SEO Recap covering October 14, 2026: Google SERP poisoning claim and the Non-Canonical Cabinet URL series

Built by Stephanie Chung·5 min read·6 stories

Daily summary of what matters in SEO, GEO, AEO, and AI search generated with Claude Code (beware of hallucinations)

AI in Search / AI Overviews

Shopify switched on WebMCP tools for every Liquid storefront on August 5, and 89 of 122 stores answered a standard MCP tool listing. Orders from AI-powered searches grew nearly 13 times year over year in Q1 2026.
Agent-readiness is mostly Shopify's rollout, not merchant work: 76 of 86 product pages load the same WebMCP adapter and 83 of 89 agents.md files are near-identical copies of a Shopify template, with only 5 written by merchants.
Zero of 122 stores passed all 9 purchase-decision checks. Common gaps include delivery dates missing everywhere, pickup never declared (93 stores list shipping only), and no tool to cancel, change, or return an order.
Most policy data is not machine-readable: 70% of stores state return windows only in prose, and just 7% of product pages expose structured merchantReturnDays.
Being cited or ranked does not fix the action layer. Fixing markup once (labeled selectors, named buy buttons, announced cart confirmations) serves screen readers, keyboard users, and browser agents together.

Google is testing placing the AI Mode vertical left sidebar on standard web search results, adding the super G logo, chat history, a way to start an AI Mode chat, and settings.
Spotted by Damien Andell, the test would make AI Mode more accessible and brings it a step closer to being the default for Google Search.
Marie Haynes noted the sidebar surfacing prior Gemini conversations, though she later clarified these came from the Gemini overlay on Android rather than the Gemini app merging with AI Mode.

Google's Non-Canonical Cabinet video tackles URL edge cases
Sources: Search Engine Journal · Search Engine Roundtable

Gary Illyes and Cherry Prommawin launched "The Non-Canonical Cabinet," a Search Central series on edge-case technical questions, with the first episode published October 14 covering URLs and a robots.txt episode next.
Illyes called keywords in a URL "an incredibly small signal" that only matters when Google has nothing else to go on and becomes basically irrelevant once content exists. He noted Google can detect URL spamming.
Google follows IETF standards (RFC 1035, RFC 1123) for domains and does not allow emoji in domain names, but Googlebot will crawl and process emoji in URL paths, which browsers percent-encode.
Illyes reaffirmed hyphens as the recommended word separator, since underscores cause Google to treat joined words "as one long word."

Martin Splitt said Google runs a converter that turns PDFs into HTML and indexes that HTML, not the PDF itself, so the rest of the pipeline works only with HTML.
Because of the HTML conversion, links embedded in PDFs are followed like normal links, and PDFs can rank in regular search, AI Overviews and AI Mode.
Image-only PDFs may not rank for text. If you can highlight and copy the text into a text editor, Google can convert and use it for search.
Mueller said HTML is generally preferable to PDF for ranking because it is easier to interact with, though PDFs remain fine when formatting, print, or forms require them.

Nozzle reports non-standard SERPs started appearing September 20, peaking at 25% of results, and have run 5-11% per day since October 1. The firm calls it poisoning because Google returns a complete page with wrong results and no indication they are wrong.
Normal results keep 88.5% of the same sites across repeated searches for a keyword, but poisoned results retain only 20.7%, producing artificial ranking volatility.
Poisoned SERPs lean toward user generated content, with Facebook, Instagram, LinkedIn, Quora and Reddit appearing more often while MapQuest, Yelp and YouTube appear less.
Expensive features show up less: AI Overviews appear 34.9% of the time versus 54.2% in baseline, and "people also search for" drops to 52.7% versus 87.2%. Image packs nearly double to 44.2% versus 23.0%.
Local targeting made no difference across 21,508 national-versus-local pairs (7.5% and 7.4%), and advertising and shopping results stayed stable.

The Helpful Content Update purged SEO-first sites with no presence outside search. HouseFresh lost over 90% of traffic, falling from ~4,000 daily visitors in September 2023 to ~200 in early 2024, while Reddit surged ~190% after its Google data-licensing deal.
Google is building a loyalty ecosystem through Preferred Sources, follows, Discover profiles, and subscription linking, rewarding brands and creators with an actual audience.
86% of Google clicks go to a site the searcher did not name, leaving only 14% navigational out of more than 5 trillion annual searches.
Per iPullRank data cited, 52.3% of searches by 18-24s end without a click versus 42.7% for 55-64s, and 55-64 year olds are 7 times more likely to click an ad than the youngest cohort.
Measure incremental uplift rather than last-click attribution, and diversify into video and creator-led formats that are booming in Discover.

Published on October 15, 2026