SEO Recap covering August 11, 2026: ChatGPT's in-house index and GSC's AI report going live

Built by Stephanie Chung·5 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)

Resoneo read 1,249 ChatGPT answers captured in July and found that OpenAI's in-house index (tagged "labrador") served pages the same way whether or not the publisher had a content deal, with identical format, length, and freshness.
In Resoneo's paid thinking-mode sample of 16,407 search results, the split was:
- ~75% scraped from Google
- ~24% from OpenAI's in-house index
The index stores a title and ~200 characters from the page. Across 463 pages with an H1, 387 snippets (83.6%) began with the H1, and template elements above the first paragraph (kickers, dates, alt text) eat into that limited window.
The work backs Suganthan Mohanadasan's July correction to his earlier "licensed tier" reading. Around July 21 OpenAI stopped tagging results with the fetching pipeline, the signal both investigations relied on.

Traditional metrics like clicks, CTR, and rankings only capture the outcome of a decision already made, a blind spot widened as AI Overviews and conversational search push more evaluation off the website.
"Decision Distance" measures the semantic gap between a user's underlying decision drivers (functional, emotional, social) and the messages a brand communicates across the journey.
The 4 steps to calculate it:
- Identify decision drivers from customer language (value for money, trust, convenience, social proof, quality)
- Measure how strongly your own messaging reflects those drivers
- Compare the 2 profiles using sentence embeddings and semantic similarity
- Reduce the gaps through messaging and content changes
Treat it as a complement to traditional metrics, not a replacement, and use search queries plus social listening, interviews, support chats, and CRM data to build the audience profile.

A TransUnion survey of 100 senior marketing and tech leaders found a "confidence-readiness paradox": 89% expect AI marketing investment to grow and 64% are confident of hitting AI goals, but only 42% rate people readiness high and 36% rate data and process readiness high.
Just 48% say they have enough platform-level AI visibility to optimize confidently. 69% cite walled garden blind spots and 70% cite cross-channel blind spots limiting measurement of AI's impact.
AI search citation tracking is framed as the search-specific piece of a company-wide measurement failure CMOs already recognize, not a standalone SEO tooling problem.
Concrete steps for SEO teams:
- Stop pitching AI search visibility as a separate line item and tie it to the CMO's known measurement gap
- Audit whether customer data is connected before buying another AI visibility tool
- Build incrementality testing to show lift tied to AI citations, not just citation counts

The generative AI performance report appears to be fully rolled out, showing as an expandable tab under the main performance report across sites with enough impressions in AI features.
It reports impressions, pages, countries, devices, and dates for URLs appearing in generative AI features in Search and Discover, but still includes no click or query data.
Sites without enough impressions in generative AI features will not see the report.

Anthropic will embed invisible text watermarks in new Claude models worldwide and attach C2PA signed metadata to generated .svg, .png, and .jpg files, applying across the API, Claude apps, Claude Code, and cloud partners.
A detected mark only shows Claude may have processed the content, not that it authored it. Proofreading, translation, or summarizing someone else's text can still trigger a mark.
The move fulfills the EU AI Act's Article 50(2) transparency code, which had ~190 signatories by late July including Google, Meta, Microsoft, Mistral, and OpenAI.
GPTZero's Alex Cui and researcher Jonas Geiping both said heavy paraphrasing can strip watermarks, and Anthropic has not yet published its detection mechanism.

Running one entirely human article through several AI detectors returned wildly different verdicts, from 100% AI to fully human, and pre-ChatGPT articles from 2014, 2019, and 2021 were also flagged as partly or mostly AI.
Stanford research found detectors disproportionately misflag non-native English speakers, and a 2026 ScienceDirect study found small human edits let genuinely AI-generated text bypass detection.
Several vendors sell a detector alongside a paid "humanizer" that beats detectors, monetizing writer fear rather than reliably verifying anything.
Evaluate content on accuracy, originality, real expertise, and usefulness instead of detector scores, which aligns with Google's stance of rewarding helpful content regardless of how it was produced. Indiana University's Kelley School bans AI detection tools as unreliable.

Martin Splitt's former X handle @g33konaut, cited across many SEO news sites, was picked up by a search engine indexing service after he let the account lapse during the Twitter to X transition.
The takeover mirrors expired-domain abuse and complicates the many published articles that linked to or cited Splitt's old posts.

Published on August 12, 2026