Most people pick fonts by how they look. Good agents pick fonts by role: which is display, which is body, which is mono.
Start from roles, not names
Every output needs only three roles: display for headings, body for content, and at most one mono for labels or code.
Open the font pairing collection, then filter by mood and context — not by "what looks nice".
Propose three, test with real copy
The correct instruction to an agent:
Suggest 3 different font pairings from lamandesain.com/font-pairing
for a wedding photographer portfolio. Explain each choice,
then wait for my pick. No default pairing.
Test each option with one real headline and one real paragraph from the brief. Any font that fails at small sizes is out.
Lock the choice and record the reason
After the user picks, write the decision into the brief: "Display X + body Y because Z". The full rule is in pairing choices.
Locking stops agents from silently swapping fonts halfway — a classic AI slop pattern.
The most common mistakes
- Using more than three font families in one output
- Display and body both decorative, so hierarchy disappears
- Forgetting to test contrast and minimum sizes on the target viewport
Fix these three and your typography already beats 80 percent of generic AI output.