
Sign up to save your podcasts
Or


Show Notes: Google's Next Update Won't Touch the Index, and Nobody Will Announce It
Sources and links referenced in this episode:
The sleep framework: A new way for LLMs to retain knowledge in continual learning (Google Research blog, October 27, 2026). Ali Behrouz and Vahab Mirrokni. The post the episode is built on.
https://research.google/blog/the-sleep-framework-a-new-way-for-llms-to-retain-knowledge-in-continual-learning/
Language Models Need Sleep: Learning to Self-Modify and Consolidate Memories (arXiv 2606.03979). The underlying paper, by Behrouz, Hashemi, Javanmard and Mirrokni.
https://arxiv.org/abs/2606.03979
Introducing Nested Learning: A new ML paradigm for continual learning (Google Research blog, November 7, 2025). The Hope architecture and continuum memory system that the sleep framework builds on.
https://research.google/blog/introducing-nested-learning-a-new-ml-paradigm-for-continual-learning/
A Q&A on Google Search updates (Google Search Central, November 2, 2023). Google's own statement that Search is updated thousands of times a year and only notable updates are shared.
https://developers.google.com/search/blog/2023/11/q-and-a-on-search-updates
Our Search Liaison on 25 years of keeping up with search (Google Keyword blog, 2022). Danny Sullivan interview with the 800,000-plus experiments figure.
https://blog.google/products/search/danny-25-years-of-search/
When the Training Data Cutoff Becomes a Ranking Factor (Duane Forrester Decodes, March 2026). The parametric-versus-retrieval piece this episode extends.
https://duaneforresterdecodes.substack.com/p/when-the-training-data-cutoff-becomes
Accurate Facts Can Still Produce a False AI Answer, and Courts Are Starting to Hold the Platform Responsible (Duane Forrester Decodes, October 2026). The entity-conflation piece referenced on consolidation risk.
https://duaneforresterdecodes.substack.com/p/accurate-facts-can-still-produce
CitationIQ. AI visibility measurement.
https://citationiq.com
My latest book: The Machine Layer.
https://www.amazon.com/Machine-Layer-Visible-Trusted-Search/dp/B0G2WZKM59/ref=sr_1_1
Show Notes: The Traffic Pool Is Shrinking, and Which Channel Fills It Matters Less Than We Think
References in the order they come up in the episode.
• Duane Forrester consulting: Work with me directly on traffic, AI visibility, and agent readiness.
• CitationIQ: Tracks how your brand appears inside AI answers.
• Marie Haynes, GEO is a distraction: Twelve months of GA4 data across 100 sites comparing ChatGPT and LLM referrals with Google organic.
• The SEO Works, AI referral traffic benchmark: 95 sites comparing traffic and conversions from organic search and AI assistants.
• Kevin Indig, Does AI drive more direct traffic?: 632 tests across 12 sites looking for AI visibility to show up in branded search and direct traffic.
• Kevin Indig, Organic traffic is down, but it's not all AI: Similarweb data for 10,000 domains each in the US and France, covering organic, paid, and AI referral trends.
• PPC Land, Google ad clicks gain 8% in Q3: Paid click and cost-per-click growth from Alphabet's Q3 2026 10-Q.
• Kevin Indig, What channels grow when organic search shrinks?: The follow-up showing email as the steadiest-growing channel.
• Shi, Zhu and Gu, Answering Without Referring: Bocconi University preprint on ChatGPT outbound clicks and search displacement.
• Search Engine Journal, Google search traffic to the open web: Similarweb data on clicks reaching the open web per 1,000 US searches.
• Google Search Queries Hit a Five-Year High, and Search Revenue Didn't Follow: Last week's piece on Alphabet's Q3 results.
• Google, AI in Search is driving more queries and higher quality clicks: Liz Reid's 2025 post describing organic click volume as relatively stable.
• Getting Indexed No Longer Means Getting Visited: My earlier piece on what being indexed earns once answers replace links.
• Alphabet Q3 2026 earnings call transcript: Source for Google's remarks on agents, UCP, and advertising in agentic experiences.
• Wadi and Ma, Does Rank Still Matter?: McGill study of AI agents choosing among randomized hotel listings.
• Wadi and Ma, Whom Do AI Agents Work For?: Companion study on how an agent's assigned principal changes its choice of sponsored listings.
• Storyboard18, OpenAI scales back ChatGPT shopping ambitions: Coverage of OpenAI pulling back from Instant Checkout.
• PPC Land, Universal Cart reaches US Search and Gemini: Universal Cart launch details and the May scan of public UCP profiles.
• Chrome for Developers, WebMCP: Google's guidance on exposing site tools to AI agents through WebMCP.
• Cloudflare, new AI traffic options: The Search, Agent, and Training controls and the September 15 defaults.
• The Machine Layer: My book on how machines decide what to surface and act on.
Google Search Queries Hit a Five-Year High, and Search Revenue Didn't Follow
Links and references from this episode:
Alphabet earnings miss reflects first real sign that AI is taking a bite out of search (CNBC) - CNBC's report on the search miss and analyst reaction
Alphabet Q3 2026 earnings coverage (CNBC) - headline numbers, free cash flow, and estimates
Alphabet Q3 2026 earnings release (PDF) - Alphabet's official release, including the constant-currency tables
Alphabet Q2 2026 earnings release (SEC) - Alphabet's Q2 results, used for the currency comparison
Alphabet Q3 2026 earnings call - Alphabet investor relations page for the call
Alphabet Q3 2026 earnings call transcript (Motley Fool) - full transcript, including Schindler's comments on AI Mode monetization and click-through rates
Alphabet stock suffers record fall (R&D World) - coverage of the record one-day stock drop
The SERP Stops Being the Default Between 2028 and 2030 (Duane Forrester Decodes) - my forecast on when the results page stops being Google's default
Cohorts, Clusters, and the Coming AI Ad System (Duane Forrester Decodes) - my August 2025 piece on ad systems that build the ad at query time
The Machine Layer (Amazon) - my book on staying visible and trusted in AI-driven search
Show Notes: Accurate Facts Can Still Produce a False AI Answer, and Courts Are Starting to Hold the Platform Responsible
Links referenced in this episode:
• Regional Court of Aachen order, September 24, 2026: the court's original order (in German) barring Google from spreading AI Mode answers that attached another man's criminal case to a doctor.
• Munich ruling, full decision in English translation: Transparency Coalition's reprint of the Regional Court of Munich decision holding Google liable for AI Overviews that falsely tied two publishers to fraud.
• Google pushes back against the Munich ruling: Canadian Lawyer's coverage of Google's statement that it would appeal, including its "specific and narrow errors" comment.
• Two German courts now hold Google liable for what its AI overviews say: The Decoder's report on the Aachen order, and the source for the Munich appeal ending in a settlement.
• Berlin court refuses injunction over perfume "dupes": Prof. Thomas Hoeren's analysis on the IPKat blog of the Berlin trademark ruling that treated AI Overviews as a search result format.
• Microsoft's Copilot falsely labels a court reporter a criminal: The Register's 2024 report on a journalist accused of the crimes he covered.
• Google users are less likely to click on links when an AI summary appears: Pew Research Center's study behind the 1%, 8%, and 15% click figures.
• Section 230 and generative AI platforms: Crowell's analysis of why Section 230 coverage for AI-generated text is still an open question in US courts.
• Search Engines Were Built to Send Out Traffic: my earlier piece on responsibility moving from the results page to the answer.
• The Machine Layer: my book on how AI systems decide what to trust and what that means for businesses.
Show Notes: Getting Indexed No Longer Means Getting Visited
The opt-out setting and the rules behind it
Google Search Central Help: Search generative AI control: Google's documentation for the Search Console setting that excludes a site from AI Overviews, AI Mode, and generative features in Discover, including what it does not affect.
Search Engine Journal: Google Search Console AI reports rolled out worldwide: Coverage of the August 31, 2026 global rollout of the opt-out control and the Generative AI performance report.
GOV.UK: CMA secures fairer deal for publishers and improves Google search services in UK: The UK Competition and Markets Authority's June 2026 conduct requirement covering the opt-out, attribution, and no-retaliation rules.
Research on clicks and AI answers
Pew Research Center: Google users are less likely to click on links when an AI summary appears: Browsing data from 900 US adults showing an 8% click rate on traditional results with an AI summary versus 15% without.
Wang, Gleason, Bart, Wilson, and Metaxa: AI in Search Reduces Publisher Referrals Without Improving User Experience: Preregistered field experiment with 1,100 Google users; forced AI Mode cut external click-through by 18.8 percentage points (preprint, not yet peer reviewed).
Is the remaining traffic better?
Microsoft Clarity: AI traffic converts at 3x the rate of other channels: Study of more than 1,200 publisher and news sites on AI referral volume and sign-up conversion.
Kaiser and Schulze, Marketing Science: ChatGPT Referrals to E-Commerce Websites: Peer-reviewed analysis of 973 e-commerce sites over twelve months comparing ChatGPT referrals with traditional channels.
The travel agent precedent
U.S. Bureau of Labor Statistics: Travel Agents, Occupational Outlook Handbook: 2025 employment of 61,500, a flat 2025 to 2035 projection, and AI tools named as a factor limiting demand.
Perk: How online booking has changed the travel agent landscape: Analysis of BLS data showing roughly a 70% drop in travel agent employment between 2000 and 2021.
Travel Weekly: How many travel agents in U.S.? Depends on whom you're asking: ASTA's dispute with BLS counts and its use of the Census figure.
Host Agency Reviews: How much do travel agents make?: Industry income research on independent advisors and the earnings edge for luxury and adventure specialists.
rom Duane
The SERP Stops Being the Default Between 2028 and 2030, and the Place It Moves To Already Exists: Last week's forecast for when the ranked results page stops being Google's default first view.
The Machine Layer on Amazon: Duane's book on how AI search and answer systems work underneath.
Show Notes: The SERP Stops Being the Default Between 2028 and 2030, and the Place It Moves To Already Exists
Google's own statements
Google: A new era for AI Search (I/O 2026): Liz Reid's I/O announcement; Gemini 3.5 Flash as the default model in AI Mode, the rebuilt search box, AI Mode passing one billion monthly users.
Search Engine Roundtable: Google Chrome Won't Make AI Mode the Default Search Experience: Rajan Patel calling the Chrome Canary flag an error and saying Google isn't planning to make AI Mode the Chrome default.
Search Engine Journal: Google Rolls Out New "Web" Filter for Search Results: The May 2024 launch of the links-only Web filter, announced by Danny Sullivan.
Google: Q2 2026 earnings call, remarks from our CEO: Sundar Pichai on AI Mode costs hitting their lowest level since launch and AI Overviews and AI Mode merging into one experience.
Alphabet Q2 2026 earnings call transcript: Full transcript, including the 1% decline in Network advertising revenue.
Google: Everything you need to know from 2026 (Google Marketing Live recap): Google's own GML recap covering ads inside AI Mode and the "reinventing what an ad is" statement.
Google Search Central: Mobile-first indexing has landed: Google declaring mobile-first indexing complete in October 2023, the seven-year precedent used for the forecast timeline.
Adoption and behavior data
Pew Research Center: Google users are less likely to click on links when an AI summary appears: Browsing study of 900 US adults; 18% of searches produced an AI summary, with clicks on traditional results falling from 15% to 8% of visits.
AI News: Similarweb's 2026 Generative AI Landscape on AI Overviews: AI Overviews on 43% of US searches, up from 15% a year earlier; AI Mode visit growth; the 5.4% increase in query length.
Similarweb: Zero-Click Marketing and the 2026 data: AI Mode at 0.34% of Google searches from January to April 2026, and AI Mode referral rates of 1.6% to 2.5% against 17% to 19% for traditional search.
OpenAI: Scaling AI for everyone: OpenAI's February 2026 announcement of 900 million weekly ChatGPT users.
PPC Land: Google faces bipartisan brief urging tougher remedy: Coverage of the amicus brief in the DOJ appeal, including its point that no industry-standard measurement of AI Overview prevalence exists.
Regulation and litigation
Advanced Television: CMA imposes Google publisher AI opt-out: The UK Competition and Markets Authority's June 3, 2026 conduct requirement on AI Overviews and AI Mode opt-outs.
SCL: European Commission investigates Google for using online content for AI purposes: The December 2025 EU investigation into AI Overviews, AI Mode, and publisher content.
MLex: US DOJ, states argue appeal of Google Search ruling, ask for harsher remedies: The July 2026 cross-appeal brief, including the "competitive window may close" warning.
Related reading from Duane Forrester Decodes
Cohorts, Clusters, and the Coming AI Ad System: My August 2025 piece describing query-time, auto-generated ads, the system now arriving inside AI Mode.
Three Predictions for 2027, and Why You Won't Be Able to Check Them: The 2027 predictions piece referenced in this episode.
Google Now Has the Math to Rank Without an Index, and the Results Page Does Not Survive It: Last week's piece on the DeepMind research behind ranking without an index.
The Machine Layer: My book on how these systems work underneath search and AI answers.
Show Notes: Google Now Has the Math to Rank Without an Index, and the Results Page Does Not Survive It
References mentioned in this episode, in the order they come up.
Google Now Has the Math to Rank Without an Index, and the Results Page Does Not Survive It. This week's article on Duane Forrester Decodes. Link to be added on publication.
Google DeepMind Develops New AI Search Ranking Model. Roger Montti, Search Engine Journal. The piece this episode builds on; the clearest walkthrough of the dual encoder and cross encoder mechanics.
Autoregressive Ranking: Bridging the Gap Between Dual and Cross Encoders. Google DeepMind, UMass Amherst, UT Austin, January 2026. The paper: the linear-growth proof for dual encoders and the SToICaL training loss.
Rethinking Search: Making Domain Experts out of Dilettantes. Metzler, Tay, Bahri, Najork, Google Research, 2021. The proposal that started the arc; carries its own disclaimer that it is not a product roadmap.
Transformer Memory as a Differentiable Search Index. Tay et al., Google Research, 2022. The DSI prototype: a single Transformer mapping queries directly to document identifiers.
DSI++: Updating Transformer Memory with New Documents. Mehta et al., Google, EMNLP 2023. The freshness problem: continually indexing new documents causes forgetting of old ones, plus the mitigations.
WordNet. Princeton University. The lexical database used as one of the two evaluation sets; its noun hierarchy supplies the ground-truth rankings.
Alphabet Second Quarter 2026 Results. SEC 8-K, Exhibit 99.1. Source for the $63.27 billion Google Search and other figure and the 17 percent growth rate.
Google Marketing Live 2026. Google, May 20, 2026. The announcement of Gemini-built ad formats for AI Search, tailored to the individual query.
Cohorts, Clusters, and the Coming AI Ad System. Duane Forrester Decodes, August 2025. Where Intent Vector Bidding was coined and the query-time ad system described.
Mechanism Design for Large Language Models. Dütting et al., Google Research. Best paper, The Web Conference 2024. The token auction: advertisers bidding on generated text token by token.
The Machine Layer. Duane's book on how content earns its place inside AI retrieval systems.
Show notes: AI Changed the Buyer in Months. Businesses Change in Years.
References mentioned in this episode:
LLMs Are Time Machines That Don't Tell You How Far You Went - Last week's piece on the confidently underinformed buyer, path metadata, and why citation is presence, not traffic.
Pew Research Center: Americans and AI 2026 - Survey of 5,119 US adults (February 2026 fieldwork). Chatbot adoption, daily use, and information search as the top use.
Why Part of Your AI Authority Takes Years, Not Campaigns. And Why It Comes From Other People. - August piece on how category authority accrues in models and where it comes from.
When AI Has Nothing On Your Company, It Describes Someone Else. - Why a gap in what models know about you gets filled by a substitute, and why content audits miss it.
The GEO Tooling and Data Market Has a Trust Problem - The tooling-first instinct and why the measurement market needs scrutiny before purchase.
Google Analytics: What's New (AI Assistant channel, May 13, 2026) - Google's own description of the GA4 AI Assistant channel and what it tracks.
Introducing AI Performance in Bing Webmaster Tools Public Preview - Microsoft's February 2026 announcement of citation reporting across Copilot and Bing AI answers.
duaneforrester.com - Consulting on GEO and SEO.
The Machine Layer - The book, on Amazon.
Show Notes: LLMs Are Time Machines. The Problem Is Where They Drop You.
Sources and links referenced in this episode:
Melumad & Yun, PNAS Nexus, October 2025. Seven experiments, 10,462 participants, on how learning from AI syntheses compares with learning through web search.
https://academic.oup.com/pnasnexus/article/4/10/pgaf316/8303888
Pew Research Center, July 2025. Browsing-panel study of 900 US adults across 68,879 Google searches, measuring click behavior when an AI summary is present.
https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/
Fisher, Goddu & Keil, Journal of Experimental Psychology: General, 2015. Nine experiments on how internet search inflates self-assessed knowledge.
https://pubmed.ncbi.nlm.nih.gov/25822461/
Lee et al., CHI 2025. Microsoft Research and Carnegie Mellon survey of 319 knowledge workers on AI confidence and critical thinking.
https://dl.acm.org/doi/full/10.1145/3706598.3713778
Lewandowski, Information Research, 2026. Exploratory study on information regret and answer engines.
https://doi.org/10.47989/ir31ISIC65159
When AI Has Nothing On Your Company, It Describes Someone Else. The companion piece on why this failure is invisible to content audits.
https://duaneforresterdecodes.substack.com/p/when-ai-has-nothing-on-your-company
The Machine Layer.
https://www.amazon.com/Machine-Layer-Visible-Trusted-Search/dp/B0G2WZKM59/ref=sr_1_1
Everyone Is Quoting the Same Number and Getting Different Answers
References from this episode:
Cloudflare, Crawl-to-Refer Ratios on Cloudflare Radar - the primary source for everything in this episode. The full formula, the June 19 to 26 2025 sample week, the user-agent aggregation, the Google ASN exclusion, and the native-app Referer caveat that says the calculations may overstate the ratios by an amount that is unclear.
Cloudflare, Control Content Use for AI Training - the separate June 2025 figures I mentioned, where Anthropic sits at 73,000 to one against the 70,900 in the post above. Same publisher, same month, different window.
Cloudflare, From Googlebot to GPTBot: Who's Crawling and Who's Clicking - background on the crawl-to-click gap and how the ratios have moved since launch.
Cloudflare Radar AI Insights - the live dashboard. Worth pulling your own figure here rather than quoting somebody else's, and worth noting which window you selected when you do.
Google, Search updates from I/O 2026 - the source for the AI Mode scale claims, including the growth rate published without a base.
The Machine Layer - my book, if you want to go further into how these systems get measured and mismeasured. Available here.
The written version is on the Substack. If you have hit a metric in this space that fell apart once you asked what was in the denominator, leave a comment there or reach out directly. Specific examples are more useful than the general argument.
From the publisher's feed