2025-07-03 AI Mastermind
Table of Contents
- Insight - MVP First, Then Internet - The Scope-Control Principle for AI Chatbots
- Insight - Becoming Cited by AI - The New Authority Signal
Session Overview
This July 3rd session opened with a small crew and a live problem: Dirk’s ElevenLabs voice chatbot had hallucinated that he earned €500k/year when a friend tested it. The ensuing discussion produced one of the most durable frameworks of the series — the MVP First principle — which establishes that scope discipline in AI tools is not just good product design, it is trust design.
The second major thread emerged unexpectedly when Dirk mentioned that a C-suite executive in Switzerland had contacted him because ChatGPT named him as one of the top three executive headhunters in Germany when asked directly. This organic discovery of AI-era authority opened a significant discussion about how visibility and credibility are being established in the emerging GEO (Generative Engine Optimization) landscape — a domain that most coaches and consultants have not yet begun to navigate intentionally.
A third discussion explored AI governance, copyright, and data training law, sparked by Bali’s attendance at a panel discussion and her research into the EU AI Act and the Anthropic copyright case. This provided important context for how coaches and consultants should think about the legal and ethical dimensions of AI-generated content.
High-Signal Moments
- Dirk’s bot hallucinated his salary — leading to the “closed knowledge base first” principle and a practical framework for managing hallucination risk in deployed AI tools
- Kasimir’s technique of requiring citations and source text verification to catch hallucinations during development (not just in production)
- Lou’s middle-ground suggestion: restrict chatbot to pre-vetted trusted sites rather than the full Internet, balancing knowledge freshness with accuracy
- Dirk was named by ChatGPT as a top-3 headhunter in Germany — discovered via an inbound call from a Swiss AI CIO
- Discussion of GEO (Generative Engine Optimization) as an emerging practice analogous to SEO but for AI model citation
- Bali’s report on the Anthropic copyright case: training on data is acceptable, but the source of that data (scraping books without permission) is the legal liability
- Lou’s suggestion of a robots.txt-like opt-in mechanism for AI training, giving creators control over what gets indexed
- Maisie’s discovery of BotBuilders service — Lou’s useful framing: “They show you the Rolls-Royce engine and ask: do you want to understand the engine, or just drive the car?”
Open Questions
- How do you actively maintain and strengthen a position of authority once an AI model begins citing you?
- As EU AI regulation tightens, what does compliant AI use look like for coaches handling client confidential data?
- What is the minimum viable knowledge base scope for a voice AI that still serves the client relationship without hallucination risk?
- Will AI training opt-in mechanisms emerge, and if so, would coaches want their content included?
- How does the Anthropic copyright precedent affect coaches who want to train AI on client session data?
Suggested Follow-Through
- Complete the Chatbot Scope Audit before any deployment: categorize every possible question as OWNED or OPEN, build MVP on OWNED only
- Run the AI Visibility Audit: query ChatGPT, Claude, Perplexity with “[your specialty] expert in [your geography]” and document what appears
- Research GEO best practices — this is early-stage but the Dirk example shows it is already producing real business results
- For anyone building client-facing AI tools: define the bot’s “I don’t know” response before launch — this is as important as the knowledge base itself
- Stay current on EU AI Act developments if you operate with European clients or data
Additional Resources
Ideas from Chat
- Donald Kihenja: Running n8n on Docker — mentioned as his current automation infrastructure setup; relevant for members exploring self-hosted automation options
- Donald Kihenja: Sidekick browser — recommended as an Arc/Dia equivalent for non-Mac users; offers similar tab/workspace organization features
- Kasimir Hedstrom on target audience research: Both Kasimir and Don Back noted independently that female professionals in early-to-mid career are more receptive to coaching than male PhD graduates; the hypothesis is greater emotional self-awareness and lower “not invented here” resistance in women. Both have independently converged on a similar ideal client profile.
- Michael Simmons referenced as a figure whose work on AI and knowledge management the group wants to follow more closely — Don Back suggested inviting him as a guest (name provided by Kasimir Hedstrom in chat)
- Don Back on specialist selection: “I had discovered that male PhD grads are highly resistant to help… testosterone-toxicity?” — a candid insight on coaching market selection that generated agreement from both Kasimir and Bally
Derived Artifacts
- avatar-archaeologist (Avatar Archaeologist — building avatars from real client data)