Each version of a page lists all its alternates in link tags in the head, including itself, with a language code: en, es, fr and so on. An extra x-default entry names the page to show when none of the languages fits. The same set can also go in the sitemap.
The annotations have to be reciprocal. If the English page names the Spanish page as its alternate, the Spanish page has to name the English one back. A one-way link is easy to ignore. Each version should also have its own self-referencing canonical, not a canonical that points at the English page.
hreflang only works if each language lives at its own URL. A site that switches language with a cookie or the browser setting, on one URL, gives search engines nothing to point at. Redirecting crawlers by country or language causes a similar problem: the crawler never sees some versions.
The mistakes are usually small and quiet: a language code that does not exist, an alternate that returns a redirect or a 404, or a translated page missing from the set. Search Console does not flag all of them, so it pays to check the rendered head of each version.
An example from our own work
audacian.eu is published in English, Spanish, Romanian and French. English has no prefix; the others live under /es, /ro and /fr, and some pages have localised slugs, such as /es/para-estudios-de-interiorismo for the interior design page. Every indexable page lists its alternates with hreflang plus an x-default pointing at the English URL, and the sitemap carries the same alternates. Pages that exist only in English, like our insights articles, list only English and x-default. Before this setup the language lived in the browser only and every language shared one URL. Our routing middleware never redirects crawlers by country, so they always see each language at its own URL.