Use AI where it helps most

Decide which writing tasks AI can speed up and which ones you should do yourself.

Hand AI the research and the checking, share outlines and headlines with it, and keep the claim, the proof, your own story and the final edit for yourself. The time AI saves comes unevenly: it does some steps well, some badly, and some steps belong to you even when it could do them.

Priya runs a centre that teaches children (she is an invented example), and much of her writing goes to parents. Once AI had given her a few good results, she let it write everything: blog posts, emails, captions, replies to parents. For a few weeks she felt she had gained an extra member of staff. Then one parent wrote back to ask whether an email had come from Priya or from a computer, since it didn't sound like her. Priya read it again and had to agree.

Going back through her sent emails, she found the weak passages all fell in the same places. The account of how she started the centre had been written by the AI and could have belonged to anyone. Her claims about results had also been drafted by the AI, and several were vaguer than the results she had evidence for. She now uses AI for outlines and subject line options only, and writes the story and the claims in her own words.

Getting AI to sort your research first

A good share of AI's usefulness comes before the first sentence of a draft. Take a transcript from a 20-minute customer interview of the kind you ran in lesson 1.3. AI can go through it and pick out the trigger, the alternatives the customer considered, their doubts, and the phrases that stand out, quoting each one word for word. Read the full transcript yourself as well; the summary shows you where to focus and does not replace the reading.

Reviews can be handled in a similar way, whether they are about your business or a competitor's. Paste in a batch, perhaps fifty reviews, and ask for them to be grouped into what customers praise and what they complain about, with their exact phrases. The sorting you did by hand in lesson 1.4 goes faster this way.

When you are ready to plan, tell AI the question you are answering, who you are writing for, and your short answer, then ask for several outlines, say five possible ones. Pick the one closest to what you want and change it to fit. For headlines and hooks, lesson 8.2 showed how quickly AI produces variations from a brief, and the choosing stays with you.

Asking AI to check a draft

Once a draft exists, AI makes a useful second reader. Set a maximum sentence length and ask it to list every sentence longer than that, with a suggested split for each; whether a split actually reads better is your call. For the third editing pass in lesson 7.1, paste a single sentence and ask for plainer words. To test whether the piece does its job, give AI the post together with the question in its title, and ask two things: whether the answer shows up in the first paragraph, and whether anything is missing that a reader would expect to find.

Lesson 7.3 asked you to read a readability tool's results as a pointer to places worth another look. Read AI's feedback the same way, and make only the changes you agree with.

Four things AI should not write for you

The core claim is what you promise, and who you promise it to. Deciding it is a business decision that rests on what you are able to deliver and what your customers need. You know both of those things, and the AI knows neither of them.

Proof means your reviews, numbers, credentials and guarantees, all taken from your own records. AI has no access to those records, and when asked for proof it may make some up.

Your personal story comes from things that happened to you: a student's face changing as the method clicked, or a customer whose cake you remade at midnight. Stories like these are often what persuades a reader most, and they carry your voice, which was what Priya's parent noticed was missing.

The final edit decides what goes out under your name. As lesson 8.4 describes, read each line the way your reader will, and check each fact.

Removing personal data before you paste

Interview transcripts, customer messages and enquiry forms tend to include names, phone numbers, details about children and other personal data. Singapore's personal data law, the PDPA, makes you responsible for protecting the personal data you hold and for using it only for the purposes you collected it for. Once you paste it into a third-party tool, it may be stored or processed somewhere outside your control. Depending on the tool, your plan and your settings, some AI tools may also use what you type to train or improve their models.

Before any customer material goes into an AI tool, do three things. Read the tool's data terms to find out how it stores your input, whether that input is used for training, and whether training can be switched off. Look up your own obligations under the PDPA in the PDPC's guidance. Then strip out names, numbers and anything else that identifies someone. Of the three, this last one is the easiest habit to build, and it keeps the words you need for your writing while leaving the person out.

Sorting your own writing steps

Write down every step you take to produce one piece, starting from the first research and ending with publishing. Next to each step, mark whether AI helps, you do it, or both, and give one reason. Be honest about the steps where AI has been giving you poor results, and about the steps you still do by hand that AI could speed up.

Write down the steps you follow to write one piece. Mark each step as AI helps, I do it, or both, and give one reason for each.

Course

Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).