diff --git a/api/app/doc_render.py b/api/app/doc_render.py
index 98544fe..156e934 100644
--- a/api/app/doc_render.py
+++ b/api/app/doc_render.py
@@ -49,7 +49,6 @@
 _MAX_PROJECT_DESC_CHARS = 520  # room for 2-3 sentences (what it is, how it works, a result)
 _MAX_EDUCATION = 4
 _MAX_CERTS = 6
-_MAX_PARAGRAPHS = 4
 
 # Pages a CV must fit: one for a graduate / early-career candidate, two otherwise.
 _PAGE_TARGET = {"junior": 1, "experienced": 2}
@@ -310,7 +309,9 @@ def _letter_contact_lines(contact: dict) -> list[dict]:
 def build_letter_template_data(letter: dict, fit: tuple[str, ...] = ()) -> dict:
     """Render-ready letter data. `fit` is a step from _LETTER_FIT_STEPS: layout classes only."""
     letter = letter or {}
-    paragraphs = [p.strip() for p in (letter.get("paragraphs") or []) if p and p.strip()][:_MAX_PARAGRAPHS]
+    # Every paragraph is kept (the model may write five, the editor can add more); fitting the
+    # letter to one page is _fitted_html's job, which only ever tightens the layout.
+    paragraphs = [p.strip() for p in (letter.get("paragraphs") or []) if p and p.strip()]
     return {
         "page_target": 1,
         "body_class": " ".join(fit),
diff --git a/api/tests/test_documents.py b/api/tests/test_documents.py
index d2e1aef..e4e4560 100644
--- a/api/tests/test_documents.py
+++ b/api/tests/test_documents.py
@@ -446,13 +446,13 @@ def test_cover_letter_sign_off_follows_the_salutation(llm, salutation, signoff):
     assert (letter["salutation"], letter["signoff"]) == (salutation, signoff)
 
 
-def test_cover_letter_renders_four_paragraphs_and_the_contact_block(llm):
-    paragraphs = _letter_paragraphs() + ["A fifth paragraph the model added."]
+def test_cover_letter_renders_every_paragraph_and_the_contact_block(llm):
+    paragraphs = _letter_paragraphs() + ["A fifth, closing paragraph the model added."]
     llm.reply = json.dumps({"headline": "Computer Science Graduate", "salutation": "Dear Hiring Manager,",
                             "paragraphs": paragraphs, "signoff": "Yours faithfully,"})
     html = doc_render.render_letter_html(_cover_letter())
 
-    assert _letter_body(html) == ["Dear Hiring Manager,", *paragraphs[:4]]
+    assert _letter_body(html) == ["Dear Hiring Manager,", *paragraphs]    # the closing paragraph isn't dropped
     assert '<p class="letter-headline">Computer Science Graduate</p>' in html
     assert ">Bristol.</p>" in html and 'href="mailto:sam@example.com"' in html and 'href="tel:+447700900456"' in html
     assert re.search(r'<p>Yours faithfully,</p>\s*<p class="sig-name">Sam Okafor</p>', html)
@@ -563,6 +563,19 @@ def test_a_full_length_cover_letter_fits_one_page(pdf_dir):
     assert _weasy("cover_letter", letter).page_count == 1
 
 
+@needs_weasyprint
+def test_a_five_paragraph_letter_at_full_length_fits_one_page(pdf_dir):
+    letter = {"candidate_name": "Sam Okafor", "headline": "Computer Science Graduate",
+              "contact": {"address_line_1": "1 High Street", "city": "Bristol", "postcode": "BS1 1AA",
+                          "phone": "+44 7700 900456", "email": "sam@example.com"},
+              "date": "15 October 2026", "company": "Acme Robotics", "company_location": "Cambridge, UK",
+              "job_title": "Graduate Software Engineer (2027 start)",
+              "paragraphs": _letter_paragraphs((60, 75, 120, 70, 55))}   # 380 words, the prompt's top end
+    pdf = _weasy("cover_letter", letter)
+    assert pdf.page_count == 1
+    assert _sentence(55, 4) in " ".join(pdf[0].get_text().split())           # the last paragraph is on the page
+
+
 @pytest.mark.parametrize("doc_type, doc, expected", [
     ("cv", agent._persona_to_cvdoc(GRADUATE),
      ["Sam Okafor", "Software Engineering Intern", "Pantry Planner", "BSc (Hons) Computer Science"]),