What We Learn Running AI Agents Against the Open Web

One agency spent $840k on manual link building over five years. We mapped where every dollar went, and why their cost per verified link was 12× higher than they reported.

Every link-building platform quotes a different number. Here's how to calculate the one that matters: cost per link still live at 90 days. And why that number is the only one your CFO should care about.

Campaign one teaches you everything about your vertical. Campaign two should cost less, take less time, and deliver more verified links — because you kept what you learned.

Retainer pricing hides variable costs. Cost-per-link pricing reveals them. Here's which one actually works, and why most agencies choose the one that's slowly killing them.

We detect challenges and hand them to a human. That choice caps autonomous throughput — and it is the right trade for anyone who plans to keep their domains.

Field mapping starts as a similarity search against forms we already solved, and only falls back to inference. The embedding step pays for itself by the second campaign.

Anchor text, follow status, indexation, screenshot, HTML snapshot, timestamp. What we store per check, and why re-verification schedules matter more than the first result.

A slow submission stage should never stall discovery. How a state machine per campaign lets every stage retry and resume without corrupting the run.

Scores between 40 and 59 are not a scoring failure, they are a routing decision. Treating manual review as a first-class state changed the qualification machine.

Five hundred descriptions in one category will collapse toward each other unless you measure it. The similarity threshold we settled on, and how we enforce it pre-write.