Once AI makes expertise cheap to fake, what still signals competence? Enforceable promises, not badges
When AI produces convincing reports and proposals at near-zero cost, signals built on costly effort stop working — the paper's substitute is a verifiable promise backed by damages.
Key numbers
- Framework: generative AI is modelled as compressing the discernible headroom between what machines produce cheaply and what buyers can distinguish; below a critical headroom no separating equilibrium in production-side signals exists.
- Consequence: the market pools — high-competence providers earn no premium and exit if they have outside options, an Akerlof lemons dynamic.
- Substitute: an outcome-contingent signal — a warranty backed by damages D with ex-post verifiability φ — restores full separation at any level of AI capability whenever φ·D ≥ v, where v is the value of a solved problem, under four stated and priced preconditions.
- What does not work: provenance certification priced as a type-independent stamp (for example C2PA-style) cannot restore separation, while a verifiable commitment to forgo the AI frontier re-imposes the pre-AI cost function.
- Extensions: civil-procedure costs set a minimum ticket size below which court-enforced warranties cannot sustain separation, and under liability insurance the signal-effective quantity is retained, collectible exposure.
Method and sample
arXiv paper “The Last Costly Signal: How Generative AI Collapses Competence Signaling and Why Liability Sustains Markets for Expert Services” (2026-07-28): signalling theory formalised for expert services as credence goods, deriving the conditions under which outcome-contingent promises restore separation, with preconditions, falsification conditions and a planned preregistered conjoint experiment with German-speaking B2B decision-makers.
What it means for brands
The first thing AI flattens in expert services is looking professional; what actually separates providers is verifiable commitment and liability. For law, consulting and accounting firms, being cited correctly by AI and clients means publishing verifiable service boundaries and responsibility terms, not just credential lists.
How SourceSight handles it
For regulated and professional-services projects SourceSight first checks whether AI can retrieve credentials, scope and liability pages, then the content factory structures those facts — credentials are themselves verifiable signals worth writing down clearly.
Limits
The paper is formal modelling plus a preregistered experiment design rather than completed empirical results, and its four preconditions may not hold in every real market.
References
Run it on your own project
What the same prompts return across 10 engines is only knowable by running them.
Related research
Cross-category recommendations: +36% add-to-cart on item pages, +15% on cart pages
Recommending general merchandise to grocery shoppers — a milk frother with the milk — has long been hard; LLM retrieval plus real-time cart context moved production A/B metrics.
Industries · 2026-10-07Home services: 22% of homeowners already use AI to find a provider, and 80% of businesses don't know how to be found
The local services funnel is moving upstream: before a homeowner opens a website or calls, AI has already synthesised reviews, reputation and service descriptions into one answer.