Skills for prompt writing

The world of writing, including language materials writing, is currently debating the question of whether using generative AI tools like ChatGPT and Gemini is legal, moral and ethical. Meanwhile, teachers and materials writers are forging ahead and – routinely – using generative AI tools to create materials for them. Alongside this usage comes the well-used term ‘prompt’ and the idea that we need to become effective as prompt writers in order to generate these materials. As materials writers, we are partly equipped to write prompts because of our experience preparing clear and concise instructions with scaffolding. But here are a few other thoughts on the subject.

Iterative prompting

Prompt-writing is relatively logical and not quite as mysterious as some would suggest. It can be an iterative process in which you keep writing short prompts until you get what you want. For example, I wanted to generate a short video script conversation between a teacher and a student. Here are seven sentence length prompts I used to build the conversation I wanted.

  • Prompt 1: Write me a dialogue between a teacher and a student.
  • Prompt 2: Limit this to ten lines of dialogue only.
  • Prompt 3: Rewrite it so the teacher is angry.
  • Prompt 4: Rewrite it for video with information on what we see.
  • Prompt 5: Add eight gaps for students to fill in. Make sure the gaps are evenly spread.
  • Prompt 6: Number the eight gaps from 1 to 8.
  • Prompt 7: Rewrite the dialogue so all the words are at CEFR B1 level or below.

And I could go on prompting AI to refine the dialogue so it included certain language items or I could have asked AI to create comprehension questions and so on.

It’s true that I could reduce the number of prompts needed by planning and composing a perfectly crafted single prompt with lots of detailed information so that AI generated a more polished version immediately. But by the time I’d done that, I may as well as well have used my sentence level approach above. And also there’s something more intuitive in writing a prompt, reading AI’s draft, prompting again and redrafting again – much in the same way I would craft a piece of text without AI.

Composing a more detailed prompt

If I want Generative AI to create me a text (e.g. for a reading lesson), then my initial prompt might benefit from being longer. Using knowledge of discourse features, I’d can define the text type in more detail with information about the length, the style, the formality, the target reader, the vocabulary I need it to include, the language level and so on. But afterwards, I’ll still need additional prompts to fine-tune OR I might and cut and paste the rough draft into word and finish it off by myself, without the need for anymore AI generation.  Here’s an example of one I could write for materials about sustainability in coastal regions:

  • Prompt: Write an informal magazine style article for 120 words for teenage learners of English at CEFR B1 level on the topic of sustainability in coastal regions including the words clean-up, beach, plastic, fishing line, recycling, volunteers, wildlife, and safe.

To see what this prompt generated, cut and paste it into your preferred AI tool. You’ll find it generates a perfectly acceptable text but that it will still need some ‘human’ editing and curating.

An additional tool, not a substitute

So what can we conclude about prompt writing? Does it require us to learn a whole new set of sub-skills. Not really. Obviously, the more detail we can provide in an earlier prompt, then the quicker we probably get to what we want. But the bottom line is that an effective prompt needs us to ask the right questions from the outset. In order to know the right questions to compose the right prompts to generate the right text, we still need the knowledge and experience required to be a teacher and materials writer in the first place.

What has your experience in writing generative AI prompts? We’d love to hear your thoughts.

© Writing ELT Materials Limited 2024

We are happy for you to share and quote our blog posts, but please reference them and provide a link back to them. Thanks!