A Paraphrase Scores Near Zero

Wave 301; SerpApi Wins; TikTokification of Search; Publishers Cutting Off Google; New Content SEO Manager Role; GPT-5.6 Hacks Hugging Face; and Much More!

Author

Yoyao
July 25, 2026 • Est. Learning Time: 11 minutes

In partnership with

FIRST …

SEO content used to be a copying exercise.

Pull the top ten. Inventory their terms and entities. Cover them. That's the entire reason content optimizers exist. It's why Floyi has one.

But matching the SERPs and competitors is table stakes now. There is more.

Floyi took that baseline and added AI response coverage: AI Overviews, AI Mode, Gemini, and ChatGPT. 

Cover what the SERPs say and what the AI answers say. That's the new floor.

Then comes the part many are talking about, but almost nobody is working on.

Originality. 

Google has patents on it, scored as information gain: the facts your page introduces that no competitor or AI can. 

Originality is measured as semantic distinctness, not different wording.

So a paraphrase scores near zero. 

Everyone's advice is to go find a statistic to add.

It’s not “original” if you’re able to search for it. 

You can't automate that from public data. A tool that only reads what's ranking can only suggest a remix of what's ranking.

It has to come from your brand. Your positioning. Your client results. The documents nobody outside your company has read.

Floyi already holds all of it. Your brand foundation, your audience and up to 2,000 of your own documents.

That's what made this week's release possible. Floyi reads your knowledge base and surfaces the angles only you can back up.

And when strong angles are missing data in the knowledge base, Floyi refuses to invent numbers and asks you for it.

Cover everything they cover. Then say the one thing they can't.

AD

How owning AI deployment expands your career

Across product, ops, and CX teams, a new kind of role is taking shape: the person responsible for making AI actually work, day to day. In this roundtable, three people living this shift share what it's really like: Simone Santiago Broad (Yoco), Yelva Espinoza (Zumba Fitness), and Fin's Dave Lynch. You'll hear how they carved out these roles, what the job looks like across industries, the skills they'd hire for, and the challenges they're tackling right now.

Watch the full conversation on demand.

SEO + GEO

Ben Wills reports that his July dataset covers 403K+ prompts across 10 models, 145,289 recommended domains, and 18 external signals. The data shows correlation with LLM visibility, not a formula for causing it.

  • Being linked from search results mattered more than simply showing up: This was the clearest pattern found. It still explained only about 11% of the difference in LLM visibility, so it is an input, not a shortcut.

  • Homepage topic match mattered more than before: The signal from words in a site's title, description, and page copy rose sharply since May. He gives title words the most weight, description words next, and body copy the least.

  • Authority worked best at the domain level: Site-wide authority signals were more useful than measurements from individual hosts or pages.

  • Entity and community signals helped: Brands connected to Wikidata and discussed on Reddit showed stronger patterns than a basic Wikipedia presence.

Danny Goodwin reports that a federal court dismissed Google's two DMCA claims against SerpApi and paused discovery while Google decides whether to amend its complaint. The court permanently dismissed claims tied to Search results without copyrighted content. It allowed Google to revise claims involving copyrighted results because its SearchGuard allegations did not show the system acted with a copyright owner's authority.

My Take: The decision turns on a narrow legal question, not a blanket approval for SERP scraping. Anyone building on search-result data still needs to account for platform terms, technical limits, and the cost of a collection model that can survive a legal challenge.

Mikael Arajuo says AI Mode is shifting search from document retrieval toward longer, platform-contained sessions where follow-up queries and cited source fragments matter as much as a click. He points to Google Creator Profiles, citations from social platforms in AI responses, and a reported 40% month-over-month increase in U.S. follow-up queries as signs that creator identity is becoming more visible in search. His recommendation is to build attributable, verifiable content and map how named experts appear across the sources Google can connect.

Margarita Loktionova shares that Semrush and Kevin Indig studied 1,094 ChatGPT categories from January through June 2026, covering more than 50,000 brands and 600,000 citations. The research defines a topic owner as a brand mentioned in at least four of five related prompts with a five-percentage-point lead.

  • Most topics remain open: Only 15.2% of categories had a clear owner. Another 31.2% had an emerging leader, while 53.7% had no brand appearing in at least three of five prompts.

  • Broad SEO strength did not decide ownership: Topic owners had more branded search volume in 55.7% of comparisons, more organic traffic in 48.4%, and a higher Authority Score in 52.5%.

  • Margin made ownership durable: Clear owners stayed in first place in 90.4% of month-over-month comparisons. Categories that changed leaders had a median lead of 1.3 percentage points, versus 2.9 points where the leader held.

Andy Crestodina analyzed 28.9 million sessions across 97 B2B lead-generation sites through June 2026, with every GA4 key event manually categorized by intent.

  • AI referrals converted at a higher rate: Visitors from AI sources were three times more likely to become leads than other organic sources. The pattern appeared on roughly two-thirds of the sites.

  • Volume was still tiny: AI sources sent 140,000 sessions, or 0.5% of all traffic. Crestodina notes that GA4 records clicks from AI Overviews and AI Mode as Organic Search, and some AI-app visits as Direct.

  • ChatGPT led the measured sources: ChatGPT accounted for 82.3% of AI-referred visits. It converted 2.1% of visitors into leads, compared with 0.5% for Google Search, while Google delivered 100 times more traffic.

Alexandra Bruell reports that Reddit, USA Today, Politico, the Economist, People Inc., and Reuters are reassessing how they work with Google as AI answers reduce outbound search clicks and disrupt publisher revenue models. Reddit has discussed restricting Google's access to content for AI use as the companies negotiate over a 2024 deal reportedly worth $60 million a year that is nearing its end.

Maryanna Franco reports that she and João da Silva tested 12 U.K. athletic-apparel brands across 14,140 API runs on ChatGPT, Gemini, Perplexity, Claude, and Google AI Overviews. Changing only the category in the prompt from athleisure to athletic footwear moved New Balance from a 1% to a 90% recommendation rate, while lululemon fell from 90% to 0%. Franco says the pattern reflects category coding, where a brand's Knowledge Graph description and third-party content history together determine which category queries an AI system associates with it.

My Take: This is why brand and audience must be the first things you set before anything else. When I build a topical map, I start with the categories customers actually use, then make sure the site, expert content, and external mentions reinforce the same association. A strong entity in the wrong category still misses the recommendation.

Roger Montti connects Liz Reid's Google I/O comments about agentic search with a February 2026 patent for delivering results after an initially unanswered query. The patent describes storing queries that lack a sufficiently high-quality, authoritative, or complete answer, then monitoring for new or updated information and proactively notifying the user. It outlines six trigger scenarios and allows delivery through notifications, later assistant conversations, or another device.

Barry Schwartz reports that Google Search head Nick Fox says AI Mode and AI Overviews send billions of clicks to websites each week, while Google Search overall sends billions daily. Fox did not say whether AI features send more clicks than before. Barry notes that Search Console's AI performance reporting does not include clicks, leaving publishers unable to compare AI-feature click-through rates with traditional search.

My Take: A big claim of BILLIONs - but no proof with data. The number I want is the click, conversion, and revenue change for a comparable query set before and after AI features appear. Until that exists, treat broad platform claims as context, not a traffic forecast.

Margarita Loktionova says that the Content SEO Manager title, absent from Semrush's 2023 data, now represents 20% of U.S. content marketing listings and ties Content Creator for the highest volume. Her review of 1,002 Indeed listings found SEO in 62% of senior roles and 59% of non-senior roles, while analytics appeared in 49% and 40% respectively. Senior listings also mentioned AI in 43% of cases, with 11% specifically naming AI Search, SGE, or AEO.

Matt G. Southern reports that the European Commission adopted binding Digital Markets Act decisions requiring Google to share anonymized query, click, view, and results-position data with eligible rival search engines and search-enabled AI chatbots. Google does not have to share its ranking algorithms, FastSearch system, or sensitive data such as account details, search histories, timestamps, and rare or long queries. Eligible providers need at least 50,000 monthly EU users, an operating-history or investment threshold, security screening, and an independent audit before data access begins.

SEO + GEO Ripples

  • Google Warns Not To Place AI Labels On Corners Of Images: Barry Schwartz reports that Google changed its AI-content-label guidance from a defined 5.5% image margin to a broader warning not to place labels in the very corners for ad images. Google says varied rendering formats can still trim a label, so ad and creative teams should test how disclosures survive across placements.

AI

Ashley Capoot reports that OpenAI said GPT-5.6 Sol and an unreleased model escaped a sandboxed evaluation environment, reached the internet, and exploited a Hugging Face vulnerability while trying to find information to cheat an evaluation. Hugging Face called it an incident driven end to end by an autonomous agent system. OpenAI says it is tightening containment, monitoring, access controls, and evaluation practices.

Update: OpenAI and Hugging Face are conducting a joint forensic investigation, patching the disclosed vulnerability, and bringing Hugging Face into OpenAI's trusted-access program so its teams can use the models to strengthen defenses.

OpenAI is rolling out Health in ChatGPT to logged-in U.S. users aged 18 and older on web and iOS, including Free, Go, Plus, and Pro plans. With permission, users can connect Apple Health, supported medical records, One Medical, or Function Health so ChatGPT can compare results, summarize changes, and use relevant health context across conversations. OpenAI says connected health data and conversations that use it will not train foundation models or target ads, and users can disconnect accounts at any time.

Zanna Iscenko and Scott Strand report that Google's first AI & Economy ATLAS study analyzed 15 million aggregated, de-identified interactions across the Gemini App, AI Mode, and Gemini API. The study found AI use in 68% of U.S. occupations, yet a typical job uses it for about 21% of tasks and fewer than 10% of work interactions fully automate a task. More than 86% of interactions occurred outside work, often around household research and administrative tasks such as taxes, licensing, and fines.

OpenAI introduced a small-business program built around virtual training, in-person AI academies, guides, and partner workflows for ChatGPT Work. The program names partners including Dropbox, Shopify, Intuit, Slack, Atlassian, and Wix. OpenAI says a prior Small Business AI Jam saw 78% of participants build a functional AI workflow in one day, while 42% reported saving more than five hours a week.

AI Ripples

  • Introducing Claude Opus 5: Anthropic released Claude Opus 5 at $5 per million input tokens and $25 per million output tokens, matching Opus 4.8's price. The company says it leads its software engineering and knowledge-work evaluations, while targeting longer-running agent workflows. Anthropic also added beta API support for mid-conversation tool changes and automatic safety fallbacks.

  • Introducing Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber: Gemini 3.6 Flash uses 17% fewer output tokens than Gemini 3.5 Flash on the Artificial Analysis Index and is priced at $1.50 per million input tokens and $7.50 per million output tokens. Google also released 3.5 Flash-Lite that’s used for Google Search now, while the cyber-focused model is heading to a limited CodeMender pilot for governments and trusted partners.

  • Anthropic’s $1.5B copyright settlement approved; only 350 authors opted out: Ashley Belanger reports that a federal judge approved Anthropic's $1.5 billion settlement with authors, with about 91% of affected authors and publishers filing claims and only 350 class members opting out. The court reduced requested legal fees to about $101 million and noted the settlement requires Anthropic to destroy covered works.

  • Google Faces Class Action Over Books Used To Train Gemini: Three publishers, Scott Turow, and S.C.R.I.B.E. filed a proposed class action alleging Google used books and articles supplied through Google Books, Play Books, and Scholar to train Gemini without permission. No court has ruled on the claims, and the case concerns sourcing routes that robots.txt controls such as Google-Extended do not govern.

  • Pichai Says Google Needs Gemini 4 To Compete At The Frontier: Sundar Pichai said Google needs a larger Gemini 4 base model to improve coding and agentic coding at the frontier. Gemini 3.5 Pro remains in partner testing with no confirmed release date, while Gemini 4 is in pretraining.

CONTENT

Dan Petrovic analyzed 13,147 scraped Engadget pages using Shannon entropy and gzip compression, then removed 7,933 product-hub templates to study 5,214 genuine articles from 2006 to 2023. Median article length rose from 1,614 characters in 2006 to about 10,000 to 10,500 during 2015 to 2017, then fell to roughly 4,200 to 4,900 by 2022 and 2023.

His companion analysis of The Verge found a different trajectory: 152,192 dated articles more than doubled in median length from 3,961 characters in 2011 to 8,376 in 2023, while length-adjusted compressibility stayed flat.

LINK BUILDING + DIGITAL PR

Vince Nero analyzed 14.3 million outreach emails sent from 1,785 BuzzStream domains between June 2025 and June 2026. He frames the findings as correlation, not proof that domain age directly causes better deliverability.

  • Young domains lagged: Domains aged one to two years averaged a 21.6% open rate, compared with 29.6% for three to four years and 34.9% for five to six years.

  • Experience did not erase the gap: Seasoned agencies and new solo senders both averaged 25.3% on young domains across 95 domains.

  • Authentication was not the full explanation: Fully authenticated one- to two-year-old domains averaged 17.9% opens, versus 27.9% for domains older than 16 years. New custom domains under three years also trailed Gmail at 21.2% versus 30.0%.

SOCIAL MEDIA

Common Thread Collective reports that TikTok plans to pilot a managed-services program in August that takes over ads, listing optimization, creator recruitment, and content production for Shop sellers. The program carries a $10,000 upfront fee plus a 10% to 20% sales commission, with TikTok using GMV Max for paid campaigns and producing hundreds of AI-generated videos. Sellers would primarily list products and distribute samples to influencers.

Facebook introduced Facebook Verified, a free, selfie-based badge for eligible users aged 18 and older who meet its trust and safety standards. The badge will appear on profiles, Marketplace, Dating, and Groups first, with Feed posts planned later. Facebook says the badge confirms that a profile belongs to a real person, not that the company endorses the person or guarantees their trustworthiness.

YouTube clarified three categories that can make a channel ineligible for the YouTube Partner Program: generic or repetitive material, off-putting content built around shock or manipulation, and AI personas presenting information on sensitive topics such as health, law, finance, or politics. YouTube says the clarification does not change its existing policy. Videos that fail monetization rules can remain on the platform if they follow Community Guidelines.

Social Ripples

  • Introducing Seller, an App for Facebook Marketplace Sellers: Facebook launched Seller, a dedicated iOS app for U.S. Marketplace sellers aged 18 and older. It combines AI-assisted listing creation, a unified inbox, inventory management, and performance data in a marketplace that now sees 430 million items listed each month globally

  • Making thumbnails easier on YouTube: The YouTube Team says YouTube Partner Program creators can now upload custom thumbnails for Shorts, with broader access planned later. Desktop users can choose from three suggested frames, while Ask Studio can generate and revise thumbnails for long-form videos through chat.

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

How to Keep Your Retainers From Leaking Data

Unrecorded hours don't announce themselves. They just don't get billed. Timeglass captures every hour on the retainer automatically and shows you hours against budget as it happens.

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.