Chapter 1: Drafting and Summarising With AI

1.1 What a writing assistant actually is

ChatGPT, Microsoft Copilot, Google Gemini and Claude are built on large language models. They were trained on enormous amounts of text and learned which words tend to follow which. When you ask a question, the model writes an answer that is statistically plausible. That is why the prose is smooth, and also why it can invent a book title, a court case or a statistic that never existed. The industry calls this "hallucination". A calmer way to think of it: the tool is a quick, well-read drafting partner who has never been asked to show their sources.

Some assistants can search the web and show links. That helps, but it does not remove the need to open those links and confirm they say what the summary claims.

1.2 Low-risk tasks it already does well

• Rewording: making a blunt email polite, or a rambling one shorter.
• Restructuring notes: turning meeting scribbles into headings and bullet points.
• First drafts: a thank-you letter, a product listing, a school newsletter paragraph.
• Explaining jargon: "Explain this energy bill line as if I am new to this."
• Brainstorming: ten title ideas, five questions to ask a landlord.
• Proofreading: tools like Grammarly or the editors built into Word and Google Docs catch typos and clumsy phrasing.

Notice the pattern. In every case you supply the facts and remain the judge of the result. The AI is shaping words, not supplying truth.

1.3 A safe drafting routine

1. 1
   Strip the sensitive bits. Replace names, account numbers and addresses with placeholders such as [CLIENT] before pasting anything in.
2. 2
   Give context and a role. "You are helping a small bakery owner reply to a customer complaint. Keep it warm and under 120 words."
3. 3
   Ask for options. Request two or three versions so you compare rather than accept the first.
4. 4
   Check every fact. Names, dates, prices, quotes and references get verified against the original source.
5. 5
   Make it yours. Re-insert real details, read it aloud, and change anything that does not sound like you.

1.4 Summaries: useful, not gospel

Asking an assistant to summarise a long report or PDF is genuinely helpful for deciding whether to read the whole thing. The risk is subtle: summaries tend to drop caveats. "The drug reduced symptoms in a small trial of older adults" can become "the drug reduces symptoms". Numbers can be rounded wrongly or attached to the wrong year.

A good habit is to ask: "List every number and claim in your summary, with the page or paragraph it came from." Then spot-check two or three against the original. If the tool cannot point to where a claim came from, treat it as unconfirmed.

1.5 Prompts that tend to work

Short, specific instructions beat clever tricks. Say who the reader is, how long the text should be, what tone you want and what to avoid. For example: "Rewrite this for a neighbour who is not technical. Keep my three dates exactly as written. No exclamation marks." If the result drifts, say what is wrong rather than starting over.

1.6 Where to stop

Use AI drafts as a starting point for legal letters, tax questions or health information, never as the final word. The tool is not regulated as a solicitor, accountant or doctor, and it does not know your full situation. For anything with money, safety or legal consequences, bring the draft to a qualified person. Schools and universities also set their own rules on AI in coursework; check them before submitting assisted work.

Maya Okafor
Chapter editor · reviewed February 2026