What we doAI VisibilityAtlasAboutInsightsContactTalk to us
Digest

The Great Decoupling: Rankings Are Up, Citations Are Somewhere Else

August 24, 2026 · Vikram Jayanand

Week of August 24, 2026 · SEO · GEO · AIO · Google Ads

The Big Picture

Google shipped its third spam update of 2026 and kept expanding AI-native reporting (Search Console's new Generative AI performance reports), while the citation landscape kept shifting under B2B marketers' feet: the overlap between top-10 organic rankings and what AI engines actually cite has collapsed to roughly 17 to 38%, and Claude's B2B referral share has grown more than 14x since the start of the year. The throughline this week: ranking well and getting cited by AI are increasingly two different jobs, and B2B teams need to budget for both.

That gap is the story worth sitting with. As recently as mid-2025, a page that ranked in the top 10 on Google had roughly a 70% chance of also showing up as an AI citation, so traditional SEO work was doing double duty. That's no longer true. AI engines are now pulling from a narrower, differently-weighted set of sources than the ones Google's organic algorithm rewards, favoring third-party authority and digital PR coverage over the on-page and technical signals that move organic rankings. Meanwhile, buyer behavior is shifting faster than most teams' budgets: over half of B2B software buyers now start vendor research inside an AI chatbot rather than Google, up from 29% a year ago, and Claude alone went from 1.4% to 20.5% of B2B AI referral traffic in eight months.

Put together, that means a site can be winning on rankings and still be invisible in the AI answers buyers are actually reading, or vice versa. The practical takeaway is that "SEO" and "GEO" now need separate tracking, separate tactics (E-E-A-T and earned coverage for citations, technical/on-page work for rankings), and separate line items in the plan, not one bucket labeled "search."

Google AI Mode & AI Overviews

Gemini 3.7 now powers parts of AI Mode

Google Search's AI Mode is now sometimes running on Gemini 3.7 Flash, part of a steady cadence of model upgrades feeding AI Mode and AI Overviews throughout 2026.

Search Console's Generative AI reports are rolling out slowly

Google's new "Search Generative AI performance reports" in Search Console (announced in June) give site owners dedicated views of impressions across AI Overviews, AI Mode, and AI features in Discover, broken out by page, country, and device. It's the first real visibility Google has given site owners into how they show up in AI features.

Key stat: The reports currently show impressions only (no clicks, CTR, or query-level data), and the rollout remains limited to a subset of UK site owners, with no confirmed date for global availability.

  • Useful for spotting which pages get pulled into AI Overviews, even without conversion data yet.
  • B2B teams with UK properties should check access now; everyone else should watch for the wider rollout.

Sources: Introducing Search Generative AI performance reports, Google Search Console Finally Shows AI Overviews & AI Mode Impressions, Search IO 2026 updates

GEO / AEO / AIO

B2B AI referral traffic is reshuffling fast

Claude overtook Perplexity in monthly referral sessions to B2B sites in March 2026, and on B2B brand sites Claude now drives 20.5% of AI referral traffic, up from just 1.4% eight months earlier, a 386% surge between January and July. Citation behavior still diverges sharply by engine: Perplexity averages 21.9 citations per response versus ChatGPT's 10.4, while Claude cites social sources far less often (3.99%) than AI Overviews (15.3%) or Perplexity (11.3%).

The ranking/citation gap keeps widening

Research from GEO firm Brandlight found the overlap between top Google organic links and AI-cited sources has fallen from about 70% in mid-2025 to below 20% in early 2026, meaning strong SEO rankings are no longer a reliable proxy for AI citation.

Key stat: More than half of B2B software buyers now start vendor research inside an AI chatbot instead of Google, up from 29% about a year ago.

  • AI search visitors convert at a significant premium: Semrush reports a 4.4x lift, Ahrefs an internal case study found 23x.
  • E-E-A-T signals and third-party digital PR coverage remain the strongest lever for earning citations, since LLMs weight external authority over on-page optimization.
  • Platform-specific tuning matters: what earns a citation on ChatGPT won't necessarily work on Perplexity, Gemini, or Claude.

Sources: AI search marketing: how B2B brands get cited in 2026, Claude GEO Statistics 2026, How ChatGPT, Claude & Perplexity Choose Brands to Cite

Traditional SEO & Organic Search

Google's August 2026 spam update finished rolling out

Google's third spam update of the year began August 18 and completed roughly 2.5 days later (finishing the morning of August 21), applying globally across all languages. Google framed it as stricter enforcement of existing spam policies rather than a new rulebook: no new policies were announced.

Key stat: Third-party trackers (AccuRanker, Mozcast, Semrush) logged multiple distinct volatility waves through August (roughly August 1 to 3, 5 to 6, and 12 to 13) even before the confirmed spam update, a reminder not to over-attribute every ranking swing to a single named update.

B2B organic CTR keeps eroding where AI Overviews appear

B2B technology queries now trigger AI Overviews 82% of the time (up from 36% a year ago), one of the steepest AI Overview growth curves of any industry tracked. Ahrefs' 300,000-keyword study found position-1 organic CTR falls from 1.41% to 0.64% when an AI Overview appears above the results.

  • Sites hit by the spam update should audit against Google's published spam policies before assuming a penalty: volatility does not equal a penalty.
  • Flat or declining organic traffic doesn't necessarily mean flat pipeline: several studies note AI citations are partially offsetting lost clicks with higher-intent visitors.

Sources: Google's August 2026 Spam Update Is Now Rolling Out, Google August 2026 spam update done rolling out, Why B2B Organic Traffic Is Dropping (But OK)

New Protocols & Standards

llms.txt adoption is growing, but Google says it does nothing for rankings

Adoption estimates vary wildly by cohort: a fixed 219-host panel put adoption at 51.8% as of August 3, while a broader Tranco top-1,000 tracker puts it at just 8.7% as of June. Google's own documentation, updated in June, states llms.txt has no effect (positive or negative) on Search rankings or AI Overviews; Search simply ignores the file. No major AI lab (OpenAI, Google, Anthropic, Meta, Mistral) has publicly committed to reading or acting on it in production as of Q1 2026.

FAQPage rich results are officially gone

FAQ rich results stopped appearing in Google Search as of May 7, 2026, and Search Console API support for FAQ rich results ends this month (August 2026). Remaining active rich-result types include Article, Product, LocalBusiness, BreadcrumbList, Organization, Review, and Recipe.

Key stat: llms.txt currently sees real usage from IDE agents and MCP servers (Cursor, Claude Code, Copilot, Windsurf), just not from the search engines and AI answer engines B2B marketers are actually trying to influence.

  • Don't deprioritize JSON-LD structured data for AI citations: Perplexity, ChatGPT search, Claude, and Google AI Mode all parse JSON-LD to help ground citations, even where llms.txt is ignored.
  • If you built an FAQ schema strategy around rich results, that traffic driver is gone; the schema itself may still help AI grounding, but not SERP rich results.

Sources: llms.txt in Practice: Adoption Data, Evidence, and Setup, The State of llms.txt in 2026, A 2026 Schema.org Cheatsheet for Google Rich Results and AI Citations

Google Ads Updates

Budget-constrained Target CPA/ROAS campaigns change August 17

Starting August 17, budget-constrained campaigns began optimizing more consistently toward the advertiser's indicated target even as budgets change, following the July 6 rollout of Google's new Bid Target Adjustment Tool.

ChatGPT is a bigger paid-traffic source from Google than most realize

An iPullRank analysis of 13.1 billion search events (9.1M opted-in users, October 2024 to December 2025) found ChatGPT ranks sixth among destinations clicked from Google Search (behind YouTube, Google's own properties, Reddit, Facebook, and Wikipedia), but stood out for an unusually high share of paid clicks specifically.

Key stat: Demand Gen's auto-generated video feature now builds ready-to-run video ads directly from uploaded product screenshots and logos, no production team required, which lowers the bar for B2B teams without video budgets.

  • AI Max for Search continues expanding rollout; combine with Performance Max reporting improvements to see channel-split impact if running both.
  • For B2B prospecting specifically, Demand Gen remains better suited to targeting by job title/company size across YouTube, Gmail, and Discover than PMax's broader automation.

Sources: Google Ads Is Changing Target-Based Bidding: August 2026 Update, ChatGPT draws more paid clicks from Google than any other top destination, Google Ads Updates In 2026

Suggested Actions This Week

  • Check whether your UK properties (or agency clients) have access to the new Search Console Generative AI performance reports, and start baselining AI Overview/AI Mode impressions now, before click data arrives.
  • Audit any content built around FAQPage schema: the rich-result payoff is gone, but keep the JSON-LD in place since AI engines still parse it for citation grounding.
  • Don't treat llms.txt as a ranking or citation lever; if you publish one, do it for agentic/IDE tooling use cases, not SEO or AI Overview visibility.
  • Pull your own AI referral mix (GA4 or an AI-visibility tool) and check whether Claude's referral share has grown on your site the way it has industry-wide, it may warrant its own GEO tuning pass separate from ChatGPT.
  • If you were hit by ranking volatility this month, cross-reference the dates against Google's confirmed August 18 to 21 spam update window before assuming causation.
← Back to Insights