# Field check: do SERP API settings change how "poisoned" Google results look? (ryandoser.com, best-serp-api) Written 2026-11-16 BEFORE any field-check data was collected. Nothing in the October live-check raw files has been opened or analyzed for this plan (their analysis is pre-registered below). ## Why On 2026-11-16 Ilias Ism posted a "clean SERP rate by provider" chart (Oxylabs 96.8%, Cloro 95.8%, Bright Data 61.3%, Serper 20.8%, DataForSEO Live 9.7%). Three of those five figures match cloro's own vendor benchmark (cloro.dev/blog/google-poisoned-serps, "12 hours to 06:00 UTC 31 Oct", three reference queries). Separately, Google started wrapping organic links as google.com/goto (confirmed to SERoundtable Aug 26, 2026), DataForSEO Live resolves "approximately 50%" of them while Standard resolves all (DataForSEO Nov 2 update), and DataForSEO shipped `resolve_poisoned_results` on 2026-11-16. Question for readers choosing a SERP API: how much of a "poisoned" score is the route and settings you pick? This is a small FIELD CHECK, not a study. It is labeled that way on the page. ## Arms (same keywords, same round, run back to back) - A. DataForSEO `serp/google/organic/live/advanced`, default params (resolve_goto_urls false). - B. DataForSEO live/advanced, `resolve_goto_urls: true`. - C. DataForSEO Standard queue `task_post` priority 1 (goto resolved by default), `resolve_poisoned_results` false. - D. DataForSEO Standard queue, `resolve_poisoned_results: true`. - E. Apify `apify/google-search-scraper`, countryCode us, languageCode en, 1 page, desktop. (An independent scraping route, not a named SERP API vendor from the chart.) All: United States (2840), English, desktop, depth 10 (one page). ## Keywords (frozen) - Set K: the 100 keywords frozen on 2026-10-14 in `../poisoned-google-serps/live-check/keywords.json`. - Set N: 20 navigational brand queries with one official domain each (listed in `nav.json`, frozen at this commit). No Ryan or client property. ## Rounds 2 rounds, the second starting at least 60 minutes after the first ends. Every raw response saved to `raw/r//.json`. ## Definitions (written before data) - Site = registrable domain (www/subdomains folded, SL2 list as in the October analyzer). - Raw site = from the `domain` field (DataForSEO) or the URL host (Apify), as a domain-matching tool would read it. - Real site = raw site, except when the raw site is google.com and the URL is a goto/redirect link: then the host of `website_name`/breadcrumb (DataForSEO) or the decoded target URL. If no real host can be recovered, the slot counts as unknown. - Organic top 10 = first 10 `organic` items by rank. - Masked share (metric 1) = share of organic top-10 slots whose raw site is google.com. - Baseline = latest DataForSEO Labs historical SERP snapshot dated 2026-08-01 to 2026-09-12 (`../poisoned-google-serps/live-check/hist_b2/`), used only when it has at least 5 distinct organic top-10 sites, and only for non-local keywords (kind != near_me and no local_pack in that arm-round's response). - Baseline overlap = share of baseline top-10 sites present in the pull's organic top 10. - Poisoned-pattern pull (metric 2) = baseline overlap below 0.5. Computed twice: with real sites (primary) and with raw sites (the "naive" reading). - Brand erased (metric 3, set N) = official domain absent from the organic top 10. Real and raw readings. - A failed or empty response counts as failed (reported separately, never dropped silently). ## Reported, whatever the result Per arm: masked share, poisoned-pattern rate (real, raw), brand-erased rate (real, raw), failure count, cost. By pull and by keyword. The arm-to-arm difference on the same keywords is the finding. Rates are "in this check", with dates. ## Caveats stated on the page Baseline is 2-3 months old, so some real churn reads as "below half" in every arm. That is why arms are compared to each other, same hour, same keyword. One country, one device, two rounds, convenience keywords. Apify is one scraping route, not a stand-in for Oxylabs, Cloro, Bright Data, Serper or SerpApi (not tested: no accounts). Ryan uses DataForSEO daily and links to it with an affiliate code: disclosed. ## October re-analysis (pre-registered here, data already collected Oct 14-15, never analyzed) 500 pulls (5 rounds x 100 keywords) from DataForSEO live/advanced. Report for context only: masked share and poisoned-pattern rate (real and raw) with the same definitions. Labeled "October 14-15 pulls, 5 of 9 planned rounds". ## Kill bar - Cost: stop if field-check spend passes $5 (planned about $2.50). - If an arm fails on more than 30% of pulls, its rates are reported as inconclusive. - Story bar: if no arm differs from arm A by at least 10 percentage points on metric 2 (real), the page says the settings did not change the result in this check. ## Checks A second script written independently recomputes metrics 1-3 for 50 random pulls (seed 20261116) from raw JSON. 50/50 agreement required (labels are deterministic code). Claude agents/scripts collected and labeled. Disclosed. ## AMENDMENT 1 (2026-11-16, before any metric was computed) Disclosure: written after viewing one raw record per arm for "hubspot" (n01) in round 1 (arms A and E), to check field formats. No metric had been computed. - Real-site recovery for a masked DataForSEO slot (domain google.com + goto URL), in order: (1) the decoded `url=`/`q=` target if it is a plain http URL; (2) `breadcrumb` only when it starts with http; (3) `website_name` matched against a fixed platform table (the text before " ยท ": LinkedIn, X, Instagram, Reddit, Facebook, YouTube, TikTok, Wikipedia, Pinterest, Quora, Yelp, Amazon, Medium, GitHub; mapped to their registrable domains); otherwise unknown. Reason: social-profile results carry breadcrumbs like "1.8M+ followers", which are not hosts. - Masked (metric 1) = raw site google.com AND the link is a goto/redirect link. Google's own /search links (Apify returns these as organic rows) are Google pages, not masked results. They stay as google.com in both readings. ## AMENDMENT 2 (2026-11-16 ~23:50 UTC, after round 1, before round 2 starts at 00:21 UTC) Reason: DataForSEO's 2026-11-16 update says `resolve_poisoned_results` also works on Live ($0.004 per task); the Live docs page read before round 1 did not list it. - New arm F, round 2 only: DataForSEO live/advanced with `resolve_poisoned_results: true` (resolve_goto_urls left at default false; real sites recovered by the Amendment 1 rule). Same keywords, params and metrics as other arms. - F is reported separately as "round 2 only" with its own denominators. Arms A-E keep the original definitions and pooling. Round-1 results for A-E were seen before this amendment (disclosed). Cost adds about $0.48, total field check stays under $5. ## AMENDMENT 3 (2026-11-17 ~00:05 UTC) Round 2 (and arm F, which existed only in round 2) cancelled on Ryan's instruction to publish sooner. Published results are round 1 only (2026-11-16, 22:57-23:20 UTC). The "2 rounds" in the plan was not met, and the page says one round. ## Hash log (UTC timestamp after each SHA-256 line refers to the PLAN.md version just hashed) 2026-11-16T22:56:43Z 7720cd45b4cde82f95c990d5ccd582c7d1537072f55f46d571a6e5279436e599 PLAN.md 159a8f54074b6e6c0b223700a60d709f605697104ccb772193f46bfc56491a8c nav.json 4e162541d2f6a7913c4065f101123408b2e3209e89ef8558fbee22eadf66f17d ../poisoned-google-serps/live-check/keywords.json 2a4e6fe47ca63f27be02e6e8aab9d39526da52ca3cd5358cb52f75050ada7dfd PLAN.md 2026-11-16T23:04:41Z 8637c6287b6a55374781521cd2fe6a3f97fafab6f62a671f26b82df1aae02583 PLAN.md 2026-11-16T23:47:53Z 66cde670dbb5473faf7660ebaa784434f4ffd28d78cda98fb358fd7d81655452 PLAN.md 2026-11-17T00:11:28Z