SEO Recap covering August 25, 2026: ChatGPT's new search language and Google's AI spam crackdown

Built by Stephanie Chung·9 min read·14 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)

AI visibility now means optimizing what SEO controls and mobilizing the rest of the org
Source: Search Engine Land

  • Being cited by AI in informational answers does not mean being recommended when a buyer asks what to purchase, because AI applies different criteria when it shifts from retrieval to advice.
  • Analysis of large brands found products dropped from AI recommendations for reasons outside SEO's control:
    • Higher maintenance requirements than competitors
    • Missing capabilities for the buyer's specific application
    • Consistent reports of poor support for complex issues
    • A component with a reputation for frequent failure
    • Cloud connectivity reported to drop frequently
  • When the cause is product, support, or policy, the SEO/GEO team's job is to bring cross-functional owners a defined business problem: which prompts lose, why, how often, and which revenue it affects.
  • Pacesetter programs create 2 layers of ownership. The GEO team owns monitoring and diagnosis while Product, Support, or Finance owns the fix.

ChatGPT rebuilt its search tool with a new compact query language
Source: Search Engine Journal

  • Between August 16-20, OpenAI replaced JSON search calls with a pipe-delimited line format, one search per line, and dropped the search_queries fan-out field entirely.
  • Each line carries a call type, query, a number, and sometimes a domain, with call types spanning verticals:
    • fast for web search
    • product for catalogue lookups on physical goods
    • business for places lookups with a location
    • image for image search
    • genui_run for rendered widgets like stock charts and maps
  • The third field appears to be a recency window in days matched to how fast the answer goes stale (2 for a stock price, 7 for football results, 30 for commercial product research, 90 for earnings).
  • A domain slot lets ChatGPT visit a named brand's site directly, filled from memory, so a stale or wrong domain (profound.ai instead of tryprofound.com) can waste a probe.
  • Reddit was fetched with a 365-day window on a tools query and a 3,650-day window on a coffee query to gather opinion on brands ChatGPT already picked, yet none of the 84 Reddit threads in one retrieval pool were credited in the answer.

International SEO now needs machine-recognizable E-E-A-T
Source: Search Engine Journal

  • AI does not inherit authority across markets. Expertise must be evidenced in a form the model recognizes as belonging to that market, the same lesson link building taught years ago.
  • Trained on near-identical content across a brand's 40 regional sites, a model can collapse them into one composite representation, flattening the localized signals meant to prove local authority.
  • Local professional credentials often go unrecognized because models trained mostly on English content have not learned that designations like Architekt BDA, Ordre des Architectes, or First-Class Registered Architect equal "expert."
  • Practice "authority translation" by connecting author credentials to the institutions, certifications, and bodies behind them, and give each market genuinely distinct content rather than repeated localizations.

Mueller says nothing special is needed for generative AI responses in Search
Source: Search Engine Roundtable

  • John Mueller said on Bluesky there is nothing special you need to do for generative AI responses in Google Search, consistent with Google's repeated line that SEO is the same as AEO or GEO.
  • Mueller admitted he did not fully understand the question, but the statement aligns with prior Google myth-busting on AI optimization.
  • Microsoft Bing disagrees with Google on whether separate GEO work is needed.

AI in Search / AI Overviews

Google brings developing-topic link carousels to AI Mode
Source: Search Engine Journal

  • Google made developing-topic link carousels live in AI Mode, showing a horizontal row of timely article cards (image, headline, source, date) between sections of an AI response.
  • The format, first introduced in AI Overviews in May, appears only for some trending searches and can highlight Preferred Sources.
  • Google says more than 600,000 unique sources have now been selected as Preferred Sources, up from more than 345,000 in May.
  • The carousel is a larger visual placement than inline citations and its links can generate Search Console impressions and clicks, though there is no filter to isolate it.

Google shows AI Overviews above stock charts for stock price queries
Source: Search Engine Roundtable

  • Google is now placing AI Overviews at the top of results for stock price queries like [GOOG stock] or [MSFT stock], above the stock chart widget.
  • This differs from a version spotted in June, where the AI Overview appeared below the stock chart rather than above it.
  • The placement pushes down the stock widget and arguably makes results less relevant for a query the chart already answers.

Google adds copyable emojis to AI Overviews
Source: Search Engine Roundtable

  • Google is now showing copyable emojis inside AI Overviews, extending a feature it added to answer boxes about a year ago.
  • Brodie Clark spotted the change, and it can appear redundantly in both the AI Overview and the answer box on the same query.

Technical SEO

Cloudflare says EmDash CMS version 1.0 is releasing soon
Source: Search Engine Journal

  • Cloudflare disclosed that EmDash, its serverless TypeScript CMS positioned as a spiritual successor to WordPress, will reach version 1.0 soon after migrating its own blog by August 12.
  • EmDash runs on Astro with securely sandboxed plugins via Dynamic Workers, addressing the plugin security problems of WordPress, which powers over 40% of the web.
  • During testing the blog handled normal load ~75 requests per second, spikes over 5,000 RPS, and a 28,000 RPS DDoS attack, with post scheduling flagged as a remaining issue.

Google is showing fewer PDF files in search results
Source: Search Engine Roundtable

  • Google appears to be showing fewer PDFs and dropping them from the index, including forms from major sites like IRS.gov and New York State, even when queries specify PDF.
  • Savanna Gray and Lily Ray flagged dramatic drops in PDF clicks in Search Console, and other SEOs reported top PDFs falling to zero impressions starting around August 12-18.
  • The IRS site itself was reported inaccessible and dropped from Google Search, though that connectivity issue does not explain missing NY State PDFs.
  • Users can still force a PDF with a filetype:pdf query, but the average searcher will not.

Google's crawlers do not parse JSON, Illyes says
Source: Search Engine Roundtable

  • Gary Illyes reiterated on Bluesky that Googlebot only downloads files and does not parse JSON. Parsing happens later, for example during indexing for Search.
  • Google uses shared infrastructure across services, so the same parser applies to systems like Merchant Center.

Organic Search & Algorithm Updates

Turn an SEO backlog into a roadmap that survives scrutiny
Source: Search Engine Land

  • A backlog lists tasks while a roadmap ties each initiative to a business outcome, an owner, dependencies, expected impact, cost, and a measurement plan.
  • Score and sequence initiatives with the SCOPE framework:
    • Strategic alignment to executive goals
    • Confidence in delivery without dependency derailment
    • Ownership and team capacity
    • Potential impact you can measure honestly
    • Effort and elapsed time
  • Mix quick wins (canonical fixes, schema on commercial pages) with long bets (internal linking rebuilds, pSEO directories) so results land before slower initiatives compound.
  • SEOFomo's 2026 survey found implementation bottlenecks and development constraints were the most common reasons SEO projects fell short of expectations.

Whitespark study of 1.8 million GBPs on category and completeness effects
Source: Search Engine Land

  • Whitespark analyzed 1.8 million Google Business Profiles across 4,209 categories. Specific primary categories were associated with a ~36% higher presence in the top 10 than generic ones (12.5% vs 9.2%).
  • Exact-match primary categories outrank adjacent ones, and well-chosen specific additional categories correlate with rankings 6-17 positions higher than a single primary category.
  • On a 0-5 completeness index (website, description, hours, photos, claimed status), moving from 0 to 5 improved average rank by ~19 positions and tripled top 10 appearance rate.
  • Skip the generic "Service establishment" category, which the data suggests is harmful, and pick the most specific primary category available.

Wix data: .org and .store outperform .com on ecommerce revenue
Source: Search Engine Journal

  • Wix found .org sites are 34% more likely to generate ecommerce revenue than .com and receive 3% more sessions, while .store sites earn 10% more revenue than .com and 2 times more than .shop or .net.
  • 82% of US Wix sites use .com, and people are 13 times more likely to use .com than any other extension, though local extensions lead in Germany, the Netherlands, and Switzerland.
  • Registrations of .store, .online, and .shop doubled in January-May 2026, .ai registrations rose 20%, and .io registrations surged 32x in May 2026.
  • Consider defensive registration beyond your brand's .com, including non-brand keyword domains and trending extensions a competitor might grab.

Reports tie Google's August spam update to mass-generated AI content
Source: Search Engine Journal

  • Reports suggest part of the August spam update targeted AI-generated content built expressly to rank, with sites automatically posting via tools like Claude Code appearing to be filtered.
  • Google recently published a research paper on a Scalable Cluster Termination System (S-CTS) designed to terminate networks of mass-generated AI spam.
  • Observers argue the trigger is the method of mass production, not AI authorship itself, since Google defines abuse as generating many pages primarily to manipulate rankings.
  • Sites with accumulated domain trust and past engagement appear more resilient after switching to LLM automation, and content narrowly focused on keywords has been hurting for the past couple years.
Published on
geoai searchtechnical seoorganic searchdigest