diff --git a/CHANGELOG.md b/CHANGELOG.md
index 4642870..d032510 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -7,6 +7,10 @@ this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.htm
 
 ## [development]
 
+### Fixed
+
+- A paragraph that follows a definition list, separated from it by a blank line, is no longer swallowed into the `dl` element: the list ends before it, and terms after the paragraph open a list of their own.
+
 ### Added
 
 -
diff --git a/internal/markdown/extension_corpus_test.go b/internal/markdown/extension_corpus_test.go
index cf478c4..f1b49c8 100644
--- a/internal/markdown/extension_corpus_test.go
+++ b/internal/markdown/extension_corpus_test.go
@@ -81,6 +81,10 @@ func TestDefinitionLists(t *testing.T) {
 			"<dl>\n<dt>Term</dt>\n<dd>\n<p>Def</p>\n<p>more</p>\n</dd>\n</dl>\n"},
 		{"definition inline content", "Term\n: A *bold* claim",
 			"<dl>\n<dt>Term</dt>\n<dd>A <em>bold</em> claim</dd>\n</dl>\n"},
+		{"definition paragraph after list", "Term\n: Definition\n\nParagraph after.",
+			"<dl>\n<dt>Term</dt>\n<dd>Definition</dd>\n</dl>\n<p>Paragraph after.</p>\n"},
+		{"definition paragraph between lists", "Term\n: Def\n\nBetween\n\nTerm 2\n: Def 2",
+			"<dl>\n<dt>Term</dt>\n<dd>Def</dd>\n</dl>\n<p>Between</p>\n<dl>\n<dt>Term 2</dt>\n<dd>Def 2</dd>\n</dl>\n"},
 	}
 	for _, tc := range cases {
 		t.Run(tc.name, func(t *testing.T) {
diff --git a/internal/markdown/parse.go b/internal/markdown/parse.go
index ff961ae..ae6a6d4 100644
--- a/internal/markdown/parse.go
+++ b/internal/markdown/parse.go
@@ -614,6 +614,9 @@ func (p *parser) finalise(n *Node) {
 			n.content = trimTrailingBlankLines(n.content)
 		}
 	case kindList, kindDefList:
+		if n.kind == kindDefList {
+			p.liftLeakedTerms(n)
+		}
 		n.tight = !blocksAreLoose(n.children)
 	case kindDocument:
 		p.gatherFootnotes(n)
@@ -623,6 +626,60 @@ func (p *parser) finalise(n *Node) {
 	}
 }
 
+// liftLeakedTerms moves the paragraphs that never became terms out of a
+// definition list. A paragraph may sit directly inside the list as a
+// term waiting for its definition line; when that line never arrives the
+// paragraph is prose after the list, and the list ends before it, since
+// a dl element carries nothing but terms and definitions. Terms and
+// definitions after the leak form a list of their own.
+func (p *parser) liftLeakedTerms(dl *Node) {
+	leak := -1
+	for i, c := range dl.children {
+		if c.kind == kindParagraph {
+			leak = i
+			break
+		}
+	}
+	if leak < 0 {
+		return
+	}
+	parent := dl.parent
+	base := 0
+	for i, c := range parent.children {
+		if c == dl {
+			base = i + 1
+			break
+		}
+	}
+	tail := dl.children[leak:]
+	dl.children = dl.children[:leak]
+	insert := func(n *Node) {
+		parent.children = append(parent.children, nil)
+		copy(parent.children[base+1:], parent.children[base:])
+		parent.children[base] = n
+		base++
+		n.parent = parent
+	}
+	var regroup *Node
+	for _, c := range tail {
+		switch c.kind {
+		case kindParagraph:
+			insert(c)
+			regroup = nil
+		case kindDefTerm, kindDefItem:
+			if regroup == nil {
+				regroup = &Node{kind: kindDefList, parent: parent}
+				insert(regroup)
+			}
+			c.parent = regroup
+			regroup.children = append(regroup.children, c)
+		}
+	}
+	if regroup != nil {
+		regroup.tight = !blocksAreLoose(regroup.children)
+	}
+}
+
 // trimTrailingBlankLines removes the trailing blank lines of an indented
 // code block. Every content line carries its newline, so the result of a
 // non-empty block ends with exactly one.