Why my AI writing assistant still needs me

Lynda Harris | October 29, 2024

Bionic hand and human hand touching at the fingertips

AI still needs a human touch. Image by cottonbro studio / Pexels licence

When I first (guiltily!) asked AI to write an email for me to achieve a certain purpose, I was shocked at the result. Really? It looked great!

But wait. It wasn’t quite right for my audience, so I set about making some quick tweaks.

A few tweaks to what AI wrote — and more

The words didn’t sound like me in places, but that was an easy fix. The US spelling… oops, I should have told AI about that. I fixed the spelling, and made the text warmer and a bit more direct with just a few changes — a couple of contractions, and some plainer, more familiar words.

I made a couple of sentences clearer (AI went into overdrive with a lengthy explanation) and split up another one.

Then I remembered an important aspect that I’d left out of the prompt. Bother! I added a new paragraph to the draft. Then I wondered if I should adjust the prompt and run it again. So I did. Back came another neatly composed email that covered all the necessary content. It even included headings. Great!

Well, not quite. A second look revealed that I needed to redo the tone tweaks. And this time the content was presented in a different order, and from a slightly different angle.

The tone was sort of okay, but not quite. I had to really think about that. I could also see that a couple of the headings could be much more useful.

And, unhelpfully, AI had introduced an idea that would puzzle my reader in their situation. Sigh!

Finally finished

I got there in the end — and maybe the process was quicker compared with writing from scratch. But what I discovered was that AI still needed me! It needed my brain, my ‘human-ness’, and my knowledge of what complete looks like. It needed an understanding of how my reader might react to certain content.

Most of all, it needed an awareness that, without me, something might feel ‘off’ to my reader.

Conclusion! Writing with AI has limitations

My early experiment confirmed that AI can be a marvellous tool for us humans to use. ‘Tool’ is the operative word though. When we’re using a hammer, a kitchen knife, or a pair of pliers, we are clearly in charge. We choose the tool that best matches what we think the task requires. The tool is only useful with our direction, insight, and control. We usually find that the tool has limitations and have to make compromises. So it is too with AI.

Query AI on this concept and it will self-confess its inherent limitations — at length! Here’s my curated list.

1. Lack of deep contextual understanding of a topic

I’ve seen AI produce text that is technically accurate, but clearly lacking in depth and the unique insights that a human with lived experience would automatically include in any given situation. AI can’t reflect what it doesn’t know.

2. Ignorance of nuances

Writing, particularly in various professional and government contexts, usually requires a nuanced understanding of tone, cultural sensitivities, and specific audience needs. AI rarely copes with such subtleties.

3. Inconsistent quality control and accuracy

AI models are trained on large datasets, but these datasets might not cover all possible scenarios or the most recent developments, leading to outdated or incomplete information. Equally, AI may not have the depth of understanding required for highly specialised fields, resulting in errors or oversimplifications.

I’ve been quite perturbed at various ‘facts’ AI has thrown me, perfectly mixed with the truth on various topics. AI can’t reflect what it hasn’t been trained on, but often joins dots and makes assumptions anyway.

AI-generated content always needs human oversight to verify the accuracy of the content, even (sometimes especially) when it appears to be referenced well.

4. Lack of consistency

Although AI can produce technically correct text, it may not produce text consistent with your own style or house style (unless trained to do so).

Often AI struggles to maintain consistency over long documents or across multiple pieces of content. AI may not always remember previous context or maintain the same style and tone.

5. Bias and ethical concerns

AI can inadvertently perpetuate biases present in the data it was trained on, affecting the overall objectivity and fairness of the text. Recognising bias and ethical concerns is a uniquely human skill.

These concerns, which are so evident now, are not new. But it’s interesting to see them play out in real time at our fingertips. And, while it’s clear that AI cannot stand alone, experience shows that it can be enormously helpful — if we know what we’re doing.

Still needed: the human connection and knowing how to write well

A year or so down the track, many prompts and many ‘tweaks’ like mine later, the answer is clear. To create rich, effective, technically correct content of any kind, AI needs you to:

My so-called tweaks, set out below, are in fact what made my email work.

The real conclusion

AI + human + plain language principles = success!

My tweaks were key to authenticity, credibility, and effectiveness. Unsurprisingly, they neatly reflect the internationally accepted plain language principles embodied in The Write Plain Language Standard.

Applying these principles took a bit of thinking and time, especially 2, 3, and 4.

1. The purpose of the document is clear at the start

I wrapped the purpose up in the prompt.

2. The content supports the purpose of the document

I noticed missing content because my prompt wasn’t specific enough.

3. The structure of the document is clear and logical to the reader

I reordered content — then got two versions from two prompts and had to decide which was best based on what I knew about my audience.

4. The headings signpost the key content

I made some headings more useful, and inserted others.

5. The paragraphs are mostly short and focused on one topic

I edited to make sure the paragraphs were a good length.

6. The sentences are mostly short and straightforward

I fixed a longish sentence that had unnecessary words, and I split up another one.

7. The words are precise and familiar

I changed some complex words to more familiar ones.

8. The tone supports the purpose of the document

I made several changes so the overly formal tone sounded like me.

9. The layout and presentation help the reader absorb the message easily

I changed the font to our house style, tweaked heading styles to make them more visible, and opened up the dense look of the text.

10. The document is error-free and consistent with your style guide

I made several tweaks consistent with our house style, such as adding spaced em dashes and aligning bullet points to the left.

Putting it another way, my tweaks resulted in an email that met the criteria for effectiveness noted by the International Plain Language Federation — that the intended reader can easily find what they need, understand what they find, and use that information.

Read the international definition of plain language

If you want to make AI work, the human factor plus plain language are still key. Plain language helps foster a positive connection between AI and us.

Clearly stated: AI can’t do without us.

Learn how to get the most out of AI

We offer customised AI writing workshops for teams and individuals. Contact us to find out more.

Let’s talk

Insights, tips, and professional development opportunities.

Discover more from Write

Subscribe now to keep reading and get access to the full archive.

Continue reading