
To use genAI as a development professional, first, understand that genAI has risks. Use it for what it’s good at and review all outputs more carefully than you would your own writing.
If you are a development professional or a government official who deals with international organizations and donors, you or your colleagues may already use generative AI (genAI) to help you write. Or maybe you are curious to learn how. But as good as genAI is (and it’s developing rapidly), it’s not perfect. So how can you best use it?
In this article, I talk about the risks of using genAI, the kind of tasks it’s good and bad at, and some considerations when using it in the international development field.
(A note for “native” English writers who are strongly opposed to genAI.)
Disclaimers
First, some definitions and limitations.
- By “genAI”, I refer to chatbots like ChatGPT, Gemini, and Claude.
- I’m looking at the free versions. The paid tiers have higher abilities,1 but I’m assuming that most users are using free versions.
- GenAI is not my professional focus. What I write here is based on my experiences with free and paid chatbots,2 what I observe in my clients’ work, and what I learned from reading.
- GenAI is changing rapidly. What I say here will probably become out of date in no time!
- As an editor, I do not use genAI tools for editing tasks. Please see my AI use policy.
- More on this below but: beware of what you put into any genAI product. Follow your organization’s AI policy.
The risks of using genAI
Before you use genAI for English writing, it’s important that you understand the risks of using tools that are still far from 100% reliable. Expand each to learn more.
Using poor genAI content can harm you, your organization, and those who rely on you
GenAI can produce what looks like fluent English, but that output may be confusing, inappropriate or, at worst, wrong. Using problematic genAI output can
- erode the credibility and reputation of the document, you, and your organization/government
- damage or harm those who rely on your work
We know that credibility and trust are fundamental to development work and diplomacy.
What to do: Use genAI with even more care than you would when writing your own words.
GenAI products pose privacy and data protection risks
When you use ChatGPT, your data goes into the servers of OpenAI, the company that develops and runs the chatbot. That data could be used to train the model behind the chatbot (your data goes into the model, which others use).
You can opt out of this, but your data can still be kept on that company’s servers for as long as their terms specify. That means your data is beyond your control—a problem especially if you are working with sensitive information.
It also raises murky questions about intellectual property rights (AI-generated content cannot be copyrighted).
What to do:
- Only put in information that is OK for anyone to see. You could anonymize your data (take out any identifying and sensitive details) before putting it into genAI, for example.
- Change genAI settings to “do not train model” (ChatGPT calls it “make it better for everyone”—turn that off).
- Follow your organization’s AI policy.
You might also consider setting up a large language model (LLM) locally, on your computer. Using a local LLM means no data leaves your computer.3
Some organizations have their own private LLMs for their staff to use securely.
- I highly recommend this excellent intro to privacy and data protection and AI.
- The European Commission offers free AI tools (I haven’t tried them). Since they’re bound by EU privacy laws, could they be safer than private genAI products?
How to use genAI effectively for writing
So! How can you use genAI effectively to help you write English in the development field? In general:
- Use genAI for narrow, well-defined tasks. GenAI can follow specific instructions better than broad ones.
Identify possible subject-verb disagreementsis better than.4Find any grammatical mistakes - Ask it to suggest changes only, not directly make changes. This gives you control over what to change and what not.
- Tell it what to do (give “good” examples and words/terms to use, for example) and also what not to do (specify words/terms and formatting to avoid, for example).
- Work on small chunks of text. The genAI chatbot can only keep a limited amount of text in its working memory. Do not tell it to reorganize long text for flow or structure.
- Use genAI to brainstorm and help you clarify ideas. GenAI is also very good with summarizing text.
- If the chat starts going awry, start a new chat. You can copy and reuse your prompts.
Good prompting is essential for using genAI effectively, but prompt engineering is beyond the scope of this post and my skills. You’ll have to explore more on your own. I put a little bonus bit at the bottom to get you started.
Effective writing needs structure, wording, and design that lets your reader find, understand, and use the information. Toggle the sections below to see the kinds of tasks genAI is good and weak at for (1) structure and (2) wording. (I haven’t tried genAI for design yet.)
For structure
Structure is crucial to effective writing because it helps the reader follow along your thought process, as well as find and use the information they need.
GenAI is good for brainstorming and clarifying your ideas on how you might structure your document. However, you cannot just ask it to “reorganize” your content. You need to work in steps and give detailed instructions.
Good for … ⭐
- Suggesting an outline for your document type (for example:
I’m drafting a memorandum of understanding (MOU). What are typical sections of an MOU?) - Brainstorming what each section might contain (for example:
What should go under the Preamble section of an MOU?) - Clarifying which section a piece of info could go into (for example:
In which section should I mention what each party will contribute?)
Weak for … 👎
- Reorganizing very long text (especially with a broad prompt like
).Reorganize this text
GenAI does not “think” on its own whether, for example, there is repetitive or inconsistent text (something a human might notice) and how that might be fixed—especially in a long document.
GenAI best handles small chunks of text. It also needs detailed, specific instructions. Break down big tasks into smaller, well-defined steps and take genAI one step at a time. (See the bonus bit about prompts to get you started.)
For words and sentences
GenAI can be useful at the words and sentences level. Use it to help you write short, active sentences to make it clear who is doing what, and to use words that are most familiar to your intended readers and users.
Good for … ⭐
- Shortening sentences
- Improving awkward phrasing
- Suggesting more direct words
- Finding incorrect grammar (but it can be wrong so do not trust it entirely)
- Identifying acronyms (so you can choose the minimum necessary)
… And many other specific tasks. GenAI is also very good with summarizing.
Weak for … 👎
- Writing from scratch. GenAI writing is bland. And it may look correct and smooth but be difficult to understand.
- (Re)writing freely to shorten or
. GenAI may choose the wrong words, delete the words you want to use, or produce contextually irrelevant writing.make clearer
Special considerations for international development
1. Focus on getting the information across clearly, rather than on elegant wording and presentation.
In international development, you are working with counterparts with many different levels of English ability. The text may also be translated (possibly via machine translation). But all parties must understand what the document says, which means making the meaning explicit.
Use genAI to help you organize the information well and write clear sentences that convey your intended meaning. Don’t be satisfied with smooth, elegant-looking language that conveys nothing.
2. Review and adjust for diplomatic tone.
You know best how to mention sensitive topics to partners. GenAI tools do not have this insight or experience. Make sure the tone and wording of the writing is appropriate to its audience and purpose. Ask a human colleague or supervisor for help, if you’re unsure. (But you can still use plain language principles, even in diplomatic writing.)
3. Beware of words and sentences that reinforce biases and stereotypes.
The large language models behind genAI tools reflect the biases in their training material. Be extra careful to review AI-generated text for potentially harmful language that reinforces negative biases. (Online language guides can help you navigate some of the evolving terminology.) We are all working to promote human rights, and that includes through our writing.
Summary
On security and safety
- Humans must guide the process and check all outputs.
- Follow your organization’s (government’s) AI policies.
- Do not put in private (identifying), sensitive, confidential information, especially into commercial genAI tools. Anonymize the data before putting it into a genAI tool.
- Change genAI settings to “do not train model” (do not “make it better for everyone”).
- Learn more data privacy and security in genAI.
On effective use of genAI
- Use genAI for specific, well-defined tasks. Learn how to prompt effectively.
- Focus on making your document clear, with good structure and words familiar to your intended audience. Remember that your document will be used and read by professionals who are not always comfortable with English or may be translated.
- Review all outputs, not only for factual correctness, but for appropriate tone and language that does not perpetuate harmful biases.
Suggested resources
- AI Sidequest newsletter for writers, editors, marketers, and educators (Mignon Fogarty): She helps you stay updated on genAI news in an understandable, non-technical way.
- Privacy and Risks with Generative AI free e-course (AI Literacy Institute)
- The 8 Best AI Courses for Beginners (Zapier)
Anything other advice? Anything else that you would add to these lists? Please leave a comment!
Bonus: Getting started on prompting
The prompt is crucial to using genAI effectively. Prompting is a huge new field, and you’ll have to explore it yourself as it’s beyond the scope of this post and my skills, but here are some starting ideas:
- Give genAI concrete instructions and examples. You will not get good results with short, vague prompts (
) without explicit instructions. Tell it both what to do and what not to do (for example,Write the introductionDo not use the following words: ...). - Work in steps (generate an outline → give your notes and ask genAI to generate the document, following the outline → go back and check → fix specific problems).
- Use genAI to help you craft your prompt. Tell it what you’re trying to do and ask it to write a prompt.
- Check what genAI products will let you upload reference docs. (Google’s NotebookLM, for example, can work with ONLY the resources you feed it.) Use that to help you get more relevant outputs and prevent the tool from hallucinating (invent things that are not true).
The basic prompt: CARE: Structure for Crafting AI Prompts, by Kate Moran, NN/group. Try also Effective Prompts for AI: The Essentials (MIT Sloan Teaching & Learning Technologies). And if you want to explore an extensive list, see Best Prompt Engineering Resources (Bilgin Ibryam, The Generative Programmer).
A note for “native” English writers who are strongly opposed to genAI
Those of us who grew up with English are lucky that English is the de facto common language in much of the international development field. We can afford to discuss the ethical concerns surrounding genAI—exploitation of workers, plagiarism, environmental impact—and condemn genAI use, but that’s our privilege.
Cover image by allreadyserviceprivat on Pixabay.
Footnotes
- For example, see information-literacy expert Mike Caulfield’s article.
- free ChatGPT and Claude and Perplexity, ChatGPT Pro, paid Anthropic API, and local large language models such as Qwen3 and Llama3
- Sorry, I can’t suggest any further resources to get you started. But researching technical things like this is where genAI chatbots like Perplexity can be very helpful).
- The strikethrough means I don’t recommend using this as a prompt.