Never Let Dad Hold the Bag

Google Plays Dumb; How People Search on ChatGPT; AI Citation Stability; HubSpot Lost 80% Traffic and Grew Revenue; Google Spam Crackdown; and Much More!

Author

Yoyao
October 31, 2026 • Est. Learning Time: 17 minutes

FIRST …

Happy Halloween to everyone celebrating tonight.

Don’t be like me and eat all the candy your kids picked up 😅 

Let’s get into all of this week’s candy below.

Floyi just got a sweet update before Halloween 🍬 🎃 

Make sure your article is ready to be quoted by LLMs.
Floyi’s Page Quality Scorecard now checks the first sentence of every article section and shows you exactly which openings to fix.

Floyi v2.18.0 also adds search demand for every topic, multiple GSC accounts, and AI Review for Topical Maps, which flags misplaced and near-duplicate topics.

SEO + GEO

Derek Perkins reports that Google has been returning SERPs that effectively change the query itself: the result page title still shows the full keyword, but the organic results answer just one word of it, a fault Nozzle traced across 248 million SERPs with data through October 21, 2026. It's the companion study to Nozzle's earlier Poisoned Google SERPs research, which caught the opposite fault where the query stays intact and the keyword's usual sites vanish.

  • Since September 16, page 1 or page 2 ignored part of the query on 13.9% of SERPs judged on both pages, against a 0.06% baseline, and hit 39.9% at the September 28 peak.

  • Page 2 broke first, with a fifth to two fifths of page 2s a day ignoring part of the query from September 17 to 29, and 93.6% of those page-2 faults ran under a page 1 that still answered the query sent.

  • SerpApi first reported the fault publicly on May 4, and Nozzle's weekly counts show March bursts that peaked at 0.3% of SERPs a week before the week of September 28 hit 25.0%, seventy times the July peak of 0.35%.

  • Of 497,362 affected page-1 SERPs since September 16, 66.2% dropped exactly one word of the query and 51.5% carried exactly one word on most of their results.

  • The organic results kept only 18.5% of the keyword's usual sites against a 90.0% baseline, while Google's own elements mostly held, led by shopping keeping 88.6% of its usual items against a 91.7% baseline and the image pack 81.9% against 90.6%.

  • The suggestion boxes followed the one word: "People also search for" went from 69.5% full-query suggestions on normal SERPs to 18.7%, and "People also ask" went from 49.5% full-query questions to 29.5%.

  • Where a tracked site was one of the keyword's usual results, it fell out of the top 10 on 69.8% of affected pages against a 2.0% baseline, then the keyword's next SERP within three days answered the query again 85.8% of the time.

  • More than half of these pages also fail the poisoned-SERP test at 54.0%, sharing its fingerprints: pagination stops by page 4 on 43.0% of them against 0.2% normally, and one domain fills three spots or more on 32.2% against 10.0%.

  • Belgium ran highest of all target countries at 4.8% with Japan lowest at 0.6%, and events and entertainment was the worst-hit US industry at 3.0% since September 16.

Google outlines new ways they’re making sources and web pages easier to reach from AI Mode and AI Overviews. Google says Preferred Sources usage has grown more than 7x since the start of 2026.

  • Inline links move higher: Google is making links to sources appear higher in AI responses more often.

  • Official pages get a dedicated test: A new section at the top of some responses will highlight official sites and explain why each link may help.

  • News carousels expand: Google is rolling out timely-article carousels to more locations and languages, with Preferred Sources highlighted.

  • Related links get more visual weight: Links at the end of some AI responses will gain thumbnails and a more prominent design.

  • Creator profiles become easier to reach: Google has started testing clickable creator and publisher names on video results in AI Mode.

Resoneo's reconstruction of Google's KScorer schemas shows how Google selects a query interpretation before deciding which answer or search result features to show. The report documents 56 signals across 12 families, including clicks, the top results, entities, language, and user context.

  • The top results help select intent

  • Google uses interaction data to predict the score human quality raters would give an answer card. Confidence in that click signal is 50% at 20 impressions and about 83% at 100.

  • Meaning is checked against page content

  • Local intent has its own evidence, include distance to nearby entities, local search and click patterns, and the average quality of the top three local results.

  • Eleven signals vote before results arrive, 41 help score interpretations as the results are assembled, and six signals assess whether an answer card is useful.

JC Chouinard built an interactive dashboard that makes OpenAI's Signals v3.0 data browsable in a browser, covering 27 million sampled consumer messages from July 2024 to September 2026. Pick a prompt type, then filter every chart by country, audience, or period, no download needed.

Mateusz Makosiewicz shares a study that followed about 37,500 AI citations for 20 days across Google AI Mode, ChatGPT, Google AI Overviews, Gemini, and Copilot, starting from day-one citations and tracking whether each one survived the next day and how long it lasted. Citations churn hard and mostly come back.

  • Between 50% and 83% of citations disappeared by the next day, from 50% on AI Overviews to 59% on ChatGPT and 83% on Copilot.

  • About 75% of lost citations returned within the study window, half of them inside three days, and on AI Overviews and ChatGPT roughly 40% were back after a single missed day.

  • Only 20% to 27% of day-one citations disappeared and never returned during the study.

  • Citation age predicted returns: 96-99% of citations that had lasted 8 days or more came back after going missing, while one-day citations never returned 34-41% of the time.

  • Pages ranking in Google for the same question were about 7x more likely to hold a citation for a week or more on Google's AI tools (29% vs 4% on AI Overviews), while Domain Rating and backlinks barely mattered, with long- and short-lived citations coming from similarly strong sites (DR 64-78).

  • Median monthly organic traffic of cited AI Overview pages climbed from 8 for one-day citations to 88 for citations lasting 15 days or more.

  • When a question named a brand, that brand's own site stayed cited for all 20 days in about 30% of cases, 4-7x more often than the average cited site, and it held its citation from one day to the next about 75% of the time.

  • The "Best X" category was the least stable, with 64-76% of cited pages on ChatGPT and AI Mode changing from one day to the next.

Mitul Gandhi publishes a refreshed seoClarity estimate of ChatGPT's search-equivalent activity, arriving at about 7.5% of Google's modeled daily search-query volume, or 1.24 billion search-like messages every day. Across July to September 2026, Seeking information carried 20.13% of sampled messages and Practical guidance 29.43%, and seoClarity counts a quarter of Practical guidance as search-equivalent, for a 27.49% combined share.

Scaled against an estimated 4.5 billion daily ChatGPT messages, that produces the headline 1.24 billion, or about 952 million search-like tasks once you assume 1.3 messages per task. Google supplies the denominator with a disclosed five trillion annual searches plus an assumed 20% growth, roughly 16.44 billion queries a day, and the published sensitivity table shows the range: counting half of Practical guidance lifts the estimate to 1.57 billion, while counting none lands at 907 million.

Shadrack Wanjohi breaks down 83,835 YouTube citations captured across 182,631 AI answers and shows YouTube is mostly a Google-side opportunity where small channels with buying-question titles do the winning. Foundation's B2B Citation Index ran 5,000 bottom-of-funnel software prompts in 13 weekly runs from June 29 to September 21, 2026.

  • Google AI Mode and AI Overviews generated 98.2% of all YouTube citations, with AI Mode at 51.9% and AI Overviews at 46.3%, against 1.8% from ChatGPT.

  • Google's AI features cited YouTube in 36.0% of answers against 2.1% on ChatGPT, about 17 times as often.

  • Google cited YouTube in 46.5% of answers to "best of" prompts and 19.1% of pricing prompts.

  • Citation rates ranged from 63.2% for project management tools down to 13.2% for ABM platforms across the 80 software categories, with operations software at 51.1% and product tools at 25.5%.

  • "Best" or "top" appeared in 59% of top-100 cited titles against 10% of videos cited a single time, and a year in the title carried a 57.3% to 19.5% gap.

  • 48.4% of Google's YouTube citations carried a timestamp, 71% of those pointing inside the first five minutes at a median of 2:28, while ChatGPT produced zero timestamped video citations.

  • 5-to-15-minute videos made up 72% of the top-150 group against 43.4% of videos cited once, and 68.7% of the top-150 were less than a year old.

  • Median views of the top-150 videos ran 1,207 against 1,719 for videos cited a single time, and 42% of the top-150 came from channels under 10,000 subscribers.

  • The top 100 videos drew only 12.6% of appearances with 52.7% of citations landing on videos cited once, while the 3,016 videos appearing in at least five of the study's weekly runs weighted 66.4% of all appearances.

Barry Schwartz reports on an opt-out case study from Cyrus Shepard that measured a 0.9% click-through rate on impressions inside Google AI Overviews and AI Mode. Cyrus used Search's generative AI controls to block one site from both surfaces, which took AI impressions there, about 8% of the site's total, to zero immediately. Ordinary organic clicks also fell during the block in a pattern the study calls the Search SERP Halo Effect, and pages that had shown in AI Overviews regained their average position once the block came off.

Roger Montti reports that Google's Q3 earnings call showed Search revenue growing 12% year over year to $63B, with the strongest year-over-year query growth in five years tied to AI Mode adding queries. Ads inside AI Mode queries earn less per query than traditional search ads are earning. Ads are scaling across both AI Overviews and AI Mode, with international expansion and multiturn ads running in the US.

Sundar Pichai framed the future as task completion, with users finishing bookings, forms, appointments, and purchases inside Google and agents working as the unifying layer across Search, Gemini, Chrome, Android, and Workspace. He also concludes that a task-based future makes no promise about sending clicks back to the open web.

John Mueller told David Iwanow that AI-era shortcuts may rank for a while but the systems eventually recognize and discount them.

  • Listicles aren't inherently spam: Mueller said a single listicle can be legitimate content, but relying on thousands of them as an entire strategy is risky, because tactics that start winning can get recognized and discounted as spam.

  • The catch is a matter of time: Google expects tricks to be used "in a particularly bad way and then at some point the systems catch," and all the effort gets thrown out.

  • Mass-produced AI content repeats history: Mueller compares it to keyword stuffing and white text on a white background, saying systems eventually call out patterns that are "hurting actual businesses."

  • Crawl Stats is underused: he recommends checking the Search Console report for unusual crawl activity or rising crawl times, since catching problems early beats recovering from a big cut in crawling.

  • Long redirect chains aren't automatically a problem: Google learns the current destination and crawls it directly, so investigate chains when diagnosing a real issue instead of counting hops.

Brooke Osmundson recaps Google's Ads Decoded episode from October 28, where Google Ads' Ginny Marvin and Search's Shivani Mohan described people researching earlier, asking longer questions, and leaning on human experience.

  • Research starts earlier: Marvin reported that people use conversational search for brainstorming well before they reach a purchase decision.

  • AI Mode queries run 3x longer: Google's US AI Mode report backs the length figure and puts multimodal use at more than one in six US AI Mode searches.

  • Users still want human perspectives: M ohan wrote that users "appreciate the streamlined clarity of AI summaries, yet they actively crave the rich detail, lived experience, and earned authority of the open web," and she pointed to forums, podcasts, and video reviews as the places they find it.

  • Check one, message the need: review customer questions and sales conversations for problems people describe before they know a product name, then test copy against that need.

  • Check two, make comparing easy: landing pages should answer price, compatibility, setup, and limitation questions before a quote request or purchase.

  • Check three, connect the proof: link reviews, demonstrations, and customer stories that address a known concern, then give visitors a defined next step.

  • Check four, measure the delayed sale: review whether comparison-page visitors return and convert over the normal purchase window, since impressions and click-through rate alone won't show it.

Danny Goodwin reports on a Findrix study that tested 18,400 buyer questions across 35 US markets and collected 55,198 answers from ChatGPT, Gemini, and Google AI Mode. Google AI Mode linked only to its own pages in 46.5% of product answers, and 57% of the links in its product answers led to Google Shopping. The Google-only pattern ran strongest in fashion and beauty, led by swimwear at 60.4% of answers, while local services questions put just 5.1% of links on Google pages and software questions 1.1%.

Greg Jarboe interviews HubSpot CMO Kipp Bodnar about a blog that lost most of its organic traffic while revenue kept climbing. Bodnar's answers in writing frame what HubSpot measures now and what it no longer defends.

  • HubSpot's blog once topped 20 million organic visits a month, lost roughly 80% of that traffic, and saw nearly 5 million visits vanish in 30 days when Google rolled out AI Overviews.

  • Visitors arriving from AI answers convert at about 3x the rate of traditional search visitors, and deal cycles run about half as long.

  • Revenue grew nearly 20% in 2025 on the back of channels HubSpot built between 2021 and 2023: The Hustle, a podcast network, a creator program, and a YouTube network.

  • About 90% of HubSpot's content leads now come from those newer channels, with the blog down to about 10%, and the company pruned more than 3,000 non-converting pages.

  • HubSpot rebuilt its chat to answer real questions, and by Q4 2025 it was generating more than 10,000 sales meetings a quarter.

  • An AI-written, per-recipient email program improved conversions 82% against a segment-based baseline, and AI lifts the broader email program by an average of 50%.

  • Qualified leads from AI answers grew 1,850% between Q1 2025 and Q1 2026, and Bodnar concedes the base was small and the attribution correlational.

  • Buyers now write prompts of 40 to 100 words against two to four words in traditional search, a shift Bodnar says content has to answer in plain, chunked sections.

Kevin Indig analyzes three years of monthly Similarweb channel data for 10,000 US and 10,000 French domains and finds the aggregate organic decline is real but uneven, with ads and algorithm updates eating clicks that often get blamed entirely on AI. The dataset covers September 2023 through August 2026 across ten channels.

  • Organic visits fell 13.2% in the US, from 19.9 billion to 17.3 billion, and 24.4% in France, from 4 billion to 3 billion, between August 2024 and August 2026.

  • Organic's share of total US traffic slipped from 20.5% to 18.3% since September 2023, and France's fell from 27.3% to 23.1%.

  • Losses concentrate hard: 62.5% of US sites and 66.6% of French sites declined, and about 1% of domains account for 40.2% and 48.5% of gross lost visits.

  • Median small sites are roughly flat, up 2.7% in the US and down 1.9% in France, while large sites shed 13.5% and 27.4% on a total-visits basis.

  • Publishers lost 41.5% of organic traffic in the US and 42.5% in France, and stripping publishers out cuts the US decline to just 7.8%.

  • Paid search traffic grew 21% in the US against organic's 13.2% loss and 17.3% in France against 24.4%, with paid gains covering about 36.6% of US organic losses and 6.7% in France in absolute terms.

  • Ad load is part of the story: Google's worldwide Sponsored results grouping launched Oct. 13, 2025, the average organic result now sits below the desktop fold with ad blocks roughly double in size on a 46,000-head-term sample, and Google's Q2 2026 earnings show paid clicks up 13% with CPC up 3%.

  • AI referrals are the fastest-growing channel yet tiny, at 2.85% of US and 2.44% of French organic-search traffic, about one in 200 visits, with technology, civic services, and careers overrepresented.

Roger Montti reports that Google expanded its user-generated-spam guidelines to put platforms, tools, and services that host user content on notice that they may face search penalties for what their users publish. The companion sections tell platform owners how to limit exposure: noindex or block pages built for direct sharing, collaboration, or utility through the robots meta tag, the X-Robots-Tag header, or robots.txt. He reads the changes as preparation for enforcement against publishing platforms that get abused, though the guidelines stop short of promising it.

Danny Goodwin reports on an SEOjobs.com survey of 279 SEO and search professionals that found 43.7% more worried about AI than a year ago, while 32.6% felt no change and 23.7% felt less concerned. Career fears cluster around structure rather than replacement: shrinking teams led at 45.5%, keeping up with AI followed at 44.8%, and tougher competition followed at 44.1%, with role elimination itself at just 24.4%. Concern concentrates at the edges, with job seekers the most alarmed at 61.1% very or extremely concerned and six-to-ten-year veterans at 42%.

Simone De Palma separates pop-up implementation choices from interstitial panic: overlays are not banned, and intrusive implementations don't trigger manual actions, but they can still erode search performance over time. Her framework runs every pop-up through fixed questions, then rules each use case by verdict.

Brent Csutoras recaps Steve Huffman's Mod World AMA of Oct. 24, where Reddit's CEO pointed the platform's next chapter at deeper community participation instead of more search and AI referral volume. The home feed will work to push users into subreddits, subreddits regain more custom appearance control, and a follow-versus-join split would let people observe first, comment later, and earn vote rights through contributions other members appreciate. On integrity, he counted spam takedowns dropping from 45 minutes to 2 over the past year and bots below 2% of activity, while conceding that some flagged accounts are humans pasting ChatGPT output and computer-using agents will blur detection further. AutoMod took the criticism, with 30% of posts from new users banned automatically and the vast majority of those posts fine.

Kelsey Duncombe-Smith announces Discord's biggest opening of community content yet, with discussions from its Discoverable Servers becoming findable through ordinary web search. Server Web Pages go live for every Discoverable Server on Nov. 16, 2026, with automated Channel Summaries the same day, and individual Channel Messages join web results from mid-December in members' own words, posted without usernames or avatars. Admins pick which public channels take part while members hold one switch that decides whether their own messages appear, and channels feeding the web get labels so everyone knows before typing. The defaults lean open, with public channels participating unless admins opt them out and the member setting enabled outside the EU, UK, Canada, Brazil, Australia, and South Korea, where legal restrictions start it off, and teen accounts are excluded entirely.

Google's Gary Illyes, interviewed by Aneta Samkoff on the Odys Podcast, said bot impersonation is now a bigger problem because of AI, since companies hunting content to train their models keep posing as Googlebot to slip past defenses that would otherwise block them. Illyes says anyone can fake a crawler's user-agent in about 30 seconds, and his honeypot tests logged requests claiming to be Google from IP addresses unrelated to Google, so site owners should verify the IP and reverse DNS records rather than trusting the user-agent string. He treats robots.txt as a suggestion rather than a security mechanism and pushes network-level defenses like a web application firewall, with tokens such as Google-Extended separating what AI work gets to use a site's content.

SEO + GEO Ripples

  • Google Gemini Stopped Attaching UTM Parameters To Cited URLs: Barry Schwartz reports that Google stopped appending utm_source=gemini tags to links inside Gemini's answers around September 28, three weeks after it started adding them. Desktop visits still carry the gemini.google.com referrer, but the iOS app opens links in its own WebView with no referrer, so those sessions land in GA4 as Direct and the GeminiiOS user agent in server logs is the only trace.

  • Google Search May Be Bringing Back 100 Search Results Parameter: Barry Schwartz reports that Google looks to be quietly restoring the num=100 parameter, which it dropped in September 2025 as part of its scraping defenses. Brodie Clark caught a query showing 99 organic results triggered by a knowledge-panel click, while Schwartz couldn't replicate it in his browsers.

  • Google Does Not Show All Spammy Links In Search Console: Barry Schwartz reports that Google's Search Console link report only shows a sample of links and doesn't break out spammy ones, disavowed links included. John Mueller told an SEO on Reddit not to anchor on whether Search Console shows the links and to ignore random spammy links outright, consistent with Google's long-standing position that Penguin 4.0 devalues bad links rather than demoting sites. He also digs at agencies selling disavow work off third-party spam scores.

  • Google Business Profiles Asking Businesses For EIN: Google is asking some businesses for an Employer Identification Number and a privacy policy URL when updating parts of their Business Profile.

  • Google Adopts New AI Training Control: Google plans to voluntarily adopt the IETF AIPREF working group's AI training control, using a shared policy token in robots.txt that expresses training preferences across participating AI providers in a single step, with Google implementing in 2027. John Mueller said Google will interpret the token as disallowing Google-Extended where that control isn't explicitly set, and company-specific directives keep working alongside it.

  • Microsoft Bing Separates AI Training From AI Search Grounding With New Controls: Bing is also splitting AI training from AI search grounding with three settings letting publishers pick training, grounding, or both while staying eligible for Bing search results. Temporary grounding controls arrive in Bing Webmaster Tools on December 15, 2026, honoring begins February 1, 2027, old NoArchive/NoCache directives must migrate by March 1, 2027, those stop working on September 30, 2027, and traditional Bing rankings are unaffected.

AI

OpenAI also adds its name to the shared AI training preference effort, announcing it will honor a common robots.txt signal that lets site operators send one training preference to multiple participating AI providers instead of maintaining a separate rule per crawler. OpenAI says implementation lands in the coming months, with its existing per-provider controls continuing to apply and be respected. The mechanics live in the IETF Internet-Draft written by Anthropic's Kevin Kelley, which binds a preference to paths through ordinary robots.txt groups keyed on a new policy token, User-Agent: __usage_ai-training, where a Disallow under that group reads as a preference against training and an Allow reads as a preference in favor.

The draft's example is a three-line group that sets the token to Disallow: / with a single Allow: /press-kit/ carve-out, meaning every path carries an opt-out unless a directory is explicitly chosen.

Google introduces Gemini 4 Carbon, a frontier model built on Gemini 4 Argon, which launched just two weeks earlier. Intro pricing matches Argon at $2 per million input and $10 per million output tokens with a 95% discount on cached input, moving to $4 and $20 after the introductory period.

OpenAI releases GPT-6.1 Terra, a fast, efficient model for the communication and computer-based tasks that fill a workday, from drafting proposals and interpreting technical drawings to filling out forms and editing videos. Standard API pricing lands at $1 per million input tokens and $5 per million output tokens, 50% below GPT-5.6 Terra and GPT-6.1 Sol.

Google introduces Genie 4, a natively multimodal real-time world model that simulates playable worlds with interactive dialogue, ambient audio, and music. Google lists four building blocks behind it: real-time shaping of the simulated world, realistic physics with believable human actions, native audio as a first for a real-time world model, and complex multi-character events. Availability starts as a research preview for trusted testers, with access planned for Google AI Ultra subscribers in Project Genie and early access through the Gemini API and AI Studio.

Dan Taylor reports that OpenAI applies a central safety check named Aura to its web-browsing tools, and flagged sites get refused no matter what their robots.txt or local allow settings say. Blocked decisions come from a global blocklist for security, safety, and compliance reasons, cache for 24 hours on the local client, and surface as a generic browser-security-policy error with no way to request review. His example client, an international media publisher, has been blocked since May 2026: server logs show zero ChatGPT traffic, competitor sites load fine in the same test sessions, and Chrome's "Always allow" permission does nothing.

The diagnostics worth taking are his 7-step routine: test branded URLs and a service page live in ChatGPT, repeat in Codex's browser tool, run competitor site checks, look for agent errors in server logs, confirm robot rules, gather monthly logs as evidence, and only then escalate to support. Check the crawl-data contrast: OpenAI's training crawlers logged only about 5,000 successful connections in 30 days against 477,000 for Google and 564,000 for Anthropic, so an agent refusal moves users toward whichever competitor sites load instead.

WHAT I'VE BUILT FOR YOU

Floyi - Build Topical Authority that wins in Google and AI Search. Don’t just plan your content strategy - make it unstoppable.

ReplyDeck - Be active on Reddit without living on Reddit. Schedule replies, stay present in the threads that matter, and get your time back.

TopicalMap.com Service - Let us do the heavy lifting. We handle the research, structure, and strategy. You get a custom topical map designed to boost authority and dominate your niche and industry.

Topical Maps Unlocked 2.0 - Unlock the blueprint to ranking success. Master the art of structuring content that search engines (and your audience) love - and watch your rankings soar.

LIKE DIGITAL SURFER?

Find me and others in the Digital Surfer Discord community.

I’d also love to know what you think and if you have any ideas for the newsletter. Reply or email me at [email protected].

I’d also appreciate it if you shared it with fellow digital surfers.

{{rp_personalized_text}}

Have a great week taking your SEO and digital marketing to another level!

And don’t forget to drag the Digital Surfer emails to your Primary Inbox 🌊 

Reply

or to participate.