SEO Recap covering July 30, 2026: AI Overviews near half of searches and Similarweb's AI search data

Built by Stephanie Chung·7 min read·10 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)

Turning news articles into liquid content for AI search
Source: Search Engine Land

  • Nikita Roy argued at ONA25 that the article is no longer the unit of journalism in an AI-mediated world, and the piece pushes "liquid content" that adapts to a viewer's context, location, and format preference.
  • For AI surfacing and citation, the author recommends concrete structure:
    • Use the inverted pyramid and avoid burying the lede
    • Add NewsArticle structured data
    • Include bullet-point summaries at the top of long pieces
    • Use subheaders and highlight key questions or quotes
    • Leverage internal linking for topic authority
  • The Future Newsrooms Study 2026 found 64% of newsrooms still build stories around channel destination versus 21% around audience preference.
  • Monetization angles include a 2025 FT Strategies "journalism as a service" model licensing data via APIs, plus tools like Google's Gemini Notebook for multimodal reformatting.

Similarweb data shows AI search layering on top of Google, not replacing it
Source: Search Engine Journal

  • Similarweb found 461 million of ChatGPT's 494 million users (95%) also used Google in the same window, and search still draws 3.3 billion monthly unique visitors versus 655 million for AI chatbots.
  • Only 6.8% of U.S. ChatGPT answers included an external link as of May 2026, up more than fivefold from ~1% a year earlier, meaning 93 of every 100 answers send no one anywhere.
  • Aleyda Solis notes 65% of URLs ChatGPT cites sit 2 or 3 folders deep, while 58.8% of AI referral traffic lands on the homepage, so cited pages and clicked pages are largely different URLs.
  • 3 changes help:
    • Split reporting into a citation-rate KPI and a separate referral and conversion KPI
    • Stop treating AI visibility as one category and pull your own category leaderboard
    • Match content to each platform's actual audience rather than its overall size

Entity mapping works on Google but does little inside language models
Source: Search Engine Journal

  • Entity mapping has a real target on Google, whose Knowledge Graph is a curated object built from structured data and independent corroboration, and Google's AI answers resolve entities against that same graph.
  • A language model's parametric memory holds no node, row, or panel for a brand. Knowledge is a distributed statistical pattern learned from how the wider web describes you, so sameAs links and schema barely move it.
  • Interpretability research cited in the piece shows entity knowledge and relational knowledge live in different parts of a model and do not map onto the subject-predicate-object triples schema produces.
  • Measure outputs directly by tracking whether you appear when a model answers from memory versus from a live fetch, since schema-to-citation correlations reflect well-sourced pages, not causation.

How to measure marketing when AI owns discovery
Source: MarTech

  • Traffic is a weaker proxy for awareness as buyers research in AI conversations without visiting sites, so teams should track demand, engagement, and assisted conversions instead of page views.
  • Recommended signals to monitor:
    • Direct traffic, social mentions, and brand-name search volume as delayed AI-driven interest
    • Share of voice across Reddit, YouTube, and LinkedIn where models source citations
    • Multi-touch attribution over 30- or 90 day windows
    • Returning-visitor ratio and pages per session
    • Downstream intent like pricing calculators and comparison pages
  • As AI filters top-of-funnel discovery, shrinking traffic paired with rising repeat visits and deeper consumption signals a site is serving more qualified buyers.

AI in Search / AI Overviews

Why Google's generative AI Search Console data misleads marketers
Source: Search Engine Journal

  • The AI Overview and AI Mode impression data added to Search Console shows presence without click numbers or query detail, so impression growth can hide a collapse in traffic, leads, and conversions.
  • Google's single-position rule assigns position one to every URL inside a top AIO block, so a link buried in an accordion earns the same rank as a featured citation, inflating rank numbers.
  • Averaging an AIO position one with an organic rank 4 into a 2.5 average position merges distinct SERP features and hides real ranking drops.
  • Prioritize organic revenue, leads, and brand citations, replace GSC CTR averages with first-party analytics, and adopt median position over average position.

AI Overviews now appear on nearly half of Google searches
Source: Search Engine Roundtable

  • Similarweb data puts AI Overviews on ~43% of all queries and Semrush puts it near 48%, up from ~13% to 20% a year earlier.
  • Informational queries trigger AI Overviews at even higher rates, and AI Mode visits are growing faster still.

Google blames source websites for a race-based AI Overview answer
Source: Search Engine Roundtable

  • Users flagged that AI Overviews gave differently framed responses to near-identical queries about wanting a white versus a black doctor.
  • Google responded on X that the response was pulling particular web pages for that specific search, said it did not appear in the Gemini app, and said teams would work to improve it.
  • Barry Schwartz notes the irony that Google claims its results, including AI Overviews, are works protected under the Copyright Act while blaming source websites for this answer.

Technical SEO

A Python keyword clustering tool using TF-IDF and HDBSCAN
Source: Search Engine Land

  • A refactored clustering script groups keywords into topics using TF-IDF vectorization fed into HDBSCAN, a density-based algorithm that does not require setting the number of clusters in advance.
  • HDBSCAN labels outliers as -1 and excludes long-tail queries that do not fit any topic, producing cleaner clusters than k-means.
  • The recommended data source is the Search Console BigQuery export rather than the UI, which caps at 1,000 rows and is sampled.
  • The workflow uploads a .txt keyword list, cleans the text, and exports an Excel file with both grouped clusters and a keyword-by-keyword breakdown, with tunable sensitivity and min_cluster_size parameters.

Structuring multi-location SEO without geographic page bloat
Source: Search Engine Land

  • Use the smallest number of location pages needed to represent the business, since excess city pages cause internal competition, fragmented authority, conflicting local info, and can resemble Google's doorway abuse.
  • 4 often-conflated concepts:
    • Physical locations, which deserve authoritative pages
    • Regional markets, which need a hub only when they add context
    • Service areas, which do not automatically warrant a URL
    • Cities the brand wants to rank in, which is a goal not a page type
  • Reusable template content is fine. The vulnerability is a template with no local differentiation, so variable content should carry address, hours, staff, directions, local photos, and FAQs.
  • Use one authoritative URL per location, applied consistently across navigation, GBP, and structured data, with crawlable HTML links rather than JavaScript-only location finders.

Organic Search & Algorithm Updates

Selling SEO and AI services when incentives reward the signature, not the result
Source: Search Engine Land

  • The core problem is incentive design. Sales reps are rewarded for closing, renewing, and shortening cycles, not for whether the SEO or AI work actually delivered.
  • AI has made expectation-setting worse because brands sell AI visibility with even less control over when a name appears in ChatGPT or AI Overviews, using prompt screenshots, third-party visibility scores, and dashboards as proof.
  • What reps actually need before pitching:
    • Clear boundaries on what can never be guaranteed
    • Qualification criteria and approved claims
    • Defined customer responsibilities and realistic case studies
    • A feedback loop and early access to delivery specialists
  • Tie part of sales compensation to account quality or early retention, since an engagement that collapses within 90 days is delayed churn, not a sale.
Published on ·Last updated on
geoai searchtechnical seoorganic searchdigest