The Problem No One Wants to Admit
Most GTM teams have quietly crossed the same line: AI is writing first drafts of sales copy, case studies, and campaign content. The productivity case is obvious. The risk is subtler — and Dave Kellogg named it cleanly in Terminators, Dogs, and Skills.
The framework is a cocktail party test. Imagine your content being read aloud at an industry event by someone who actually lives in that market. Would insiders nod, or would they quietly notice something's off? Kellogg calls the failure signal 'the dogs bark' — buyers sense inauthenticity without being able to articulate exactly why. They just disengage.
This matters more in B2B than almost anywhere else. Enterprise buyers are domain experts. They know which analysts they trust, which forums they actually read, which pain points are real versus vendor-fabricated. AI trained on the open web produces plausible-sounding content — but plausible is not the same as fluent.
Two Failure Modes Worth Auditing Now
Kellogg identifies two specific patterns that cause AI copy to fail the test:
Wrong social context. The copy cites the wrong authorities, references the wrong watering holes, or uses the vocabulary of an outsider. Think of a healthcare IT vendor whose content references analyst firms no CMIO reads, or a manufacturing software company that writes like a SaaS startup. The words are fine. The signals are wrong.
Wrong problem. The content addresses a pain point no one in that segment actually prioritizes. This is the more dangerous failure because it looks polished. It just doesn't land — and it won't generate pipeline because it's solving for a problem that isn't in the buyer's top three.
Both failures are diagnosable before content ships. Kellogg built an AI skill specifically to flag them — a reviewable prompt layer that evaluates draft copy against insider-fluency criteria before a human reviews it.
The Actionable Move
If your team is using AI to produce GTM content at volume, add an audit step before human review, not instead of it. Build a prompt (or a formal AI skill, if your stack supports it) that asks:
- Does this content cite sources, communities, and authorities this specific buyer segment actually trusts?
- Is the problem being addressed in the buyer's stated top priorities — or in what we wish they cared about?
- Would someone who has worked in this industry for ten years find anything here that rings false?
This isn't about slowing down AI-assisted content. It's about routing the right QA step to the right tool before your CMO or a senior sales leader has to catch it manually. The goal is making AI-assisted content genuinely insider-fluent — not just grammatically correct and on-brand.