What success rate should I expect? On tier 3 US portals with a well-built pipeline: 88–96% first-pass on quality residential, 70–85% on budget residential, and single digits on datacenter. If you are below 80% with residential IPs, the problem is almost always your request profile, not the proxies.

The short answer

For most real estate data pipelines in 2026: rotating residential for the detail tier, datacenter for discovery, ISP for anything that needs a stable regional identity, and mobile only where nothing else gets through. Block images, parse JSON-LD before HTML, rotate per request outside of pagination, and match locale to IP geography.

Get those five things right and provider choice becomes a cost optimisation rather than a survival question. Start with the tested shortlist on our best proxy providers page, verify with your own 5,000-request pilot, and let measured success rate — not marketing copy — pick the winner.