StephEO

SEO Recap covering July 20, 2026

Built by Stephanie Chung·9 min read·12 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)

How AI-driven personalized search changes SEO strategy
Source: Search Engine Land

  • The piece argues AI systems (AI Overviews, AI Mode, ChatGPT, Perplexity, Claude) now tailor answers to the individual, so two people asking the same question get different responses.
  • The author says search and social are converging, with LLMs pulling from YouTube, Reddit, LinkedIn, X, TikTok and reviews, so brands should manage a portfolio of searchable assets rather than just HTML pages.
  • Recommended tactics include descriptive alt text, video transcripts with speaker attribution, structured data, and original research AI can cite.
  • Google's Preferred Sources and Discover Follow features are highlighted as ways to build loyal audiences; Google previously said people are twice as likely to click through to a Preferred Source.

Google.com is now AI Mode's second most-cited domain, Profound finds
Source: Search Engine Journal

  • Profound reports google.com's citation share in AI Mode rose 8.4x between April 15 and June 30, driven almost entirely by Business Profile cards and Product Knowledge Panels.
  • Local queries pull citations from Business Profile cards (biggest movement in hospitality/travel, home services, restaurants, real estate, healthcare); product queries surface Product Knowledge Panels instead of retailer or brand links.
  • Earlier data cited by SEJ found Google made up 17% of 1.3 million AI Mode citations (about 20% with YouTube), and 43% of AI Overview responses referenced Google's own results.
  • The takeaway: these cards are fed by inputs you control (Business Profile data, product feeds), so optimizing them is more tractable than sitewide work, though it only secures the citation, not the click.

AI answers about business locations are often wrong, tests find
Source: Search Engine Journal

  • A Searchable test found AI tools returned at least one false fact about 64% of UK high street retailers, with wrong postcode the most common error (one in 10 answers, even when the town was specified).
  • In 13,365 questions across 165 London businesses, 93% had at least one basic fact wrong or missing; smaller businesses drew errors more often (50%) than larger ones (32%).
  • By platform, Perplexity gave inaccurate answers 10% of the time versus 5% for Gemini and 4% for ChatGPT, so checking one tool tells you little about the others.
  • Recommended audit: run a fixed question list across AI Overviews, AI Mode, ChatGPT, Gemini and Perplexity, sort factual errors from reputation issues, then reconcile Google Business Profile, website and location pages and pursue corrections at cited third-party sources.
  • Context stats: Whitespark found AI Overviews on 15% of direct local-intent searches but 92% of informational ones; SOCi found ChatGPT recommended 1.2% of 350,000 locations; BrightLocal found 45% of consumers now use AI tools for local recommendations, up from 6%.

How marketers can use the Model Context Protocol (MCP)
Source: Search Engine Journal

  • The author describes MCP as an open standard (likened to a USB-C port) that connects AI assistants like ChatGPT, Claude, Gemini and Copilot to your files, analytics, CMS and search data via a one-time, code-free connection.
  • MCP does not make a model smarter; it lets the assistant pull live data and act, so connecting search/AI-visibility platforms, web analytics, CMS and CRM turns generic advice into brand-specific priorities.
  • Example prompts by team: SEO asking which pages slipped in position, content teams drafting briefs where citation share is under 15%, PR checking which channels shape AI answers.
  • Cautions: audit data quality first (MCP amplifies bad inputs), start with one trusted source, scope permissions to read-only, and treat outputs as a first draft to verify.

AI in Search / AI Overviews

Google says AI Search features send billions of clicks weekly
Source: Search Engine Roundtable

  • Google search head Nick Fox said on LinkedIn that AI Mode and AI Overviews now send billions of clicks to websites every week, and that overall Search sends billions of clicks daily and is growing.
  • Liz Reid reshared the claim; Google says it designed AI features to connect people to websites but stops short of saying it sends more clicks than before.
  • Schwartz counters with a zero-click study showing Google sends only 27.6% of clicks to the open web, plus Pew data and a Press Gazette report projecting Google search traffic will halve by Q3 2027.
  • He notes Google built AI performance reports in Search Console but excluded click data, and suggests Fox's post timing responded to a forthcoming NY Times story.

Technical SEO

Which technical SEO tasks are usually a waste of time
Source: Search Engine Land

  • The author argues technical SEO work should meet at least one test: it drives revenue, enables other revenue work, or prepares the site for AI-driven search.
  • Tasks flagged as low-value on most sites: pushing Core Web Vitals further once already in the "good" range, cleaning up redirects without chains or loops, crawl-budget work when important pages are already crawled, chasing every 404, removing meta keywords, and routine disavow file maintenance.
  • The piece warns against clearing every SEO tool alert, since tools lack context on intentional robots.txt blocks or 404s.
  • The recommendation: tie each task to a measurable business outcome (for example, faster load times to lift ecommerce conversion rate) before spending scarce developer time.

Google says many Search Console "errors" aren't real problems
Source: Search Engine Journal

  • John Mueller and Martin Splitt say site owners should treat Search Console as a tool for spotting patterns and trends, not a checklist of issues to fix.
  • Splitt acknowledges Google's own UI and notification emails, plus the "mark as fixed" button, wrongly imply every reported item needs action.
  • Many "not indexed" pages and 404s are normal and expected; a 404 is labeled an error because the request for a missing URL fails, not because the site is broken.
  • When indexing drops have no technical cause, Google advises stepping back to assess overall site quality rather than chasing individual reports.

Why pages get stuck in "Crawled - currently not indexed"
Source: Search Engine Journal

  • The author finds most stuck pages are "commodity content" that rehashes what already exists rather than offering first-hand experience or unique knowledge, which is what Google says it wants to index.
  • Technical causes are rarer but real: one migrated site blocked CSS/JavaScript via a robots.txt rule (Disallow: /?), so Google saw empty pages; a live URL test in GSC confirms whether Google can see the content.
  • At a Toronto Google event, a presenter said AI has lowered the content-creation threshold, and Google sometimes indexes a page temporarily to test whether it "produces happier users."
  • Mueller confirms that when Google has strong quality concerns about a site, it crawls and indexes less, so large-scale non-indexing is a signal to reassess overall quality, not a technical bug to patch.

Organic Search & Algorithm Updates

EU orders Google to share anonymized search data with rivals
Source: Search Engine Journal

  • The European Commission adopted two binding Digital Markets Act decisions: Google must share anonymized query, click, view and result-position data on fair, non-discriminatory terms, and open Android features to rival AI assistants.
  • Ranking algorithms and the FastSearch grounding system are excluded; sensitive data like account details, timestamps and rare queries are suppressed, and data can't be used to train general AI models or replicate Google's results.
  • Eligible providers need at least 50,000 monthly EU users plus an operating-history or investment test; Bing and DuckDuckGo are seen as likely to qualify quickly.
  • Timelines are long: Google's pricing proposal is due by January 2027, main Android changes by August 1, 2027, and concurrent voice activation by August 1, 2028; SE Ranking notes AI platforms were about 0.24% of global traffic in January.

Unconfirmed Google ranking volatility around July 18-19
Source: Search Engine Roundtable

  • Barry Schwartz reports signs of another unconfirmed Google ranking update over the weekend of July 18-19, with a spike on Saturday July 18 and Sunday July 19 looking even more heated.
  • Some in the SEO community say the earlier "7-Eleven" July 11 movement is now reversing; forum members report swings including one site's Google traffic up 400% with no changes made.
  • The last confirmed updates were the June 2026 spam update (June 14-26) and the May 2026 core update (May 21-June 2).

Diagnose the four types of content decay before refreshing a page
Source: MarTech

  • The author splits content decay into four types with distinct GSC signatures: ranking decay (clicks, impressions and position all down), zero-click capture (clicks down but impressions and position flat or up), intent drift (position holds but SERP format changed), and demand decay (clicks and impressions down, position holds).
  • Cited data: fewer than one in three Google searches now sends a click to the open web, about 68% end without a click (up from ~60% two years ago), and the top organic result loses around 58% of clicks when an AI Overview appears.
  • Diagnosis method uses only Search Console and a spreadsheet: pull six months of monthly clicks plus a three-month year-over-year comparison of clicks, impressions and average position.
  • Warnings on 2025 impression data: the September 2025 removal of &num=100 and a logging error inflating impressions since May 2025 make year-over-year impression drops unreliable, so verify against the live SERP.
  • Fixes differ by type: real information gain for ranking decay, unique tools/data or reallocation for zero-click capture, format re-matching for intent drift, and consolidation or pruning for genuine demand decay.

Branded search is becoming a weaker proxy for brand demand
Source: MarTech

  • The author's client portfolio shows branded customer acquisition costs average 76.6% lower than non-branded, yet branded search demand fell 11.1% over the past month despite a stable auction environment.
  • After ruling out weaker preferences, lower engagement and macro pressure, the piece attributes the gap to AI search: AI Overviews rose from 57.2% of commercial searches for one client's core keyword to 95.9% by June 2026.
  • The argument: AI now answers and compares options before users run a branded search, so branded search captures a smaller share of the decision, making it a measurement problem rather than a demand problem.
  • Recommendation: keep branded search as one indicator but weight unaided awareness, share of search and brand conversion rate more, and invest in positioning, original research and thought leadership that AI systems reference.
Published on
geoai searchtechnical seoorganic searchdigest