In part 3 of my Horror Escape Game series in Unreal Engine 5, I start building the Ghost AI and set up the foundation for a smarter enemy. I duplicate my character blueprint into a new GhostAIBP, slow the default movement speed (so it walks first), and add a Pawn Sensing component—the main thing that lets the AI “see” and “hear” the player. I tweak the sensing settings (hearing threshold, sight radius, vision angle) and I also uncheck “only sense player” because it can create bugs. Then I hook up OnSeePawn to an AI Move To so the ghost immediately chases the detected player, and I print debug strings for when it sees you and when it catches you. After that, I make sure the AI can actually navigate the house by adding a Nav Mesh Bounds Volume and visualizing it with P. If the navmesh doesn’t cover tricky areas (like slopes), I show you how I fix it by adjusting Recast NavMesh generation values—especially Default Cell Size (I set it around 5). Finally, I replace the mannequin with a real ghost model (fixing import scale via Uniform Scale 100), rebuild a simple material so it blends with the dark lighting, migrate animations from the free Horror Framework project, retarget them to the ghost, and create a 1D blend space + Animation Blueprint that drives idle/walk/run based on speed. The AI is still “a little bit dumb” here, but this setup is exactly what we’ll improve with Behavior Trees in the next part.