SEO Recap covering October 22, 2026: AI search ROI studies and Pichai's agentic vision

Built by Stephanie Chung·7 min read·8 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)

Marie Haynes analyzed 12 months of GA4 data across 100 sites and 145 million sessions, finding ChatGPT sessions were just 0.62% of Google organic volume. ChatGPT visitors converted at 13.8% versus 3.02% for Google organic, but Google still drove 96.5% of total conversions and 27 times more conversions than all LLMs combined.
SEO Works found organic search delivered 39 times more traffic than AI assistants, and for every 100 conversions from organic search AI referrals delivered 3.2. AI referral traffic converted worse than paid search (0.67x) and better than direct (2.48x).
Kevin Indig reported that AI-driven direct traffic to longtail pages was not detectable across 12 sites, with no reliable link found between AI visibility and branded search or direct traffic.
Rand Fishkin summarized the 3 studies in a LinkedIn video, with the shared takeaway that AI search sends a tiny fraction of traffic and conversions compared with Google.

Google finished rolling out the AI Overviews and AI Mode opt-out to every site worldwide on August 31, following a UK CMA conduct requirement. The Search Console setting removes content as an input to AI responses without being used as a ranking signal elsewhere.
Supporting evidence on click loss: Pew found users clicked a traditional result in 8% of visits when an AI summary appeared versus 15% without, and a Penn and Northeastern field experiment found routing searches into AI Mode for a week cut external click-through by 18.8 percentage points.
Conversion data is split: Microsoft Clarity found AI referrals were under 1% of traffic but converted to sign-ups at 1.66% versus 0.15% from search, while a Marketing Science paper on 973 ecommerce sites found ChatGPT referrals converted ~13% below organic search.
The opt-out does not affect model training (handled separately via Google-Extended), and the CMA's page-level controls are due March 2027, which would turn an all-or-nothing choice into a page-by-page one.
Rebuild content to settle questions rather than capture queries, organize sites around decisions instead of keyword silos, and separate counted referral conversions from inferred mentions in measurement.

Sundar Pichai told Alex Heath that Google envisions an agentic layer that unifies Search, Gemini, and AI Overviews rather than merging the products, saying the near-future vision is "one box."
Enterprise and consumer agents share the same model, harness, and runtime, with each task provisioned a secure micro VM running 24-7 with real-time monitoring and auditability.
Pichai said the share of agentic queries in Gemini has steadily increased, a user-driven trend Google plans to bring into Search "in a pretty deep way" and widen beyond Ultra and Pro users over the next few months.

OpenAI says ChatGPT Ads reached a $1 billion annualized revenue run rate less than 200 days after launch, with more than 100,000 advertisers, though it disclosed no CPA, ROAS, or conversion-rate benchmarks.
The ad selection system runs on separate models and code from the one writing answers, so choosing an ad does not feed back into ChatGPT's response and a late ad simply does not appear.
An "orchestrator" model turns conversation context into search queries and embeddings, checks eligibility (location, policy, budget, pacing), and submits candidates to a separate auction that handles final ranking and pricing.
Winning the auction does not guarantee an impression, since a post-auction relevance check or delivery timing can still block an ad. OpenAI logs these stages internally but did not say how much advertisers can see in reporting.

Gary Illyes said robots.txt under RFC 9309 is an access preference protocol that governs which URLs a crawler may request and does not dictate what a system does with data after crawling. To block one AI crawler, you must target its specific user-agent token.
A fictional directive like No-AI-Chicken-Training would be ignored by the parser, since Google only supports User-Agent, Allow, Disallow, and Sitemap, and treats invalid lines as ignored.
Google-Extended is a token with no user agent of its own that controls whether crawled content trains future Gemini models and grounds Gemini Apps, without affecting Search inclusion or ranking.
An IETF draft Illyes co-wrote (dated Aug. 19) would add a Content-Usage rule and matching HTTP header, with a train-ai label, but it remains a Working Group document not yet submitted to the IESG.
Cherry Prommawin advised percent-encoding non-English and emoji paths in robots.txt, since Google treats raw and encoded paths the same but older or other crawlers may misread raw characters.

Pichai said AI agents will access content 2 ways, through human-like browsing and through agent-specific protocols, and that protocols will scale better but take longer to develop.
Google is backing the Universal Commerce Protocol (UCP), an open standard for agent commerce that merchants can adopt to enable buying in AI Mode and Gemini, with a Merchant Center cart-transfer option already available and lodging and food features marked coming soon.
Google lists Google-Agent as a user-triggered fetcher for agents on its servers, meaning it generally does not follow robots.txt directives, and Google publishes the IP ranges these fetchers use since user-agent strings can be spoofed.
Asked how Google protects its ads business when agents browse, Pichai said the answer is to put users first and let the market reach a shifting equilibrium, without addressing how ads would work with agents.

Google is testing swapping the positions of the company name, ratings, description, and action button with the business photos in local knowledge panels, bringing the layout closer to its current mobile design.
The bottom section is tightened in the new design. Spotted by Ben Fisher, the test raises questions about how the reordering could affect clicks, calls, and directions to businesses.

Comparison content, roundups, and listicles still work for SEO and conversions when built for users rather than to game an algorithm. Scaled, authority-abusing versions are what spam updates target.
Smaller brands can get included in creator comparison videos that already rank for big competitors, buying exposure to those competitors' funnels through media buys, affiliate deals, or PR.
Being featured in roundups like SEO Roundtable or SEOFOMO drives crawling and discovery, referral traffic, and audience building when the publisher promotes to its own subscribers.
Google does not hate affiliate links or listicles. Lead with user intent and quality to grow naturally rather than treating the format as a ranking silver bullet.

Published on October 23, 2026