Every new discipline creates a dark side. SEO gave us keyword stuffing, link farms, and cloaking. Social media gave us fake followers and engagement pods. AI Agent Optimization will have its own version — and it's already starting.
Before you build your AAO strategy, you should know what not to do. Not because it won't work in the short term. But because the platforms are catching up faster than most people realize, and the penalties are domain-level.
The Risks Worth Taking Seriously
What "Black-Hat AAO" Will Look Like
The pattern is predictable because we've seen it before. The same logic that drove keyword stuffing will drive llms.txt inflation. If AI agents prefer businesses with clear use cases, people will manufacture fake use cases. If structured data improves recommendation frequency, people will add schema markup that doesn't match their actual services.
The problem: AI companies have a strong incentive to detect this. Their products are only valuable if the recommendations they make are accurate. Stripe's recommendation is worth something because it means Stripe is genuinely good at payments — not because Stripe gamed a system. The moment AI recommendations become gameable, they become worthless, and the companies lose.
The Competitive Intelligence Problem
This one is underappreciated. When you create an llms.txt file listing your services, pricing ranges, use cases, and competitive positioning, you're creating a machine-readable summary of your business strategy. Any competitor running an agent against your site gets that information immediately, cleanly formatted.
This doesn't mean you shouldn't create one — you should. But it does mean thinking carefully about what level of detail serves your customers versus what level of detail hands your competitors a roadmap.
- Specific pricing is worth including — agents need it to make useful recommendations, and it's probably findable anyway
- Proprietary methodology or process detail belongs on gated pages, not in llms.txt
- Client names or case study details should be kept off public agent-accessible files unless they're already public
The Principles That Hold Up
The window between "this matters" and "everyone knows this matters" is short. The businesses that use it to build genuine agent-readiness will have a durable advantage. The ones that use it to game the system will have a brief advantage followed by a hard correction.
That's how it always goes. Build for the long game.