ChatGPT Almost Ruined Santa for My Kid

Wave 305; Spam Update Came and Went; Track AI Mode in GSC; Scrape-to-Referral Insights; ChatGPT Drops Reddit; AI Local Search; X Offers $175K; and Much More!

Author

Yoyao
August 22, 2026 • Est. Learning Time: 13 minutes

FIRST …

Holidays are upon us. Probably one of the last remaining holiday seasons where there’s still magic in the air.

ChatGPT almost ruined Santa for one of my kids when they asked if Santa was real.

In August.

Then it tried to wiggle its way out.

My kid still believes.

What’s up with ChatGPT’s crying emoji? 😭 

ChatGPT never gives me an emoji. Gives them to my kids constantly.

They use my account.

Which makes me wonder what it thinks it knows about who's asking.

FREE CONTENT AUDIT TOOL

Want a topic coverage map like the one below for your domain or a competitor’s?

The tool reads and analyzes up to 100 pages, grouping them by topic.

If you want the full version with more insights and feature, try the Topical Audit tool with Floyi’s Pro trial for 14 days for $7.

SEO + GEO

Google launched its August 2026 spam update globally on August 18 at about 12:30 p.m. ET, covering all languages and regions. The rollout finished about 2.5 days later, on August 21 at 4:50 a.m. ET. Google described it as a normal spam update, not a link spam or site reputation abuse update, and says affected sites should review its spam policies.

My Take: A short rollout does not make diagnosis simple. Mark August 18 to 21 on your reporting calendar, then look for patterns across affected templates, queries, and competitors. Do not blame every ranking move on the update before you check the underlying site changes.

OtterlyAI and Press Ranger analyzed 129.3 million AI-search citations across more than 20 million URLs, seven platforms, and 91 confirmed publisher licensing deals. Their June 2026 snapshot finds that OpenAI-licensed news publishers received more ChatGPT citations per cited page, but the association did not carry across every platform:

  • Its licensed publishers averaged 10.2 ChatGPT citations per cited page, versus 6.9 for unlicensed publishers. That is a 48% premium. The study says this is an association, not proof that the deals caused the increase.

  • Perplexity licensees averaged 2% fewer citations on Perplexity, while Alphabet licensees averaged 9% fewer in AI Overviews and 14% fewer in AI Mode and Gemini. Microsoft Copilot showed an 85% lift, but it applied to licensed publishers broadly rather than Microsoft’s own licensees.

  • Best-of lists, buying guides and money advice, and product reviews and tests made up 58.8% of citations to licensed publishers. Dated news accounted for 5.1%.

  • Niche and trade media earned 75.8% of News and Media citations across 16 industries, or 213% more than mainstream media. They led in 15 of the 16 industries measured.

Suganthan Mohanadasan shows why Google’s June 2026 generative AI report is incomplete for query analysis: it gives feature impressions in the UI but no query-level data through the API or BigQuery export. He compares four workarounds, from an Excel API export and custom GSC regex to MCP tools and a free classifier that labels up to 100,000 unique queries by conversational pattern. Every method is an undercount because Google anonymizes rare queries.

Anu Adegbola reports that ChatGPT was the sixth-most-clicked destination in iPullRank’s analysis of 13.1 billion Google Search events from 9.1 million opted-in users. It also drew a higher share of paid clicks than any other leading destination in the study, while only 11.1% of searches headed toward OpenAI ended in a zero-click Google experience. Google’s zero-click rate rose about 2.6 percentage points over the study period, but paid clicks showed no meaningful change.

My Take: Google rankings and AI visibility are no longer competing scorecards. A user can start with a category search in Google, move to ChatGPT to narrow the shortlist, then return to search when buying intent is clearer. For client work, I would map that handoff from category query to AI prompt to conversion instead of treating each channel as an isolated traffic source.

Gianluca Fiorelli explains why consensus and information gain need to work together for search visibility. Consensus signals that a page is accurate and trustworthy, while original data, first-hand experience, or a distinct framework gives an AI system a reason to cite it instead of absorbing it into a generic answer. He also notes that Google has not confirmed whether its information gain patent is an active ranking system.

My Take: In case you missed it, here’s my video on Information Gain for SEO.

Microsoft Clarity now connects AI crawler activity with referral sessions through an AI Scrape-to-Referral Ratio in Bot Analytics. The report breaks performance down by AI operator and links each source directly to filtered session recordings. Clarity also flags coverage gaps when bot and referral data do not cover the same mapped domains.

Andy Crestodina reports that Direct was the first or second traffic channel in 92% of 145 Google Analytics accounts, representing 36% of 56 million sessions. GA4 also assigns traffic from bookmarks, files, apps, untagged links, and lost attribution to Direct, so it captures both brand demand and measurement gaps. Across 143 properties, Direct generated 27% of all demand and ranked as the second-largest source of leads.

Yulia Deda’s analysis of 100,000 French keywords across 20 niches shows AI Overviews arrived at scale from day one. The study also maps the sources and search patterns behind 52,564 answers with citation data:

  • Broad initial coverage: AI Overviews appeared for 52.63% of keywords, nearly twice the first-month rate SE Ranking recorded in Germany, Spain, and the Netherlands.

  • Video leads the source mix: YouTube appeared in 51.31% of answers and supplied 14.33% of citations, more than the other nine domains in the top 10 combined.

  • Local sources matter: French-language sources made up 89.34% of citations. In Legal, Finance, Real Estate, and Insurance, their share was roughly 97%.

  • Organic rankings still feed AI visibility: 60.46% of cited URLs ranked in the organic top 10, while 76.33% appeared in the top 50.

Stephanie Heitman shares that WordStream tracked 4.45 million AI bot scrapes over six months and received one referral for every 201 scrapes. Its findings show how unevenly AI platforms return value to publishers:

  • The return worsened over time: The scrape-to-referral ratio rose 19.2%, from 207:1 to 248:1, over three months.

  • AI favored expensive research assets: Advertising benchmarks, conversion-rate data, and cost-per-click studies dominated the most-scraped pages.

  • OpenAI led crawling activity: It generated nearly three times as many scrapes as Meta in WordStream’s data.

  • Page formats produced very different returns: Facebook Advertising Benchmarks (2017) generated 2.5 scrapes per referral, while Social Media Image Sizes generated 27,161.

James Allen explains how he built an E-E-A-T audit tool in Claude Desktop with Claude Code. The tool crawls a selected set of page types, captures raw and rendered content, applies a local scoring rubric based on Google’s guidance, and produces a formatted Word report. James also notes that E-E-A-T is not a direct ranking factor, so the output measures alignment with quality guidance rather than predicting rankings.

Matt G. Southern reports that Google’s generative UI is beginning to appear in AI Overviews, where it can create interactive tools and simulations inside Search. In Google’s research, participants preferred a generated interface to the top search result in 90% of head-to-head comparisons for its main prompt set and 73.5% for an information-seeking set. The tests used pre-cached results and smaller custom sites, so they do not prove how users will behave in live Search.

Barry Schwartz reports that Googlebot now applies only one pass of HTML unescaping when it extracts JSON-LD. Double-escaped entities will no longer be corrected automatically, which can affect structured data used for rich results. Google advises site owners to use standard JSON escapes or Unicode hexadecimal escapes instead.

If you're using JSON-LD for structured data, be sure to update your code to standard JSON escapes or Unicode hexadecimal escapes (like \u0026).

My Take: Floyi’s schema generator is already good to go 😁 

Ben Fisher’s study examined 14,472 AI citations across 1,487 local-business queries in 50 U.S. metros. It shows that local AI recommendations are less stable than a normal ranking report suggests:

  • Your own site still matters: 60% of citations pointed to a business’s own website, more than directories, forums, and review platforms combined.

  • One prompt is not a scorecard: Different phrasings of the same question shared cited sources only 40% of the time. Repeating the exact question produced the same top business only about 7% of the time, versus roughly 90% for Google’s local pack.

  • Each model uses a different source mix: Gemini and ChatGPT cited the same domains only 8% of the time and chose the same top business just 4.2% of the time.

  • Community sources still count: Reddit made up 13.7% of Gemini citations, exceeding the 10.3% share from all local-service directories combined.

Matt G. Southern reports that Promptwatch saw Reddit’s ChatGPT Search citation share fall from 3.83% between July 18 and August 7 to 0.52% between August 14 and 17. ChatGPT’s use of site-specific background searches jumped on August 8, but the sharper Reddit drop came six days later. Promptwatch says the cause remains uncertain and has not ruled out a data-collection issue.

Andrea Schultz outlines four technical checks for AI retrieval: crawler access, server-rendered critical content, structured data in the initial HTML, and a consistent brand entity across the web. She notes that many AI crawlers do not reliably render JavaScript, so a page that works in Google can still expose only an empty shell to another system. She recommends testing raw HTML, robots rules, crawler access, and schema rather than assuming a normal search crawl covers every case.

Matt G. Southern reports that llms.txt V2 adds two standard links that point agents from an HTML page to its Markdown version and its related llms.txt file. Sites can add those links in HTML or an HTTP header, and V2 also supports a shorter Markdown URL format. Google Search still says llms.txt neither helps nor harms rankings, so the update is for coding agents and documentation tools, not a shortcut to AI Overview citations.

Kelsey Libert’s Fractl analysis compared 4,320 responses and more than 8,500 brand references across GPT-4o, Gemini 2.5 Flash, and Claude Sonnet 4.6. It found that Google authority and AI recall usually align, but the exceptions create real brand gaps:

  • Strong SEO can still miss AI answers: 471 brands, about 5% of the dataset, had high domain authority, traffic, and keyword reach but drew few LLM references.

  • Smaller brands can punch above their SEO weight: 377 brands, about 4%, appeared in AI answers more often than their traditional search signals would predict.

  • Third-party coverage separated the outliers: Brands repeatedly featured in roundups, comparison pages, expert lists, and reviews were more likely to overperform in AI recall.

  • Model coverage is fragmented: Only 11% of brands appeared in all three models, while 77% appeared in just one.

Matt G. Southern explains that local AI answers turn a long list of businesses into a short recommendation set before a customer reaches a website. Google says these answers use its existing ranking and quality systems, then pull supporting pages through retrieval and related searches through query fan-out. Search Console can show link impressions from Google’s AI features, but not the queries behind them, whether a business was recommended, or what other assistants said.

Google Research scientists Nitay Calderon and Gal Yona tested 13 AI models on 2,150 Wikipedia facts. Their knowledge profiling framework measures the state of each fact, not just whether a single answer was right. It separates facts a model never encoded from facts it encoded but could not retrieve.

  • Five profiles diagnose the failure: The framework classifies each fact as an encoding failure, recall failure, direct recall, recall with thinking, or inference without encoding.

  • Facts can be stored but unreachable: Gemini-3-Pro and GPT-5 encoded 95% to 98% of the facts, yet failed to recall 26% to 34% of them directly.

  • Thinking closes part of the gap: With thinking enabled, the models still missed 11% to 12% of facts that they had encoded.

  • Extra reasoning helps stored knowledge: Thinking recovered 40% to 65% of encoded facts that the models could not initially answer, but only 5% to 15% of facts that were not encoded.

  • Subject-object order affects recall: WikiProfile extracts facts as ordered subject-object pairs, with the subject appearing first in the source document. A model could often recognize a reverse relationship in multiple choice, but found it harder to generate when the question asked for the relationship in the opposite direction.

SEO + GEO Ripples

AI

Google can generate custom interactive visuals and simulations in AI Mode, now available globally in English, with the feature also beginning to roll out in AI Overviews. Google has also added free, custom practice quizzes for subjects and exams including the ACT, GRE, LSAT, MCAT, and SAT. Notebooks in AI Mode will sync sources and instructions with Gemini Notebook and roll out across more than 180 countries in English, excluding the EEA.

Meta has opened Pocket to all U.S. users after testing the AI game-making app in Brazil. People can turn prompts into small interactive games, publish them to a feed, and let others save, remix, or repost them. Meta acquired the team behind Gizmo earlier this year and is shutting down the original app.

Meta says its new business features let Meta AI use Facebook and Instagram analytics, Meta ad data, and Google Workspace context in the same conversation. It can turn that information into reports, decks, and spreadsheets, then run recurring tasks and reminders. Meta also says users can compare public content and engagement patterns from similar brands.

Andrew Hutchinson reports that X is offering $100,000, $50,000, and $25,000 for the top three Grok Imagine videos based on Homer’s Odyssey. The official contest terms require U.S. entrants aged 18 or older to submit a three-to-five-minute English video with at least one minute of dialogue by August 31. Entries will be judged on submission requirements, creativity, originality, and verified Home Timeline impressions.

Dan Goodin reports that Varonis researchers found an undocumented Copilot URL parameter that could let a crafted link run a prompt inside a user’s authenticated Microsoft 365 session. Microsoft mitigated the flaw in February after Varonis reported it, then released broader fixes as CVE-2026-24301. The researchers also demonstrated that a malicious webpage could inject instructions into Copilot’s persistent memory when the user asked it to summarize the page.

Ana-Maria Stanciuc reports that Anthropic began watermarking Claude text worldwide on August 2 to meet commitments under the EU AI Act. Within days, removal tools appeared. One open-source project drew more than 14,000 GitHub stars, while U.S. Google Trends interest in “AI watermark remover” rose 60% week on week. The watermark uses patterns in word choice, so heavy rewriting can remove it. Anthropic says the mark does not identify a user or decide who owns the text.

These are two of the tools mentioned: MarkScrub repository and Watermarks Remover repository.

Manisha Priyadarshini reports that Gemini users can now turn visible watermarks on or off for Nano Banana images, Omni videos, and Lyria music. Google’s invisible SynthID watermark and C2PA metadata remain in every file. The option is limited to paid AI Ultra users in India, South Korea, and Vietnam, and is unavailable for work and school accounts.

Carl Franzen reports that SpaceXAI has opened an early beta of Grok Bot, a system of persistent AI workers that can use apps and websites through their own computer environments. The Bots can keep working after a laptop closes, learn routines by watching a user complete a task, and hand work to other Bots. Team access costs $120 per seat each month, while the individual Cursor Ultra plan costs $200. SpaceXAI has not released performance benchmarks for the product.

AI Ripples

  • ChatGPT Ads expands across Europe: OpenAI says ChatGPT Ads will enter 31 European markets through its Ads Solutions team, agency partners, and technology partners before self-service access arrives later in summer 2026. Ads will appear only on Free and Go plans, and the platform now supports conversion optimization, geo-targeting, custom audiences, the OpenAI Pixel, Conversions API, and third-party measurement.

  • Introducing ChatGPT for Teens: Built for learning, backed by protections OpenAI says users aged 13 to 17, or those its system estimates are under 18, will be placed automatically in ChatGPT for Teens. The experience adds Study Mode, homework reminders, quizzes, Learning Visualizations, Study Hours, age-appropriate safeguards, and parent controls including Quiet Hours and limited safety notifications.

  • Stripe agrees to acquire OpenRouter to help businesses optimize token routing and usage: Stripe says it has agreed to acquire OpenRouter, the model-routing platform that supports more than 400 models from over 80 providers. It plans to pair OpenRouter’s task-based routing across price, speed, reliability, and complexity with Stripe’s existing token billing tools.

SOCIAL MEDIA

Hari Srinivasan from LinkedIn says more than one million people have used LinkedIn's “seems like AI slop” feedback option since it launched two weeks earlier. Posts that receive enough feedback may now trigger an analytics message for the creator, although Srinivasan says no single report determines distribution and safeguards are meant to prevent targeting. LinkedIn says members are seeing 40% fewer views of content it classifies as AI slop than a few weeks ago.

YouTube Studio’s Copyright menu will become a broader Claims tab that groups Content ID and AI likeness restrictions. The update adds four dispute categories for likeness claims and will show both the filer and the person whose likeness was identified. YouTube says a likeness claim can restrict or block a video, but it will not create a copyright or Community Guidelines strike.

YouTube will count a public view when any video starts playing, bringing long-form and live video in line with Shorts. The earlier 30-second threshold for standard and live video will no longer define public views. Engaged views and watch hours will remain the measures used for earnings and YouTube Partner Program eligibility.

Social Ripples

  • TikTok owner ByteDance signs Motion Picture Association deal: Andrew Hutchinson reports that ByteDance and the Motion Picture Association have agreed on a framework for stronger intellectual-property protections in generative video and image models, including Seedance and Seedream. The agreement covers products such as TikTok, CapCut, and Dreamina after the MPA called for tighter restrictions in February.

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.

AD

📣 Only important people get an Intelligence Report. So why don't you? Join 250,000 Founders & CEOs getting Max's CEO Report! The quick, easy, but highly impactful end-of-day business news recap. Get It For Free, Here!

What Did You Think of This Week's Wave?

Login or Subscribe to participate in polls.

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.

You currently have 0 referrals, only 3 away from receiving LinkedIn Shout-out.

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.