Documented cases: data-led and programmatic sites, winners and losers
Collected 2026-10-25 (Asia/Dubai). Traffic figures are third-party Ahrefs or Similarweb-style estimates unless the row says otherwise. Many "case studies" are written by agencies or vendors selling services; I label each source. Causation is rarely proven.
| # | Case | Data asset | Result (with source and caveats) | Lesson |
|---|---|---|---|---|
| 1 | National Today /us/ (loser, penalty) | 850K+ AI-generated local-news URLs, rewritten from other outlets' reporting (per Futurism, via Somethinginc) | Manual action for scaled content abuse; section fully deindexed; dropped in AI Overviews/AI Mode; most ChatGPT citations vanished though a few remained. Glenn Gabe case study (primary) https://www.gsqi.com/marketing-blog/when-mt-ai-crumbles-chatgpt-follows/ ; secondary summary https://somethinginc.com/blog/scaled-ai-content-manual-action-case-study/ | No proprietary data: scaled rewrites of others' work. A Google deindex carried into AI answers in this case. One case, not a law. |
| 2 | Wise (currency pages) | Templated "X to Y" currency pair pages across ~180 currencies | Ahrefs via MV3 (agency, pulled 2026-09-21): 39.1M visits Jan 2024, peak 109.9M Apr 2025, about 65M Sep 2026, roughly -40% from peak. MV3 says the cause is unconfirmed (AI Overviews expansion and spam enforcement are hypotheses). https://www.mv3marketing.com/blog/pseo-case-study-data-zapier-wise-ramp-2026/ | Even a massive winner can fall 40% on commodity queries. Data that is public elsewhere (FX rates) is not a moat. |
| 3 | Zapier integration pages | App-to-app pair pages from a template | Same source: 10.9M visits Feb 2025 to 3.49M, below the Jan 2024 start of 4.4M. Top pages now are "best of" roundups, not pair pages. seosavages: Zapier /apps/ has 98.8% of pages at <=100 visits/month (5,604 tracked pages). | Template-only pairs without unique data decay. Ahrefs-estimate caveat. |
| 4 | Ramp (winner, calculators and proprietary data) | State/city rate tables behind calculators; monthly "AI Index" and "SaaS Vendors Leaderboard" from its own card-spend data | MV3/Ahrefs: 76K visits Jan 2024 to 394K Apr 2026 (5.2x), then 295K (-25% from peak). MV3 reports NYT/Atlantic citations of its indices (not independently verified here). | The closest match to a data layer: owned data, narrow intent, link magnet. Still volatile. |
| 5 | Flyhomes (spike, partly acquisition) | Real-estate pages, "Cost of Living/Quality of Life" templates; acquired ZeroDown (June 2024) and 301-redirected it | aioseo (vendor): 18.7K to ~2M monthly visits in 3 months; ZeroDown already had 504.5K visits and 804.2K keywords in March 2024. Page estimates: 10,083 (Apr) to 425,336 (Jul) pages (via Medium teardown snippet). The Medium teardown (title: "...The Catch Behind a Viral SEO Win") is paywalled. aioseo says Flyhomes "likely" used pSEO. https://aioseo.com/trends/flyhomes-seo-case-study/ | Headline growth was partly inherited authority. Do not attribute a spike to pSEO alone. No later-state data fetched, so durability is unknown. |
| 6 | Zillow (long-run winner) | Zestimate: automated valuations on nearly 100M homes (2011 prospectus) | Company-reported 233M avg monthly unique users (Q3 2024); ~33M visits/month is a third-party estimate. seo.to is an SEO-agency write-up that labels each figure. https://seo.to/case-studies/zillow-data-product-seo | A proprietary data product plus 20 years. Not replicable in 12 months, but the pattern (data product feeds pages and links) is. |
| 7 | Nomad List (long-run winner) | User-submitted city data, six page types; filter combinations stored as slugs | Founder: filter.db generates long-tail filter pages that "rank REALLY well" (X, 2025-01-31, L500, imp297k). viewengine.ai (vendor): ~50K+ monthly organic visits, 25.5K indexed pages (citing upgrowth.in, not fetched). |
User queries become page inventory; the data is first-party. Traffic figure is old, low-quality sourcing. |
| 8 | The dead tail: nine big templated sites | Zapier /apps/, G2, Yelp, Tripadvisor, Booking, Dictionary.com, Wise, Canva, NerdWallet | seosavages (agency lab post), Ahrefs data, 12.3M pages: 28.8% have zero estimated traffic; 96.7% get <=100 visits/month; only 0.3% exceed 1,000. Per site: Yelp 98.0% and Tripadvisor 96.7% and Booking.com 94.8% at <=100; Canva 72.6%; NerdWallet 55.8%. Ahrefs only tracks pages with some footprint, so the real dead share may be higher. https://www.seosavages.com/lab/does-programmatic-seo-work/ | Most pages in any pSEO footprint earn little; traffic is concentrated in the few that match real demand. The Benidorm analogue (Tripadvisor/Booking tails) says: build fewer, better pages. |
| 9 | Iannuttall MCP directory (small winner, daily refresh) | Scraped open-source MCP servers plus npm download stats, re-checked daily; articles rewritten on change | Self-reported on X (2025-04-22): ~92K pageviews in 30 days; 18K Google visitors; 262 Perplexity and 169 ChatGPT visitors. | Closest public analogue of a daily-changing data layer; also shows AI referrals were ~1% in his numbers. Self-reported, no follow-up fetched. |
| 10 | Shreedhan sitemap failure (small loser, ops) | 10,000-page pSEO site | X (2026-10-16), self-reported: 0 to 10,000 indexed pages and 6,000 impressions/day in 3 weeks; then -72% impressions, -65% clicks, 2,300 pages lost, blamed on a sitemap bug. | Operational bugs (sitemaps, canonicals) are a first-class risk. Unverified single anecdote. |
| 11 | Schema experiment (null result) | 1,885 pages that added JSON-LD (Aug 2025-Mar 2026) vs 4,000 matched controls | Ahrefs: AI Overviews citations -4.6% vs controls (statistically significant), AI Mode +2.4% and ChatGPT +2.2% (not distinguishable from zero). Both groups were already declining. https://ahrefs.com/blog/schema-ai-citations/ | Adding schema did not buy citations in this test. Vendor study on its own corpus; Ahrefs sells SEO/AI tools. |
| 12 | llms.txt study (null result) | ~300,000 domains | SE Ranking: 10.13% of domains had llms.txt; no correlation with AI citations; removing the variable improved the XGBoost model. https://seranking.com/blog/llms-txt/ | No evidence it helps. Observational and vendor-run. |
| 13 | Dubai DLD data platforms (market context) | Official DLD transactions and Ejari contracts via Dubai Pulse | propertyindex.ae (vendor self-description) notes that different platforms produce different medians from the same DLD records because of inclusion and grouping choices; dxbdata.io (first-party) serves 8,480 transactions across 131 areas via MCP. No traffic data. | Methodology disclosure (asking vs registered prices, medians vs indices) is the product. Competitors already expose DLD data to agents. |
Cases I looked for and could not document
- A dedicated, sourced travel/destination-data case (Benidorm-like). Nearest are Tripadvisor/Booking in #8 and Nomad List in #7.
- Lovable's decline: only a speculative X post (2025-10-12). Not treated as a case.
- Recovery cases from manual actions: only anecdotes on X (e.g. @samigrows, 2026-10-02). No verified recovery.
Pattern across cases
- Survivors own or generate the data (Zillow, Nomad List, Ramp). Losers or decliners reformat commodity data (Wise rates, Zapier pairs, National Today's rewrites).
- Even good footprints are concentrated: the top 0.1-0.3% of pages carry much of the traffic (seosavages: G2's 0.12% of pages capture 30% of section traffic).
- Third-party traffic numbers conflict widely (MV3 found Zapier quoted from 2.6M to 16M); always date and source them.
- Headlines compress timelines: Flyhomes' 3-month spike included an acquired domain's authority.