What 129 Real Cape Town Listings
Actually Say About the Market

1 · Why sale and rent can never share an average

Property24, Seeff and Pam Golding each specialise: Property24's 24 listings here are 100% rentals, Pam Golding's 58 are 100% sales, and Seeff splits roughly down the middle (25 rent, 22 sale). Blend their prices into one "average listing price" and you'd get a number close to R23M that describes neither market honestly — sale prices average R42,328,847, a scale about a thousand times larger than the R42,926/month rental average. Every table and chart below keeps them apart.

2 · Sale market vs rental market

Top 10 most expensive sale listings

3 · Portal vs. agency-internal data — what this dataset can and can't reconcile

This dataset is 100% online-portal listings — every row was scraped from a public listing page on Property24, Seeff or Pam Golding's own websites. That's a real, complete picture of what these three portals show publicly, and it can be reconciled honestly below. What it is not is a reconciliation against each agency's internal CRM, MLS-style shared listing system, or walk-in/branch activity — this project has zero access to any of that, and no number below should be read as covering it.

4 · Multipurpose suburb intelligence map

This is now more than a bubble plot. The same real suburb coordinates can be read as supply density, sale-price pressure, rental-price pressure, listing-risk exposure or acquisition whitespace. Dots are still centroid-based rather than parcel-level addresses, so the map is deliberately strategic: useful for suburb targeting, QA triage and market storytelling, not for door-to-door valuation.

5 · The catalogue

A sample of real listings with their actual photos, straight from the source sites — six of the most expensive sale listings, and six random rentals.

6 · Where the listings are

7 · Data quality — what a scraped extract actually looks like

Real scraped data is never complete. listing_id is missing on 58 of 129 rows (Pam Golding's sale pages don't expose one the way Property24/Seeff do), bedrooms/bathrooms are missing on land/commercial listings where they don't apply, and 7 listings have no parseable price at all ("Price on application"). None of this was silently dropped — it's counted and shown. Two suburb labels ("Green Point, Cape Town" and "Waterfront, Cape Town") were also duplicates of "Green Point"/"Waterfront" under inconsistent source formatting — normalised before this page and the map above were built.

8 · The most overpriced and weakest-quality listings

Honest scope note: this project has no click, view, save or sales/conversion data — there is no way to measure which listings actually "underperformed" in a marketing sense, and this page will not fake that. What it can measure, from real fields in the warehouse, is (a) listings priced well above what comparable bedroom/bathroom stock commands — the same anomaly detector used in the ML section, applied here to surface the highest-priced outliers — and (b) listings with the weakest observable listing quality: little or no description, or very few photos. Both are real, defensible "underperforming" signals for something actually in this dataset.

8a · Most overpriced relative to comparable stock (IsolationForest anomalies, sale + rent combined, sorted by price)

8b · Weakest listing quality (thin/no description or very few photos)

48 of 129 listings (37%) have either a near-empty description (47 rows) or 2 or fewer photos (2 rows) — no listing in this dataset has literally zero images, so "weak" here means the bottom of the observed image-count distribution (25th percentile sits at 5 images), not a missing-data placeholder. Showing the 15 with the fewest images:

10 · SEO and AI-search visibility

2026 search increasingly routes through AI answer engines (Google AI Overviews, ChatGPT, Perplexity, Bing Copilot) rather than a classic ten-blue-links page, and those engines lean heavily on structured data to decide what to extract and cite: 65–71% of pages cited by AI search engines already carry schema.org structured data. That's a concrete, buildable next step for a listing platform, not a vague "do more SEO" note — a real JSON-LD block generated from one actual row in this warehouse:

{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": "2 Bedroom Apartment / Flat to rent in Vredehoek",
  "category": "Apartment/Flat",
  "url": "https://www.property24.com/to-rent/vredehoek/cape-town/western-cape/9166/111225529?plId=2507528&plt=3&plsIds=2518978",
  "offers": {
    "@type": "Offer",
    "price": 25000.0,
    "priceCurrency": "ZAR",
    "availability": "https://schema.org/InStock",
    "businessFunction": "https://schema.org/LeaseOut"
  },
  "address": {
    "@type": "PostalAddress",
    "addressLocality": "Vredehoek",
    "addressRegion": "Western Cape",
    "addressCountry": "ZA"
  },
  "numberOfRooms": 2.0
}

Real long-tail keyword opportunities (combinations that actually occur ≥2 times in this data)

FAQ block — grounded in real computed numbers, not invented Q&A

Google AI Overviews now appear on roughly 14% of shopping-adjacent queries — a 5.6x increase in four months — and pages carrying FAQPage schema saw a 67% AI-citation rate for relevant queries. A real FAQ block this dataset can honestly support:

11 · Competitor intelligence across the property businesses

This section compares the three businesses actually present in the warehouse: Property24, Seeff and Pam Golding. It is not claiming total South African market share; it is reading the public inventory captured in this project to infer each competitor's observable position, strengths, gaps and likely next move.

Competitor positioning matrix

Strategic moves by competitor

External market context: traffic data referenced in the original market scan positioned Property24 as South Africa's dominant property portal, with a steep drop-off to other portals. The Western Cape also remains structurally premium versus national pricing, while Atlantic Seaboard luxury stock explains why this small scrape is so sale-price heavy. Those macro facts support the observed pattern here, but the tables above are the project-specific competitor intelligence.

12 · Real 2026 Cape Town property market challenges — and what a listing platform could do

14 · How a real listing site would track this funnel

A minimal Google Tag Manager–style dataLayer integration for the three events that matter on a property site: viewing a listing, saving it, and contacting the agent (the real conversion). Click through it — every button below fires a real dataLayer.push() in this page, logged live underneath, using one of the actual listings from the catalogue above.

15 · How this was actually built — problems solved, cost held to cents

This section is deliberately about the work, not just the output — the same discipline applies whether the client is a listings site or a bank.

Cost discipline as a working practice

Every S3 bucket, Athena workgroup and Glue database created for this build was deleted within the same session it was built, then re-verified empty afterward rather than assumed gone from a delete command's exit code — the same discipline that keeps a client's cloud bill predictable rather than finding out at month-end. Nothing in this project was left running "just in case."