Growing evidence indicates that the mammalian microbiome can affect behaviour, and several symbionts even produce neurotransmitters. One common explanation for these observations is that symbionts have evolved to manipulate host behaviour for their benefit. Here, we evaluate the manipulation hypothesis by applying evolutionary theory to recent work on the gut-brain axis. Although the theory predicts manipulation by symbionts under certain conditions, these appear rarely satisfied by the genetically diverse communities of the mammalian microbiome. Specifically, any symbiont investing its resources to manipulate host behaviour is expected to be outcompeted within the microbiome by strains that do not manipulate and redirect their resources into growth and survival. Moreover, current data provide no clear evidence for manipulation. Instead, we show how behavioural effects can readily arise as a by-product of natural selection on microorganisms to grow within the host and natural selection on hosts to depend upon their symbionts. We argue that understanding why the microbiome influences behaviour requires a focus on microbial ecology and local effects within the host.