SEO Recap covering September 15, 2026: Gemini 3.8 Live powers Search Live amid ranking volatility

Built by Stephanie Chung·10 min read·12 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)

Discovery is moving ahead of search, and AI cites earned media
Source: Search Engine Land

  • The funnel has shifted to passive exposure, preference development, confirmation searches, then purchase. Recommendation engines shape preferences before the first query, so many searches now just confirm a decision.
  • A study of over 100 million AI citations found YouTube the second most-cited social platform behind Reddit, and more than two-thirds of consumers rate YouTube somewhat or very trustworthy for purchase decisions.
  • Earned media drove 84% of AI citations, per a Muck Rack review of 25 million ChatGPT, Claude, and Gemini responses across multiple industries.
  • Google query volume hit an all-time high in early 2026 even as search's funnel position changed.
  • 3 ways to drive discovery:
    • Build a video library that answers product and brand questions whether or not you run YouTube ads
    • Keep product feed and data health complete and consistent across owned channels
    • Show up in conversations you don't own, like Reddit threads feeding AI Overviews

How to audit and defend a brand's identity in AI search
Source: Search Engine Journal

  • Audit branded queries across search surfaces (Google, Bing, Brave, DuckDuckGo, YouTube) and AI systems (AI Overviews, AI Mode, Gemini, ChatGPT search, Perplexity, Claude, Brave Ask), running each prompt several times in fresh sessions with memory disabled and per market and language.
  • Model sourcing varies: paid ChatGPT thinking mode takes about 75% of its results from scraped Google rankings and elsewhere, while the free version answers from OpenAI's own index. Brave writes the paragraph first, then searches for the links it shows, so the page you need to fix may not be listed.
  • In a February 2026 evaluation of 6 chatbots on 2,100 news questions, over 70% of inaccuracies came from retrieval failures rather than model reasoning.
  • Slopsquatting shows brand misuse reaching developer infrastructure: a malicious npm package impersonated "eslint-plugin-unused-imports," and an LLM-invented package name spread through 237 GitHub repositories of AI-generated agent skills.
  • Match defense to the problem: fix what you control, correct what you can claim, report genuine abuse, and publish a clearer first-party answer where removal isn't possible, using Organization structured data and preset alerts (registrar notifications, Certificate Transparency, DMARC, Search Console).

AI in Search / AI Overviews

Gemini 3.8 Live now powers Search Live
Sources: Search Engine Land · Search Engine Roundtable

  • Google released Gemini 3.8 Live and 3.8 Live Extended Thinking, now powering real-time conversations in Search Live starting today.
  • Rajan Patel, VP Engineering for Search, said the update brings:
    • More helpful responses with web links to dig deeper
    • Fluid multilingual support, including switching languages mid-conversation
    • More natural, free-flowing interactions
  • To use it, open the Google app, tap the Live icon, and ask verbally. You hear an AI audio response with on-screen links, can open a transcript to type follow-ups, and revisit responses in AI Mode history.
  • Search Live went global in March. Gemini now powers AI Overviews, AI Mode, and Search Live, and results will keep shifting as the models change.

Rethinking SEO for conversational, follow-up search
Source: Search Engine Land

  • The follow-up query is where intent gets revealed. Systems now remember the relationship between turns, so plan for the first question, likely follow-ups, credible proof, and the next useful action rather than a single query and landing page.
  • Google says AI Overviews and AI Mode use query fan-out, running multiple related searches across subtopics, so a page can support part of an answer without mirroring the initial wording.
  • Per a Pew Research Center study, visits with a Google AI summary produced a traditional result click 8% of the time versus 15% without a summary, and links inside summaries were clicked in only 1% of visits.
  • Google reported Lens handles more than 25 billion queries per month (May 2025), pushing visual SEO beyond filenames and alt text toward genuinely useful images, close-ups, labels, and transcripts.
  • Give people a reason to click past the summary: original testing, transparent methodology and dates, calculators and datasets, and clear next actions.

A fight, influence, and generate-demand framework for paid search after AI Overviews
Source: Search Engine Land

  • One bid rule fails once AI answers sit atop the page. Sort queries into 3 buckets: fight (commercial intent worth winning the auction above the answer), influence (education happens in the answer, so aim for share of the answer), and generate demand (zero-click has won, so fund upstream channels).
  • A July SE Ranking study of commercial queries in Google's US AI Mode found ad likelihood rose with cost per click:
    • ~54% at $10 or more
    • 32% between $2 and $10
    • ~24% under $2
  • About 70% of answers with an ad showed 2 advertisers in the block, so you rarely compete alone.
  • Classify terms as a routine: map intent from search terms and Search Console, check AI Overview presence and whether your brand is cited, then sort into fight, influence, or generate demand and revisit as the SERP changes.
  • Watch Absolute Top impression share and impression share lost to rank to see the AI block pushing you down, and introduce AI Max or broad match only where the bucket is known and tracking is clean.

Technical SEO

Google adds ad experience metrics to the CrUX report
Source: Search Engine Journal

  • Google introduced experimental Chrome User Experience Report metrics measuring how advertising affects real-world experience, drawn from opted-in Chrome field data.
  • 4 ad experience metrics:
    • Ad Count: average distinct ad frames visible in the viewport while scrolling
    • Ad Density: average fraction of visible viewport occupied by ad frames
    • Ad Weight CPU: cumulative compute load from ad scripts and assets
    • Ad Weight Network: cumulative network load from ad scripts and assets
  • Data is available via the CrUX API, CrUX History API, and Chrome DevTools' Ad panel, with a BigQuery dataset planned later. SEOs may watch whether it becomes a ranking-related signal.

Cloudflare's Disallow AI Training keeps search crawlers while blocking training
Source: Search Engine Journal

  • Cloudflare's new Disallow AI Training setting, live September 15, adds a no-training preference in robots.txt while still allowing Googlebot, Applebot, and Bingbot to crawl for search. Choosing Block now completely stops those 3 crawlers, including for search.
  • Mixed-use crawlers keep search access only if Cloudflare labels the operator Accountable, which requires meeting or committing to 4 conditions:
    • A robots.txt-style way to opt out of AI training
    • A way to opt out of AI summaries
    • URL-level visibility into pages made available for training plus search-appearance metrics
    • An assurance that opting out of training won't affect traditional search results
  • Apple, Google, and Microsoft qualify as Accountable. Amazon, Anthropic, Meta, and OpenAI are listed as Accountable because they run separate search and training crawlers, but their training crawlers stay blocked.
  • The setting maps to Google-Extended and Applebot-Extended. Bing's robots.txt no-training support isn't ready (its current opt-out is the NOARCHIVE tag), with Microsoft targeting early 2027 and Google adding URL-level transparency for Google-Extended soon.

Meta's Muse agent browses as the user, and only its engineering post says so
Source: Search Engine Journal

  • Meta launched Muse on September 8 (US only, in WhatsApp or the Muse app), an agent that opens a browser, fills forms, and checks out via Stripe's Link, running in a dedicated Muse Secure VM.
  • Across 2 same-day documents, the consumer announcement uses the words risk, attack, attacker, mistake, untrusted, and prompt injection zero times, while the engineering post uses them 39 times and offers up to $300,000 in bounties, including up to $130,000 for prompt injection affecting one user.
  • Only the engineering post discloses that Muse browses as your activity, so a designer's analytics records a human visit and retargeting fires at you while a VM did the browsing.
  • Muse runs 2 tiers: negotiated API connectors for services Meta partners with, and a real Chromium browser imitating the user for everyone else, which defeats bot rules, machine paywalls, and analytics that key on a self-identifying crawler.
  • Watch 3 things going forward: whether agents ever carry a verifiable identity (IETF Web Bot Auth), whether the connector list grows, and whether the next launch ships with one document or 2.

Organic Search & Algorithm Updates

Five ways to automate repetitive SEO work with Claude
Source: Search Engine Land

  • 5 practical uses for Claude in daily SEO work:
    • A specific daily intelligence briefing on industry, competitor, and M&A news
    • A direct connection to Google Analytics for faster data questions
    • Building an hreflang XML sitemap from just links to source sitemaps
    • Translating and localizing content and creating Asana tasks with staged due dates
    • Swapping an outdated, insecure WordPress plugin across sites
  • The plugin swap, done via Claude's Chrome extension in staging, cut a task estimated at an hour per site across 20+ sites to about 5 minutes of hands-on work per site.
  • Guidelines: act as the strategist and let AI do the work, match the tool to the task, and check and correct output because AI still fails and needs spoon-feeding at first.

ChatGPT Ads aren't paid search: 5 questions before shifting budget
Source: Search Engine Journal

  • OpenAI keeps ads separate from answers. Advertising doesn't influence what ChatGPT says, brands can't pay to rank organically, and product recommendations are selected independently from ads.
  • Advertisers get different signals than paid search: OpenAI uses conversation context, but doesn't share users' conversations, and Context Hints (topics, keywords, descriptions) don't work like exact-match keywords or guarantee an appearance.
  • Measurement now supports Pixel and Conversions API tracking, UTMs, and conversion optimization, but upper-funnel tests need assisted conversions, branded-search shifts, or CRM signals rather than last-click or CTR.
  • Answer 5 questions before moving money:
    • What job you expect ChatGPT Ads to do
    • Whether you understand the intent you're buying
    • Where the test budget comes from
    • How you'll measure success, decided before launch
    • How paid ads fit with your organic ChatGPT visibility

Anthropic shares power prompts to improve AI results
Source: Search Engine Journal

  • Anthropic published prompting techniques for Claude Fable 5.1 that can also help users of other models save tokens and get better output.
  • To cut AI writing tics, define the unwanted pattern and give an example of what to do instead. The short version is simply "Please remove all mannered prose."
  • Fable 5.1 uses less formatting than prior models, so refresh anti-formatting rules or let lists and bold appear when they aid clarity.
  • Other prompts address specific behaviors:
    • A low-effort nudge to search and verify unfamiliar or fast-moving names before answering
    • Preferring surgical, targeted file edits over whole-file rewrites
    • A "finish the whole task" prompt for autonomous work so the agent proceeds on reversible actions instead of pausing

Google ranking volatility spikes September 15
Source: Search Engine Roundtable

  • Barry Schwartz reported a big spike in ranking chatter the morning of September 15, suspecting the start of an unconfirmed Google update as tracking tools began heating up.
  • It follows earlier volatility between September 4-12, with site owners reporting sudden traffic and conversion drops, dropped keyword sets, and new pages ranking.
  • UK results drew heavy complaints of page-one spam, irrelevant AI Overviews, and stale news surfacing in Top Stories.
Published on
geoai searchtechnical seoorganic searchdigest