X posts: programmatic SEO, data layers, GEO, llms.txt, schema, penalties
On this page
Collected 2026-10-25 (Asia/Dubai) via the shared public X read tools (search_posts_all, get_posts_by_ids). Metrics are as returned by the API at collection time (likes L, reposts RT, replies R, impressions imp). URL pattern: https://x.com/<handle>/status/<id>. Nothing here was fetched from x.com directly.
Reading guide. Most posts on these topics are low-engagement and many are vendor promotion. I list only posts that carry a concrete claim, a number, or a sourced opinion. "Claimed/unverified" means the post asserts a result I could not confirm from a primary source. Posts marked [verified] have a claim I cross-checked against a fetched primary source (see cases.md / best-practice.md).
Gully / INVT. Searched "Gully INVT", "INVT learnings", "Gully data layer". Zero relevant results. search_users is not available on the shared X access, so I could not look up a handle. Gully/INVT is not identified from any public X source. Unverified; not guessed.
1. "Data layer / proprietary data" thesis (the Anthony framing)
| Post | Author (followers) | Date | Key claim | Engagement | Status |
|---|---|---|---|---|---|
| 2000583082345169008 | @thinking_slow (13.9k) | 2025-12-15 | Brands with proprietary data and a low-friction business model would have been "irresponsible" NOT to do pSEO; traffic declines are not failure, the opportunity was real. | L63 RT2 R14 Q4 imp13,588 | Opinion, balanced. Useful counterweight to "pSEO is dead". |
| 2037644134794895747 | @JHTScherck (9.6k) | 2026-03-27 | pSEO works only with "a defensible reason to rank" (unique data, proprietary metric). "Loading up the page with GPTslop aint it." | L12 imp532 | Opinion; matches Google's originality criteria. |
| 2078109439039873388 | @askOkara (78.5k) | 2026-07-17 | 2026 SEO trends list includes "original data and proprietary research" and "programmatic seo with human oversight". | L23 RT2 imp3,487 | Opinion/list. |
| 2074406486160396745 | @askOkara (78.5k) | 2026-07-07 | Google can de-index large volumes of auto-generated content lacking depth; include proprietary data; "the best programmatic seo pages do not feel programmatic." Also advises varying publish dates and 5-10 pillar pages first. | L25 RT2 imp3,068 | The "vary publish dates" advice is folklore, not in Google docs. Unverified. |
| 2050821923974807757 | @sachinaggarwal (237) | 2026-05-03 | Claims Zerodha became India's largest broker via programmatic pages on every stock/MF/ETF built from exchange data. | L0 imp73 | Claimed/unverified (causal claim not sourced). |
| 2075840231203635280 / 2075825593074037245 | replies to @om_patel5 | 2026-07-11 | "programmatic SEO without a real dataset just turns into thin pages fast" / "need the infra to aggregate demand for those citations". | L1 | Low-signal replies. |
| 2047695571159330922 | @DomJoLuna (2.1k) | 2026-04-24 | Smartest companies are "building the data layer AI" agents can use rather than "optimizing for AI". Cites Gartner "traditional search will drop 25% by end of 2026". | L0 imp49 | Gartner stat second-hand; unverified here. Shows the "data layer" phrase in the wild. |
| 2084571540880855125 | @ayan_sorouri (0) | 2026-08-04 | "Rebuilding the data layer": entity-rich JSON-LD, unblocking AI crawler rendering, third-party co-citations. | L0 | Here "data layer" = markup/entities, not a data pipeline. Terminology differs from Anthony's. |
Note on terminology: on X, "data layer" is usually either (a) vendor jargon around GEO tooling or (b) schema/entity markup. Anthony's meaning (an ingested, cleaned, proprietary dataset powering pages) is better described by the phrases "proprietary data", "unique data module", "information gain", "defensible data moat" in the posts above.
2. Success claims (all claimed/unverified unless noted)
| Post | Author | Date | Claim | Engagement | Status |
|---|---|---|---|---|---|
| 1914654175553888443 | @iannuttall (81.9k) | 2025-04-22 | MCP directory built with AI agents: ~100k pageviews/month within 30 days (35k visitors, 92k pageviews, 18k from Google, 262 Perplexity, 169 ChatGPT). Daily scrapers update stats; articles rewritten when server details change. | L187 RT10 R20 imp17,146 | Self-reported. Closest public analogue to a "daily-changing data layer". AI referrals were ~1% of visitors in his own numbers. No later-state data. |
| 1885413639386976265 | @levelsio (981.9k) | 2025-01-31 | Nomad List's filter.db records every filter combination, generates slugs, and these long-tail filter pages "rank REALLY well". A DB slowdown made him stop writing to it. |
L500 RT13 R29 imp297,035 | Primary (founder). Excellent mechanism: user queries become the page inventory. |
| 2075820626468552973 | @om_patel5 (27.3k) | 2026-07-11 | 0 to 2.15M Google impressions and 6,000+ clicks in 3 months via pSEO off a proprietary dataset plus "gseo". | L44 imp4,975 | Claimed/unverified. Note 6,000 clicks on 2.15M impressions is a 0.28% CTR, i.e. impressions are not traffic. |
| 2053628866313302516 | @KorayGubur (37.6k) | 2026-05-11 | Live fully-programmatic SaaS site: YoY clicks +950,908 (+44.2%), impressions +7.0M (+10%). Credits technical SEO, PageRank per document, micro-semantics. | L63 imp5,640 | Claimed. Practitioner with a technical-SEO angle; no site named. |
| 2030696471125778561 | @KorayGubur | 2026-03-08 | pSEO project's daily clicks 27,000 to 42,000 in 3 months from technical/semantic fixes. | L78 imp7,283 | Claimed. |
| 1976306304340852933 | @WhizofGEO (157) | 2025-10-09 | Flyhomes 18.7K to 2M monthly visits in 3 months (+10,737%). | L9 | Number is real in the source (aioseo.com, Ahrefs estimates, see cases.md) but the tweet omits that ZeroDown (acquired June 2024, 301-redirected) already had ~504.5K monthly visits in March 2024 (per aioseo); the spike is not purely pSEO. |
| 2011513951906119971 | @fba (18.5k) | 2026-01-14 | Agency claims multiple clients taken "0 to 1M+" monthly visitors via SEO. | L81 R16 imp16,168 | Agency marketing; unverified. |
| 2111176639061111077 | @Shreedhan (63 fol.) | 2026-10-16 | Own pSEO site: 0 to 10,000 indexed pages and 6,000 impressions/day in 3 weeks; then -72% impressions, -65% clicks, 2,300 pages wiped overnight, blamed on a sitemap bug. | L0 imp49 | Claimed. Small, but a useful operational failure mode (sitemap bugs). |
3. Cautionary, contrarian, penalty posts
| Post | Author | Date | Claim | Engagement | Status |
|---|---|---|---|---|---|
| 2046200379851878649 | @glenngabe (85.7k) | 2026-04-20 | National Today's /us/ directory (850K+ AI-generated local-news URLs) got a manual action and was fully deindexed; he checked ChatGPT citations too. | L34 RT8 Q9 imp11,554 | [verified] against gsqi.com post (fetched). |
| 2048038265777729575 | @glenngabe | 2026-04-25 | "Drop in Google and drop in ChatGPT." Rankings fell across Google surfaces incl. AI Overviews and AI Mode. | L8 R4 imp3,030 | [verified] same source. |
| 1948008401348489522 | @glenngabe | 2025-07-23 | From Search Central Live: Gary Illyes said normal SEO works for AI Overviews, "You don't need GEO, LLMO or anything else." | L23 RT4 Q3 imp2,520 | Second-hand report of a talk; consistent with Google's published AI optimization guide (fetched). |
| 2110439253624967595 | @rafalabradorseo | 2026-10-14 | Search Central Live Barcelona (Oct 2026): AI Overviews/AI Mode use the same crawl and index as Search; "Good SEO is good GEO"; llms.txt not needed. | L0 imp37 | Second-hand; consistent with Google docs. |
| 2108191210246443193 | @AlDev0 | 2026-10-08 | John Mueller compared llms.txt to the keywords meta tag; "ship it if you want. But it's not why you're not getting cited." | L5 imp136 | Mueller comparison second-hand; unverified. |
| 1897002863081218417 | @lilyraynyc (124k) | 2025-03-04 | Asks whether anyone has seen a "scaled content abuse" manual action; sites she saw got "pure spam" manual actions or algorithmic demotions. Gabe replies the same. | L21 imp3,721 | Practitioner observation (2025); taxonomy of penalties is murky. |
| 2105959297473761301 | @samigrows (1.3k) | 2026-10-02 | Client lost 90% traffic overnight; GSC showed manual action "major spam problems" after a tool bulk-published 200+ thin pages. | L9 imp885 | Anecdote; unverified. |
| 2074933397944689076 | @DavidGQuaid | 2026-07-08 | Reddit user got a "Major spam problems" manual action citing scaled content abuse, cloaking, scraping. | L5 imp1,547 | Anecdote via Reddit. |
| 2106973749954359435 | @Flozi_ | 2026-10-05 | Summarises schema research: AI-cited pages ~3x likelier to have schema (correlation); Ahrefs test of 1,885 pages found no meaningful lift; AI systems mostly read visible HTML. | L1 imp49 | [verified] Ahrefs study fetched. The searchVIU "ignore JSON-LD" claim is reported via SEJ, not fetched at source. |
| 1991244375444320262 | @SERanking | 2025-11-19 | "Only 10.13% of sites use LLMs.txt"; no impact on AI mentions across ~300k domains. | L1 imp274 | [verified] SE Ranking post fetched. Vendor study; see best-practice.md caveats. |
| 2023127268009431422 | @KrittinKalra | 2026-02-15 | Own pSEO site deindexed by Bing, "not random". | L6 imp109 | Anecdote; unverified. |
| 1977247556590878809 | @arre_ankit | 2025-10-12 | Says Lovable's pSEO traffic dropped because of AI content with no human review, "likely" triggering scaled content abuse. | L5 imp893 | Speculation; unverified. Do not cite as fact. |
| 2055592258695545228 | @aykut_cevik | 2026-05-16 | Rebuts Google's guide: claims 44% of LLM citations come from the first 30% of a page, FAQ-schema pages cited 40% more by ChatGPT, 85% of brand mentions come from external domains, etc. | L0 imp40 | Unsourced vendor-style stats; unverified. Directly contradicts Google and the Ahrefs schema test. Treat as hype until a methodology is shown. |
| 1995055195827720288 | @hiayoola (n/a) | 2025-11-30 | Tutorial: enable llms.txt in Yoast in 60 seconds. | L90 RT21 imp10,717 | Shows tooling adoption outpaced evidence of benefit. |
| 1991922823129075746 | @andilosing (10k) | 2025-11-21 | "llms.txt... does it help for AI? I don't know. You don't know. But it doesn't hurt." One-click generator. | L27 imp2,375 | Honest-uncertainty framing, then a product pitch. |
| 1937893163353411624 (and repeats) | @TomPick | 2025-06-25 | Shares "Beware the 'Generative Engine Optimization' Snake Oil". | L0 | Low engagement; the "GEO is snake oil" stance is common among SEOs. |
4. Tool and platform chatter (vendor posts, not evidence)
- 2046242232747925788 @Gkapogiannis (2026-04-20): Conductor "2026 AEO/GEO Benchmarks Report" (vendor report; not fetched).
- 2045591192717476030 @Gkapogiannis (2026-04-18): names Scrunch, Semrush AI Visibility Toolkit, Athena as AI-visibility monitors; "31% of US users expected to use generative AI search this year" (unsourced).
- 2074860806441726070 @johnbuildss (2026-07-08): "Bloomiro" pipes GSC + AI-visibility data into Claude (MCP-style). Vendor.
- 2084654051254280266 @naqui_s (2026-08-04): Pierview Claude Connector in the Claude Connectors Directory. Vendor.
- 2091356563495960802 @JulianGoldieSEO (171k followers, 2026-08-23): API that sends one prompt to ChatGPT/Gemini/Perplexity/Grok/Copilot and returns structured JSON. Vendor/promo.
- 2002941017243197794 @mention_network (4.4k followers, 2025-12-22; L60 R43 imp11,669): promotional explainer on where AI answers come from (pre-trained data, retrieval). Vendor content; no evidence offered in the visible text.
- 1973468460392006127 @aleyda (182k, 2025-10-01): shares Prerender AI Indexing Benchmark on e-commerce AI citations (L55 RT16); report not fetched.
5. Take-aways from the X layer
- Consensus among practitioners with larger audiences: pSEO still works only when each page has data the template did not supply; "unique data" is the repeated survival criterion.
- The loudest quantified success claims are self-reported (Nomad List and iannuttall are the most credible because the mechanism is explained).
- The most cited failure documentation is Glenn Gabe's National Today case study (fetched, verified).
- The GEO tactic debate (llms.txt, schema, chunking) splits cleanly: Google staff and independent studies say no measurable effect; vendors and a few posters assert large lifts without methodology.
- The shared X access cannot find a specific person's handle, so any Gully/INVT identification must come from internal material.