StephEO

SEO Recap covering July 29, 2026: New AI citation studies and Google's Open Knowledge Format

Built by Stephanie Chung·11 min read·15 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)

The Ghost Citation Problem: When AI Cites You Without Naming You
Source: Search Engine Land

  • Writesonic analyzed roughly 16 million brand appearances and found about 40% of AI citations did not name the source brand in the generated answer, what the authors call a ghost citation.
  • Ghost citation rates ranged from 19% to 52% by engine:
    • Perplexity 52%
    • Google AI Mode 49%
    • Google AI Overviews 41%
    • ChatGPT 37%
    • Gemini 25%, Grok 22%, Microsoft Copilot 19%
  • The study splits engines into "citers" that link often but name brands less (Perplexity, Google) and "namers" that mention brands but link less (Gemini, Copilot), so optimization should differ by engine.
  • The authors recommend tracking citations and mentions separately, building named proprietary research, and pairing the brand name directly with each claim rather than a few paragraphs away.

Prompt Injection Is Now a Brand and AI Workflow Risk
Source: Search Engine Land

  • The author notes simple hidden injection (white-on-white text, HTML comments, invisible Unicode) no longer works on modern LLMs, but more sophisticated attacks do because models cannot reliably separate content from instructions.
  • The piece catalogs live attack vectors:
    • ChatGPhish payloads in blogs and help centers that render fake alerts and QR codes inside the chat
    • Semantic embedding that tells browsing agents to recommend a competitor and steals LLM referral share
    • Multimodal attacks hidden in images, audio, and voice tone
    • Confused-deputy hijacks of customer support agents, plus a supply chain breach tied to malicious LiteLLM versions
  • The recommended defense is the Dual-LLM pattern plus four requirements: map AI access across retrieval, memory, planning, tool selection, and output, require human confirmation for high-stakes actions, audit data flow not just control flow, and vet every third-party AI vendor.

AI Visibility Rankings Are Mostly Statistical Noise, New Research Shows
Source: Search Engine Journal

  • An IQRush paper by Ron Sielinski argues a single AI visibility reading can mislead because generative models add randomness, and it offers a stopping rule: a ranking is trustworthy only when the order stops changing and the gap between top sites exceeds the margin of error.
  • Across 30 platform-topic tests, the number of cited answers needed to meet both conditions ranged from 33 to 94, and 3 tests on SearchGPT never separated their leaders even after 125 questions.
  • A University of St. Gallen team reached the same verdict independently, and the author advises measuring before and after a content change more than once and reporting a range rather than a clean single figure.
  • A separate SE Ranking analysis of 100,000 keywords found Reddit gained top 3 share in all 20 niches after the May core update, rising to 10.24% overall, with the biggest gains in experience-led niches like pets, education, and sports.

89% of AI Search Demand Has No Clear Owner Yet
Source: Search Engine Journal

  • Kevin Indig analyzed six months of ChatGPT answer data from Semrush spanning 1,094 US categories, 50,000-plus brands, and over 600,000 citations, defining a clear owner as a brand leading 4 of 5 prompts by at least 5 points.
  • In June 2026 only 15.2% of categories had a clear owner and 53.7% were unsettled, and because higher-volume categories are less settled, 89.3% of estimated AI search demand sits in categories with no clear owner.
  • Clear owners held first place in 90.4% of month-over-month comparisons, and leader changes clustered where the median lead was just 1.3 points versus 2.9 points for defended categories.
  • Citations and mentions were only weakly, slightly negatively correlated (-0.229), with the most cited domain matching the most mentioned brand just 20.8% of the time, so citations open the door but do not decide who ChatGPT names.
  • Indig recommends picking 10 to 20 priority categories, tracking the same 5 prompt types (definition, comparison, alternatives, use case, buying), and treating any lead under about 3 points as contested.

How Perplexity Actually Picks Sources
Source: Search Engine Journal

  • The author captured Perplexity's live SSE stream on a logged-in Pro account (8 captures, 7 query types, single Dubai user) and separates firm structural findings from directional single-user counts.
  • Perplexity ships a full classifier scorecard to the browser, with 16 heads each carrying a probability, a fixed threshold, and a fire/no-fire flag, plus a domain taxonomy label, so practitioners can map priority queries to the surface they actually compete for.
  • A July build re-added a per-source trust object with scoped first-party notes (level 1 credible, level 2 trusted), describing what a domain can be believed on rather than a general authority score, though the field has changed across builds.
  • Unlike ChatGPT's text bucket that answers from memory, Perplexity searched the web on all 7 queries (skip_search false) and silently escalated a how-to to Study mode with a video tab, so every query is contestable.

AI in Search / AI Overviews

The AI Halftime Report: H1 2026
Source: Search Engine Land

  • AI Mode reached 1 billion monthly active users with queries about 3x longer than classic search, and Google called it the biggest search box upgrade in 25 years at I/O 2026.
  • Growth Memo research found citation and mention overlap between engines is tiny, with 91% of citations appearing in only one of ChatGPT, Perplexity, or AI Overviews, and argues brand mentions matter more than citations for most vendors.
  • On trust, close to 75% of consumers pick the number one result in an AI shortlist, and users accepted AI Mode product recommendations as the best 88% of the time.
  • Organic referral loss landed at about 33% year over year, with 68% of Google searches ending without a click, pushing publishers toward courts and regulators.
    • A Munich court ruled Google liable for false AI Overviews statements
    • 400 newspapers sued OpenAI and Microsoft over content use
    • The UK CMA ordered Google to give publishers opt-outs and clearer attribution
  • The consumer AI market fragmented, with ChatGPT share falling from 78% to 56% year over year while Gemini rose from 15% to 30% and Claude from 2% to 10%.

Technical SEO

Google Search Console Platform Properties Now Globally Live
Source: Search Engine Land

  • Google rolled out platform properties to all Search Console accounts globally, letting owners track how their Instagram, TikTok, X, and YouTube content performs in Google Search, Discover, and News.
  • This gives owners performance data for properties they do not have developer access to for the first time.
  • Google published a new guide covering how to understand the search audience, find trending content, analyze content across platforms, optimize existing content, and compare formats.

Scaling SEO Content Updates With Claude Code
Source: Search Engine Land

  • The author details a 14-step process for diagnosing content decay and making targeted updates, starting with a locked 56-day Search Console window and tagging every section keep, fix, remove, or add to protect existing rankings and schema.
  • On an Antalya Airport transfers page, the update moved every metric the right way against the locked baseline: clicks up 23.88% per day, average position from 14.89 to 10.87, CTR up 0.30 points, and query coverage up 6.28%.
  • The team mapped the 14 steps into Claude skills and a dedicated project preloaded with brand book, terms, and pricing so batch runs enforce judgment rather than replace it, cutting a two-day-per-page manual job across thousands of pages in nine languages.

Google's Open Knowledge Format Adds Five Trust Signals
Source: Search Engine Journal

  • Google released Open Knowledge Format version 0.2, adding five trust features that let a consumer verify aspects of OKF bundles:
    • Provenance via a sources field
    • Trust via generated and verified fields
    • Freshness via stale_after
    • Lifecycle via status
    • Attestation via a new Attested Computation type
  • The sources field records signals like author, usage_count, and last_modified rather than a credibility score, letting the consuming agent or LLM infer credibility, with per-claim attribution via markdown footnotes.
  • The new Attested Computation type defines a sanctioned way to compute a value and a means to verify the calculation ran that way, and Google updated its GitHub repository and published an explainer.

Google's Mueller: Fix Conflicting Metadata, Don't Test It
Source: Search Engine Journal

  • John Mueller said Google has no publicly defined order of precedence for reconciling conflicting metadata, using in-stock availability that differs across on-page content, structured data, and product feed as the example.
  • Mueller advised fixing the conflict rather than analyzing which signal wins, and noted dates (visible date, dateModified, sitemap lastmod) are a common version of the same problem with different weights and filters that change over time.
  • Availability drifts easily because a site can serve one value in initial HTML, another after JavaScript, and a third in a Merchant Center feed, so aligning the three is more reliable than guessing.

Google's Illyes Unsure on Shifting unavailable_after Dates
Source: Search Engine Journal

  • Gary Illyes said his gut feeling is that pushing an unavailable_after date forward is fine, but Google must crawl the page again to see the new date and he would have to check the renewal case.
  • The question came from a classifieds site with roughly 5,000 core URLs and about 10,000 new listings a month, most live only 24 to 72 hours, wanting the tag to avoid 404s and manage crawl budget.
  • Illyes said the tag only affects index selection as a "you can drop this now" signal and that Google would download the page rather than use a cheap HEAD request, and Google's docs still do not cover a date that changes on every renewal.

Organic Search & Algorithm Updates

How to Prevent SEO Scope Creep From Eating Your Budget
Source: Search Engine Land

  • The author defines scope creep as work added after a scope is agreed without matching cost or timeline changes, and argues SEO is especially prone because deliverables like a "technical audit" are rarely defined precisely.
  • The piece lists concrete prevention steps:
    • Define countable deliverables instead of outcome language
    • Separate strategy and execution as priced line items
    • Build a one-page change order process before you need it
    • Price for value rather than hourly
    • Set communication boundaries and audit scope quarterly
    • Say no in writing and route requests to the change process
  • The author warns that a moderate amount of unbilled creep can cut a healthy project margin by more than half, and that a tight change process cannot rescue an engagement that was underpriced from the start.

European Search Strategy Goes Beyond Google and Bing
Source: Search Engine Journal

  • Europe is many fragmented markets: Google sits near 80% in Germany but Bing holds a genuine 10% (higher on desktop), Ecosia is growing, and Czechia's Seznam.cz holds about 12%.
  • Regulation reshapes visibility, with a DMA Article 6(11) decision forcing Google to share ranking, query, click, and view data with rivals starting January 2027 and a Munich court finding Google liable for false AI Overviews claims.
  • Marketplaces and comparison engines like Amazon, Zalando, and Idealo absorb many product queries before Google sees them, so listing titles, attributes, and review counts now do real SEO work.
  • The author flags a tokenization problem where German compounds and agglutinative Turkish and Hungarian fragment poorly in English-trained tokenizers, and warns consent banners silently block schema and AI-referral measurement, with server-side tagging as the real fix.

Microsoft Earnings: Bing Search Ad Revenue Up 10%
Source: Search Engine Roundtable

  • Microsoft reported search and advertising revenue up 10% ex-TAC (9% in constant currency) for fiscal Q4 2026, down from 12% the prior quarter.
  • Search was the primary growth driver in More Personal Computing, which fell 4% overall as Windows OEM dropped 7% and Xbox 10%.
  • Total revenue was $90 billion (up 18%), Azure surpassed $100 billion for the year, and Microsoft 365 Copilot passed 30 million paid seats.

Bing Testing Pricing on Product Images
Source: Search Engine Roundtable

  • Microsoft appears to be testing showing pricing on product images within Bing organic results, where it normally shows review stars with the price below the image.
  • Barry Schwartz also spotted two rows of product images under a single search result snippet, which looks new.
Published on
geoai searchtechnical seoorganic searchdigest