SEO Recap covering September 11, 2026: AI Overviews cutting Wikipedia traffic and longer queries

Built by Stephanie Chung·6 min read·7 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)

MJ Cachón's August dataset ran 189 branded prompts through ChatGPT and observed 1,797 sub-queries no user typed, with quote usage climbing 25-fold between a run's first and last search and sub-queries averaging 7 words each.
Google has said ~15% of daily queries are ones it has never seen before, a figure Mueller confirmed at Search Central Live NYC in March 2025 has stayed steady despite LLM adoption.
3 habits to target the long tail:
- Build opening paragraphs and H2s around longer nested phrases, using Search Console high-impression low-click queries to find variants
- Publish at the speed of news via press releases or rapid-response pages to own new phrasing first
- Write literal answers as standalone, quotable sentences that AI systems can lift whole to verify claims

LightSite AI's Stas Levitan argues mentions, citations, and share of voice are benchmark data that show what might happen, while first-party site data shows what actually happened, and confusing the 2 wastes budget.
The framework tracks 4 signals across the funnel:
- AI bot traffic as a site-level discovery impression
- Pages consumed by AI bots to reveal content priorities
- Human visits from AI search referrals
- AI click-through rate comparing machine attention to human demand
Across the dataset, ~12% of pages absorbed about half of bot impressions, and a small set of pages reread over a 4- to 6-week window drove a disproportionate share of crawl volume.
About one-third of sites in the dataset blocked at least one major AI bot, often from misaligned security, CDN, and marketing settings, so start with a technical crawl-access check.

A University of Washington working paper (Khosravi and Yoganarasimhan, updated September 2, not peer reviewed) estimates default AI Overviews reduced monthly external-search referrals to English Wikipedia by 5.45% versus German and 4.82% versus French.
The design uses AI Overviews' May 2024 US default rollout as a timing break, comparing 499,927 English-German and 530,873 English-French matched article pairs from December 2023-December 2024 with a Poisson pseudo-maximum likelihood difference-in-differences model.
The estimate works out to ~100.27 million fewer search referrals per month and ~1.20 billion annually. An earlier version using daily pageviews had reported a ~15% decline before switching to monthly referrals and adding controls.
Separately, Wikimedia reported overall human pageviews across all languages fell ~8% year over year, and Pew found only 1% of visits to pages with an AI summary led to a click on a cited source.
Google disputes that the combined external-search metric can isolate AI Overviews, while the authors note other engines carry only a small share of traffic.

Google Search Console page indexing report missing June 2026 data
Sources: Search Engine Land · Search Engine Roundtable

The GSC page indexing report is missing several days of data from June 2026 across all profiles, a reporting glitch rather than an actual Search indexing problem.
John Mueller said the gap traces to a period in June when data was delayed and the report was never updated. Google does not back-fill indexing data, so the data appears permanently lost.
The issue follows other recent GSC glitches, including a separate indexing reporting hiccup earlier in the week and a spinning-wheel bug Marie Haynes saw in the Search performance report.

Many audits amount to a crawler export or a generic list, yet cost between $2,500 and $20,000+. A real audit explains which issues matter, which are intentional, and how findings map to business goals.
SEO audits still matter for GEO because most LLMs lack their own indexes and ground answers via retrieval and trusted search results, so crawlability, indexability, and content quality carry over to AI visibility.
Add contract clauses granting a full refund if the deliverable is only a crawler export with no manual discovery, or if it matches raw LLM output with nothing found beyond it.
Buy or trial crawlers like Screaming Frog or Sitebulb to scan your own site first, and use an LLM as a second opinion while knowing it can flag intentional robots.txt blocks or thin pages that should stay.
Get deliverable expectations in writing so page count padded with images and charts does not stand in for actual analysis.

Comparing January 2025 with August 2026, impression share for 1-2-word queries fell from 42% to 24%, while 3-4-word queries rose from 33% to 48% and became the new center of gravity for impressions.
Conversion share for 1-2-word queries collapsed from 62% to 52%, while 3-4-word queries jumped from 20% to 46%, 5-6-word queries tripled from 3% to 9%, and 7+-word queries quadrupled from 1% to 4%.
Conversion rate in the 3-4-word bucket rose steadily each quarter while the 1-2-word bucket's rate declined, pointing to higher intent moving down the long tail.
3 takeaways for advertisers:
- Make long-tail and conversational query targeting a core pillar, not an experiment
- Match ad copy and landing pages tightly to the specific query language
- Reallocate budget away from broad, high-cost head terms toward higher-converting long-tail segments

Google Analytics Dashboards give a drag-and-drop, grid-based canvas where users position, resize, and align visualizations and drop dimensions and metrics directly onto charts.
6 visualization types ship at launch:
- Scorecards for high-level KPIs
- Tables for detailed reporting
- Line charts for trends
- Bar charts for comparisons
- Donut charts for breakdowns
- Funnel charts for conversion steps and drop-offs
Standard properties allow up to 15 cards per dashboard and premium up to 30. Creating and publishing requires an Editor or Administrator role, and published dashboards are shared property-wide via the Reports navigation.
API support, segments, and card-level comparisons are not currently supported.

Published on September 12, 2026