SEO Recap covering September 28, 2026: AI Mode monitoring goes global and the September spam update bites

Built by Stephanie Chung·7 min read·9 stories

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

Generative Engine Optimization (GEO) & Answer Engine Optimization (AEO)

Publisher court filing quotes OpenAI staff on links and cites Bing Chat CTR drops
Source: Search Engine Journal

  • A summary judgment brief from news publishers suing OpenAI and Microsoft quotes an OpenAI engineer writing in February 2023 that "no matter how prominently we show the links, users won't click," and ChatGPT head Nick Turley saying there was "no good reason to click" a source once ChatGPT answered.
  • The brief cites Microsoft data showing click-through rate was much lower in Bing Chat than Bing Web Search:
    • 87-93% lower for The Times's sites
    • 83-91% lower for the Daily News plaintiffs
    • 51-94% lower for Ziff Davis sites
  • CTR is defined as clicks relative to impressions in a results page or chatbot response, so the ranges don't measure traffic or visits. The passage omits when the data was collected, how many queries it covered, and how it was selected.
  • OpenAI and Microsoft argue their use is fair use and filed their own summary judgment motions Sept. 4, citing studies that their outputs aren't commercial substitutes. Judge Sidney H. Stein set Oct. 23 for oppositions and Nov. 20 for replies.

Anthropic's prompting guidance for Claude Opus 5.5
Source: Search Engine Journal

  • Anthropic advises developers to retest effort settings rather than reuse Opus 5 configs. Opus 5.5 (launched Sept. 22) defaults to medium effort, one level below Opus 5's high, and thinking can no longer be turned off.
  • In Anthropic's testing, Opus 5.5 at medium effort matches or surpasses Opus 5 at high effort on coding and knowledge work. Adjust effort first when balancing quality against speed and cost, saving xhigh and max for tasks that need them.
  • Chat apps should reconsider "think carefully" system-prompt lines. Removing one made replies start sooner with "no clear decline" in quality, since Opus 5.5 decides how much to think on its own.
  • For agents, pass elapsed time against an advisory time budget, and tag pasted external text with a random ID as one guardrail against prompt injection, not a full defense.

AI in Search / AI Overviews

Google rolls out AI Mode information monitoring globally
Sources: Search Engine Journal · Search Engine Roundtable

  • Google is opening info monitoring in AI Mode to all users worldwide, after limiting it to AI Ultra and Pro subscribers. Robby Stein, VP of Product for Search, announced the rollout on X.
  • You tell AI Mode what to track and Search continuously checks sites, forums, social posts, real-time data sources, and a Shopping Graph of 60 billion-plus products, then surfaces alerts through the Google app.
  • Search will also suggest monitoring tasks mid-search. Stein's examples were all local or shopping topics:
    • New restaurants or pop-ups nearby
    • Local holiday things to do with kids
    • Back-in-stock and price-drop updates
  • Neither Stein's post explains how Search picks the sources behind an update or whether updates carry links, so local businesses and retailers can't yet tell how often monitoring will surface their sites or drive visits.

Google plans to adopt UCP draft spec for AI Mode hotel booking
Source: Search Engine Journal

  • The Universal Commerce Protocol project published a draft Lodging Booking spec to GitHub on Sept. 25, and Google's docs say it will adopt the standard to power AI Mode hotel booking over the coming months.
  • Google began rolling out AI Mode hotel booking for US English users on August 27 with 10 hotel chains and travel platforms. The spec covers real-time pricing, availability, guest details, occupancy, check-in/check-out dates, deposits, split payments, and cancellation policies.
  • Google's FAQ says choosing the booking integration does not influence how properties or rates rank, and the business stays merchant of record, keeping its own customer data and terms.
  • Lifecycle events shared via the API let a business distinguish traffic and payments coming through Google's AI surfaces from its own site. The current onboarding path is a waitlist.
  • The Lodging Technical Council includes Amadeus, Booking.com, Expedia, Google, Hilton, Marriott, and Trip.com. The first milestone covers one room, with multiple rooms flagged as a possible future extension.

Building an AI commercial exposure model from Search Console's Generative AI report
Source: Search Engine Journal

  • Google's Generative AI report only shows visibility (impressions), with no click or query-level data. Mueller notes a directly shown link counts as an impression, while a link needing activation counts only after activation.
  • To move from visibility to exposure, layer the report with GA4 audience data, commercial KPIs, third-party AI Overview prevalence data, and server logs showing AI crawler activity by bot type (training, search, retrieval).
  • The proposed model scores AI Commercial Exposure as (V x S x T x C)^(1/4), combining AI visibility, AI substitution (optional), traffic dependency, and commercial value normalized to 0-100. Drop substitution and use the cube root for the core version.
  • Resilience layers on 4 defensibility factors: branded search, direct audience, returning audience, and content defensibility, to judge how vulnerable each section is if AI intercepts more traffic.
  • The page-level GSC Pages export is capped at 1,000 rows, so large sites are working from a sample and should export by subfolder to classify visibility by content area.

Google tests "Loading..." button on AI Overviews
Source: Search Engine Roundtable

  • Google is testing replacing the "Show more" button under AI Overviews with a "Loading..." button, spotted by Sachin Patel and Khushal Bherwani.
  • Earlier this year Google tested "Enter AI Mode" text for the same button, which did not stick. The current test could not be widely replicated.

Organic Search & Algorithm Updates

Google September 2026 spam update hits hard over the weekend
Source: Search Engine Roundtable

  • The first stage of the September 2026 spam update, announced Sept. 24 at ~9 am PT, drove heavy volatility starting Friday Sept. 25 with more movement Saturday and Sunday. The rollout is expected to take ~2 weeks.
  • What the update targets isn't fully known, though Barry Schwartz expects multiple tactics to be hit rather than one. Recent spam updates appear to hit harder than older ones.
  • Glenn Gabe reported big drops across verticals and countries, including further declines for the AI-generated Grokipedia and examples he tags "Mt. AI" and "Mt. Programmatic."

English-to-Chinese localization benchmark: humans lost 4 of 6 content types
Source: Search Engine Journal

  • A benchmark from EC Innovations and Jademond Digital scored 774 localized outputs across 6 content types, 7 workflow models, and 3 task types, judged blind by native-speaking localizers on accuracy, fluency, and cultural adaptation (disclosure: both firms sell related services).
  • Professional humans won only informational (76.9) and SEO (74.1) content, finishing 7th, 7th, 9th, and 10th of 15 on technical, product UI, UGC, and marketing. On marketing they placed 10th, 22.2 points behind post-edited Qwen.
  • Even where humans won SEO, the lead over the best AI workflow was just 2.8 points, with post-edited Qwen and Doubao at 71.3 roughly at parity. Post-edited Google Translate scored 66.7, beating every raw model.
  • Category averages hide huge within-group spread. Raw Chinese LLMs range up to 22.3 points apart on technical content, so evaluate Qwen vs Doubao vs DeepSeek individually rather than treating "Chinese LLM" as one thing.
  • For Chinese SEO, ICP filing and mainland or Hong Kong hosting can matter more for Baidu visibility than the gap between a 60.7 and 74.1 translation score, since Baidu weights site-level signals over page-level language quality.

Google briefly removes desktop Follow button from search profiles
Source: Search Engine Roundtable

  • Google removed the Follow button from desktop search results knowledge panels while keeping it on mobile, per Damien Andell. The button lets users follow a publisher or creator's search profile.
  • The button returned the next day, though it had been absent on Chrome for several days, suggesting a test or glitch rather than a permanent change.
Published on
geoai searchorganic searchdigest