Exploration favours the rarer hero

Spectare tests different atom arrangements automatically. Set an exploration rate in Settings and a share of visitors will see an experimental layout, with outcomes fed back into the learning loop so the engine gradually favours whatever converts.

Previously, when both hero component types were available, the AI assembling the page would almost always choose the text hero because a model choosing between a writable text component and an image component it cannot inspect defaults to the one it can control. This meant the image hero received around 3% of impressions despite being available, giving the learning loop almost no signal about whether it converts.

Two changes address this. Both hero components now carry explicit selection guidance: the image hero is preferred at awareness stage, where a strong visual earns the next scroll from a visitor deciding whether to stay, and the statement hero is preferred when the claim is specific to the visitor's intent rather than a general welcome. On top of this, exploration mode now has an explicit instruction to try the less-served hero medium, so the learning loop can gather real conversion evidence about image versus text rather than reinforcing an accidental bias.

The result is that hero type becomes a genuine variable in your experimentation rather than a fixed default, and the engine accumulates the evidence it needs to make an informed choice rather than one shaped by model preference.