Michael Krigsman has interviewed 897 executives about technology since 2013. Nobody had ever counted them — including him.
Dennis has a call with Michael today at 1pm Pacific. Eight and a half hours before it, he sent one prompt: inventory every CXOTalk episode, analyze what’s popular, run the full personal brand and business audit, make it visual, show what agents on Michael’s side could do, and write this article about how it went.
Here’s what shipped, how, and what it cost.
Shipped in one session — July 5, 2026
| The audit page | dennisyu.com/cxotalk — hero stats, guest wall with 12 headshots, 5 SVG charts, ten findings, the 63/100 score, seven agent blueprints |
| The workbook | CXOTalk-Inventory-and-Audit.xlsx — 8 tabs: 897 episodes, 863 YouTube videos with views, 569 guests, top 100, analysis, scorecard, agent roadmap. Zero formula errors. |
| The inventory | 897 unique episodes reconciled across four sources; YouTube reconciles to the upload: 863 + 16 Shorts + 163 live = 1,042 exactly |
| The email | Reply drafted in Dennis’s Gmail, staged for him to send before the call |
The architecture: three agents in parallel, one browser doing the impossible part
The work split four ways, all running at once.
Agent 1 — site inventory. Crawled the cxotalk.com episode directory, both episode sitemaps, and the podcast feed. Found 897 unique episodes against the site’s claimed 921 (the gap is pre-2016 webisode clips no longer listed anywhere). Bonus find: the episode sitemap silently drops a contiguous 108-episode block from March 2017 to July 2018 — broken pagination that’s been hiding those pages from Google for years.
Agent 2 — guests and entity checks. Parsed all 569 guest bios from the bio sitemaps, then verified what the machines know about Michael: no Wikipedia article, no Wikidata item, no Grokipedia page, no Knowledge Panel. This is a man cited in over 1,000 media pieces with a verified Muck Rack profile of 875 licensed articles. Earned, not wired.
Agent 3 — Ahrefs. Domain Rating 63, 1,283 live referring domains including Wikipedia, NIH, Columbia, Microsoft, McKinsey, and 664 links from his old ZDNET column — ranking for only 467 keywords worth $481 a month. And 140 AI-engine citations: Grok 91, Google AI Overviews 28, ChatGPT 2.
The browser did the YouTube harvest. No API key needed: Claude drove Dennis’s Chrome to the channel, called YouTube’s own internal browse endpoint from the page context, and walked the continuation tokens — all 863 videos with view counts in 28 loops, then Shorts and live tabs the same way. The three tabs sum to the channel’s stated 1,042 uploads exactly, which is the kind of reconciliation that makes an inventory trustworthy.
What the analysis loops found
Counting is inventory. The value came from running cohorts and splits until patterns fell out. The ten findings are on the audit page; the four that made Dennis grin:
- The AI title multiplier is 4.8×. Last twelve months, AI-titled uploads average 26,743 views versus 5,571 for everything else.
- The show lives in post, not live. Aaron Levie’s live premiere: 590 viewers. The edited upload: 42,000. The livestream is a recording session wearing a show’s clothes.
- CFO and CISO titles nearly triple CIO titles. 26.8K and 26.6K average views versus 10.1K — on a show literally named for the CxO. Two franchise formats hiding in plain sight.
- ~40 million words of transcripts are invisible to Google. Episode pages server-render about 350 words; the transcripts load client-side. Meanwhile Grok cites CXOTalk 91 times — the AI engines found the corpus anyway. Server-render it and you own both channels.
The score: 63/100 — and his Domain Rating is also 63
Scored on the Personal Brand Score rubric: Entity Home 12/20, Knowledge Panel 1/15, Search Ownership 14/15, Content Engine 12/15, Audience & Proof 13/15, Structured Data 5/10, Social Consolidation 6/10. Same pattern we found with Nixon Lee, at ten times the scale: the substance is world-class, the entity layer was never built. That’s not a criticism. It’s a to-do list with a 90-day scoreboard attached.
One reusable gotcha for the toolbox
WordPress lazy-loaders will eat your external images — ship skip-lazy from the start. The audit page’s 22 YouTube thumbnails published fine, then rendered as 1×1 placeholders: the site’s optimizer rewrote src into data-src but its restore script never fired for external images inside a raw HTML block. The fix that works across Smush, Jetpack, WP Rocket and friends is the WordPress-standard opt-out — class="skip-lazy" data-skip-lazy="1" data-no-lazy="1" loading="eager" on every hotlinked image. We verified all 22 restored by checking naturalWidth in the live DOM, not by eyeballing. Rule going forward: any raw-HTML page with external images gets skip-lazy attributes at write time, and verification means checking the rendered DOM, not the saved HTML.
What it cost
| Line item | Estimate |
|---|---|
| Research subagents (3, parallel) | ~566K tokens (metered: 258K + 181K + 128K) |
| Main session — browser harvest, analysis, builds, publishing | ~900K–1.1M tokens cumulative |
| Total | ~1.5–1.7M tokens, ~150 tool calls, one session |
| At Claude API list pricing | roughly $8–15 depending on model mix |
| What a strategy shop would quote for the same deliverables | $15,000–25,000 and three weeks |
That asymmetry is the whole point. The inventory nobody had, the analysis nobody ran, the score nobody computed — for the price of lunch, before lunch.
The real pitch: agents on Michael’s side
The audit ends with seven agent blueprints — transcript-to-article, YouTube librarian, clip factory, entity & schema, guest alumni, sponsor proof, newsletter digest — each grounded in a specific finding, each with a 90-day KPI. This is the same play we run in the Authority Engine and teach in every workshop: audit first, then train the agents on the audit, then let them run with the owner’s judgment on top. Michael has 897 episodes of raw material and 569 alumni relationships. His agents have plenty to do.
See it: dennisyu.com/cxotalk · Method: Personal Brand Score · Quick Audit (#MAA)

