The data card is the dataset's summary and its first test of credibility. It should let a reader who has never seen the data understand what it contains and what it does not. Useful cards are specific. They give record counts by year, list every field with its type and fill rate, name the method used to remove personal data, and say plainly where coverage is thin or records were excluded.
The card is also a legal document in effect. Buyers commonly ask the seller to represent that it is accurate, and acceptance checks are written against the figures it states. A card that overstates volume or omits a gap creates a problem at delivery. Write it from measurements, not memory: profile the dataset first, then copy the numbers. Update it whenever the dataset changes, and give each version a date. Outrich Data prepares a data card for each dataset it buys and supplies it with the title report. The free generator on this site produces a card in Markdown from a form.
Templates and generators help with structure, but the content has to come from the dataset itself.