Four new insights, zero new operators, three pattern updates. Two themes organized the day: the AEO citation universe has split into two distinct populations, and verification is the rate-limiting step in every domain AI has accelerated. The reasoning-mode citation gap surfaced from Kevin Indig's research and gained structural support from Mike King's rebuttal of Google's official AI search guidance. The verification constraint appeared in Anthropic's Glasswing security data and in Florin Tatulea's MCP outbound deployment rule.
Theme 1, Two citation populations, one AEO strategy
Kevin Indig's May 18 research (High-reasoning AI mode and standard mode share only 25.6% of cited domains; AEO built for one misses three in four sources the other cites) is the most concrete data point the AEO corpus has seen on reasoning-mode behavior. High-reasoning AI mode fires 4.6x more queries per answer and cites 99 domains standard mode never touches. The overlap between the two populations is 25.6%.
"The brand that wins under minimal reasoning is not the brand that wins under high reasoning."
This matters for The AEO triangle, presence, relevance, manual-action propagation because the three-layer framework (presence, readiness, business impact) was built around a single AI-search population. The reasoning-mode finding adds a fourth question: which population are you present in? The content formats that earn reasoning-mode citations skew toward primary research, expert interviews, and structured evidence. Not the optimized long-form guides most AEO practitioners build for. Mike King's May 18 rebuttal (Google's AI optimization guide contradicts its own technical research in four places; calling it 'still SEO' is a rhetorical strategy, not a neutral description) adds structural support: Google's official AI optimization guide contradicts its own technical research in four specific places, and the "it's still SEO" framing is a deliberate rhetorical strategy rather than a neutral description. The practical implication: use the 5-gatekeeper mental model (Planner, Retriever, Reader, Critic, Synthesizer) over Google's guide, and audit separately for both reasoning-mode and standard-mode citation presence.
- Kevin Indig, High-reasoning AI mode and standard mode share only 25.6% of cited domains; AEO built for one misses three in four sources the other cites. Research showing a 74.4% non-overlap between high-reasoning and standard AI citation populations, with a concrete content-format prescription for closing the gap.
- Mike King, Google's AI optimization guide contradicts its own technical research in four places; calling it 'still SEO' is a rhetorical strategy, not a neutral description. Four-point rebuttal of Google's AI optimization guidance, with the "still SEO" framing identified as the mechanism keeping AEO investment suppressed.
Theme 2, Verification as the rate limiter
Anthropic’s Glasswing update (AI has removed the vulnerability discovery bottleneck in security; the rate limiter is now verification, disclosure, and patching speed) found 10,000+ high or critical vulnerabilities across approximately 50 organizations in the first month of deployment. The constraint is not discovery. The constraint is human throughput on verification, disclosure, and remediation.
"Progress on software security used to be limited by how quickly we could find new vulnerabilities. Now it's limited by how quickly we can verify, disclose, and patch."
This extends Verification, not execution, is the irreplaceable human job with the most concrete domain evidence yet. The same shift from generation bottleneck to verification bottleneck appears in MCP outbound deployment. Florin Tatulea's Prospecting from the Trenches #132 (Connect one MCP tool and validate one workflow before building a full outbound stack; complexity compounds error before it delivers value) documents that full-stack MCP outbound fails because tooling costs and model error rates compound before any single workflow is validated. His rule: connect one tool, prove one workflow, then expand. Internal tasks can run unattended. Anything touching a prospect requires a human checkpoint until error rates are understood.
- Anthropic, AI has removed the vulnerability discovery bottleneck in security; the rate limiter is now verification, disclosure, and patching speed. Glasswing security data as concrete evidence that AI removes the discovery bottleneck and shifts the constraint to human verification throughput.
- Florin Tatulea, Connect one MCP tool and validate one workflow before building a full outbound stack; complexity compounds error before it delivers value. Staged MCP outbound deployment rule: one tool, one workflow, validated error rate, then expand.