SEO Recap covering September 2, 2026: Gemini 3.8 Flash in AI Mode and a listicle ranking study

Built by Stephanie Chung·8 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)

A new AI protocol won't fix missing organizational knowledge
Source: Search Engine Journal

  • llms.txt, MCP, markdown, and schema are delivery mechanisms, not strategies. Publishing 4 pieces of evidence through another format does not create the fifth criterion a decision needs.
  • Diagnose AI invisibility with Decision Coverage: deconstruct a complex prompt (like "best family-friendly beachfront resort in Cancun") into its criteria and check whether authoritative evidence exists for each.
  • Across more than 100 agentic readiness audits, all flagged the presence or absence of an llms.txt file but none critiqued the depth or quality of the file's contents.
  • Build the canonical knowledge base once and publish everywhere, so supporting a new format becomes a publishing decision rather than another reconstruction project.

AI in Search / AI Overviews

Google adds Gemini 3.8 Flash to AI Mode on launch day
Sources: Search Engine Land · Search Engine Journal · Search Engine Roundtable

  • Google made Gemini 3.8 Flash selectable in AI Mode the same day it released the model, for Google AI Pro and Ultra subscribers worldwide, chosen via the "+" icon in the "Ask anything" bar.
  • Robby Stein called it Google's "most intelligent workhorse model yet," citing gains over 3.7 Flash in software engineering, agentic tasks, and multi-step reasoning. Nick Fox noted it arrived less than 3 weeks after 3.7.
  • It is Google's third Flash release in 6 weeks, and Search is now part of the model launch itself rather than following a day later.
  • Free-tier AI Mode users still get only the default and no model choice. 3.7 Flash is not free even with 3.8 out.
  • For anyone testing AI Mode citations, log the model name with every test. A reading on 3.7 Flash and one on 3.8 Flash came from different models, so gaps between them do not isolate content changes.

Google DeepMind says Gemini is shifting from chatbot to AI agent
Source: Search Engine Journal

  • DeepMind SVP Koray Kavukcuoglu said Google increasingly views Gemini as an agent that takes actions alongside humans, not just a model that answers questions better.
  • Coding and software engineering were the gateway to tool use and agentic workflows, and learnings from work on 3.5 through 3.7 are converging into current models.
  • The underlying methods (deep learning, pre-training, reinforcement learning) have not changed much. What changed is an environment requiring models to infer intent, handle ambiguity, and collaborate.
  • The direction aligns with Sundar Pichai's stance that agentic AI is the future of search, framing all Google products, including Search, around accomplishing tasks.

Technical SEO

Entity SEO as the foundation of modern information retrieval
Source: Search Engine Journal

  • Entity SEO reduces ambiguity about the people, places, organizations, and products tied to a brand so search engines and LLMs build a clear, consistent picture from fragmented web evidence.
  • Work from a brand "source of truth" through 4 stages: own (audit editable assets like About pages, profiles, Google Business Profile, WikiData, structured data and sameAs), describe (page-level entity targeting, headers, internal links, schema), prove (author pages, credentials, E-E-A-T signals), and earn (accurate mentions and citations).
    • Own: controlled, editable assets and consistent NAP
    • Describe: page-level entity targeting and schema
    • Prove: verifiable author identities and credentials
    • Earn: third-party mentions that corroborate salience
  • Mentions now matter more than links because they are harder to game and are better evidence of reputation. Entity corroboration establishes who you are, topical corroboration establishes what you are known for.
  • News is an entity edge case: Top Stories is dominated by entities, and Google's news crawler may crawl an article once, so accuracy on first pass is critical.

Mueller warns against cache-busting sitemap URLs
Source: Search Engine Journal

  • An SEO observed a large site appending a fresh ?v=timestamp to child sitemap URLs daily to force refetches instead of relying on the <lastmod> element.
  • John Mueller called it a bad idea because a constantly changing sitemap URL signals the canonical URL for content should keep changing, which is bad for search engines and tracking.
  • Use clean, stable URLs and accurate <lastmod> values. Cache busting is legitimate for CSS and JavaScript files but not for sitemaps.

Organic Search & Algorithm Updates

What makes a listicle rank well in classic and AI search
Source: Search Engine Land

  • Freshness was the clearest signal: a date from the prior 2 years appeared on 66.6% of top-3 pages versus 57.3% at positions 8-10. An old or unreadable date was tied to 56% lower odds of ranking in the top 3.
  • Counted listicles (explicit item counts like "17 Best Project Management Tools") won 56.8% of direct matchups against editorial roundups and landed in the top 3 55.9% of the time versus 41.4% for roundups.
  • Bigger title numbers ranked higher: counted listicles with 51-100 in the title averaged position 4.11 versus 5.85 for titles with 2-5.
  • More section headings hurt. A one-standard-deviation increase in heading count was associated with 29% lower top-3 odds, so avoid over-structuring.
  • Self-serving listicles still grew (90.4% gained traffic, median 186% among those with starting traffic), but self-promotion itself explained only 4% of growth. The gains came from vendors investing more effort after realizing listicles feed AI answers.

Lily Ray on surviving algorithm updates and why SEO still drives AI search
Source: Search Engine Land

  • Ray traces her approach to the August 2012 Penguin update, when a link-buying strategy wiped out her company's rankings overnight, and to the 2018 Medic and helpful content updates that hit sites she believed were following Google's guidance (~40% of affected sites, in her estimate).
  • She warns GEO shortcut-chasers of precedent for a harsh crackdown, saying she does not want experimenters caught in a "GEO atomic bomb" update.
  • On content, less is more. Ray favors a smaller number of valuable assets updated over time plus presence beyond a brand's own site, rather than scaling generative content.
  • Ray calls abandoning SEO for AI search "completely blatantly false," arguing good SEO is the most sustainable path to AI visibility. Winning means being recommended naturally as the best brand, not publishing self-declared listicles.

Reframing SEO planning around users and surfaces, not keywords
Source: Search Engine Land

  • A Shopify analysis found AI-referred sessions to merchant storefronts grew 197% year over year in Q2 and converted at ~2x the rate of organic visitors in research-heavy categories, while organic search still sent more traffic than all tracked AI platforms combined and grew 12%.
  • Sort your existing keyword list into buckets: still a click (comparison, pricing, implementation), answer-only (definitions), and community-owned (Reddit "has anyone used X"). Stop writing thin articles for the middle bucket and show up as a real person in the last one.
    • Still a click: comparison, pricing, implementation
    • Answer-only: definitions, "what is"
    • Community-owned: Reddit "has anyone used X"
  • Unlinked mentions in trusted communities are an SEO-adjacent job, not brand-marketing leftovers. The thing models struggle to fake is a named person who used the product and will say what broke.
  • Search systems increasingly treat video and transcripts as source material, so record real experts and give assets titles, transcripts, and chapters a model can retrieve.

Google Trends adds category filters to the Explore page
Sources: Search Engine Land · Search Engine Roundtable

  • Google Trends now lets you filter the Explore page by predefined categories such as "All Books & Literature," with or without entering a query.
  • Applying a category filter without a query surfaces top-searched terms across specific categories, regions, and timeframes for topic ideation.
  • Adding a category to a query refines results and excludes overlapping categories, for example filtering "Java" to search marketing rather than the drink or island.

Mueller says the disavow tool does have effects, in its place
Source: Search Engine Roundtable

  • John Mueller said the disavow tool "does have effects" but "has its place & time," responding to a site that dropped after disavowing links bought from Fiverr guest posts.
  • Disavowing a handful of links will not make a site drop out of search. Mueller advised fixing the disavow file to what is actually needed (possibly empty) and looking for other causes.
  • He pointed to a broader spam update and big technical issues (like accidentally blocking the whole site) as more likely culprits, noting recovery from algorithmic changes can take many months.
  • Google has previously said most sites don't need to disavow and that the tool hurts more sites than it helps.

Bing tests dropping site names for URLs in results
Source: Search Engine Roundtable

  • Microsoft is testing removing site names from Bing search results and showing only the site's URL instead.
  • Google ran a similar test about a year ago, so the Bing experiment follows precedent.

Google Analytics bug shows zero traffic for September 1
Source: Search Engine Land

  • Google Analytics displayed no data for Tuesday, September 1, 2026, across many accounts, with GA expert Dana DiTomaso confirming the same pattern across multiple client accounts.
  • The missing data is a processing bug on Google's end, not an actual traffic drop, though Google had not yet confirmed or updated its status page.
Published on
geoai searchtechnical seoorganic searchdigest