What pulls a page into an AI answer is the clarity of the steps themselves, far more than the HowTo tag wrapped around them. Treating the markup as the thing that earns the citation gets the causation backwards, and the promise that adding HowTo schema will get you into AI answers is not one anyone can honestly make. The credit belongs to how the steps are written, because that is what a model can extract cleanly, and the tag mostly rides along.
Two facts make this concrete. First, Google removed HowTo rich results from desktop and mobile back in September 2023, so the tag no longer produces any visible result in Search; the type still validates, but it triggers nothing. Second, the evidence that schema markup itself improves AI extraction is mixed rather than settled. Microsoft has said structured data helps its systems, one independent study found no correlation between schema coverage and citation rates, and Google has stated that no special markup is needed for its AI overviews. Both of these are worth verifying again at the time of writing, because this area moves quickly.
What is steadier is that language models extract better from content laid out as clear, ordered, self contained steps. A numbered sequence where each step makes sense on its own gives a model a clean unit to lift and attribute. That structure lives in the visible writing, and it is the visible writing that gets cited. The tag, at best, is cheap insurance with an uncertain payoff, not the mechanism.
So the writer builds genuinely clear numbered steps that stand on their own, and stops banking on the markup to do the work that only the writing can do.