Video does not directly boost a page’s ranking by simply being there; its value is indirect, through engagement and dwell time, and through video-specific SERP features when the video is genuinely relevant. So the calibrated answer splits both ways: a relevant video can help, but the help comes from how people respond to it and from the extra surfaces it can win, not from the embed itself acting as a ranking signal. Irrelevant video dropped in “for SEO” helps nothing, because none of those indirect paths open up when the video does not serve the content.

The engagement path is the most common benefit. A video that genuinely fits the topic keeps people on the page longer, gives them another way to absorb the material, and can satisfy the intent more completely than text alone. Those behaviors are the kind of positive signals associated with content that serves users well, which is what ranking systems are ultimately trying to reward. The mechanism is indirect, the video improves the experience and the experience is what matters, so the lift only appears when the video actually earns the attention rather than sitting there as decoration.

The features path is the other real benefit. Relevant video, properly marked up and hosted, can surface in video-specific results and rich SERP elements, opening visibility you would not get from text alone. By contrast, an unrelated clip added because someone heard video helps SEO does not improve engagement, does not fit a video feature, and gives Google nothing to reward. The presence of a video is not the factor; the relevance and the response are.

Add video only when it genuinely serves the content, when it explains something better than text, demonstrates a process, or answers the intent more fully. When it does, host and mark it up so it can earn video features, and let the improved engagement do its indirect work. Skip the reflex of embedding a clip purely for ranking, because that embed on its own moves nothing. Treat the engagement and feature effects as worth confirming against your own data.