In Plain English
Websites can expose substantially the same content at several addresses. Canonicalisation helps search systems consolidate those versions around a preferred URL. A canonical link is a signal rather than a guaranteed command, and it should usually identify a valid, indexable page that genuinely represents the duplicates. Search systems can choose another version when signals conflict. An accidental canonical can also cause the wrong URL to be treated as the preferred candidate.
Why it matters
Consistent canonical signals can reduce confusion between duplicate versions and consolidate their signals. They also make it clearer which address should appear in search and be shared internally.
Canonical choices should also be consistent with internal links, sitemap entries and redirects. If the site declares one URL canonical but repeatedly links to a parameterised alternative, it sends a muddled message. Inspect generated templates as well as one page, because a site-wide mistake can repeat quickly. Canonicalisation is most useful for genuinely equivalent versions; it should not be used to combine unrelated pages simply because one of them performs better.
Example
Tracking parameters create several URLs for the same product page, while the canonical points to the clean product URL.
Common misunderstanding
A canonical tag is not the same as a redirect. Visitors can still load a canonicalised duplicate; a redirect actively sends them elsewhere.