· Early Childhood Teacher
It was Friday afternoon at 5:40pm and the cursor had been blinking on a blank newsletter template for twenty minutes.
Two toddlers were still in the room, both clinging to a worm-farm story they were not finished telling. I had an allergy table to update, a transition note to write for a child moving up, and a week of worm farms and herb-planting and recycling-sorting that somehow had to become something eighteen families would actually want to read. My brain was tapping out.
My partner had spent the last fortnight building a small project with an AI coding assistant. I’d been hearing about it every dinner. If he can have an AI write the code, I thought, maybe I can have an AI write the newsletter.
I opened an AI chat tool. Four prompts later, it wrote in my voice. This post is what happened in between.
What we write, and why this one was hard
I write two things on repeat for the families in our toddler room. A weekly newsletter: one document going to eighteen families every Friday before I leave. And individual learning observations: one per child every fortnight, posted to each family via OWNA. The newsletter takes about 45 minutes if I’m flowing, longer if I’m not. The observations are four to five hours per cycle, eighteen kids deep.
This particular Friday everything piled at once. Updated allergy table. Transition note. Sustainability week (worm farms, herb planting, recycling sort) written up in a way that captured what the kids actually did. The blank template won. So I tried the AI.
V1 — esteemed families
The first prompt was lazy. Plain English, no examples:
Write a weekly newsletter for parents of toddlers (2-3 years old) about our sustainability week. We did worm farms, planted herbs, sorted recycling. Make it warm and friendly.
The opening sentence came back:
Dear esteemed families, we are thrilled to share an enriching week of sustainable learning experiences with your remarkable little ones…
I closed the tab for a second and opened it again. Families in our room don’t get addressed as “esteemed families.” Nobody has a “remarkable little one.” Read it aloud and you hear a corporate childcare brand sending a quarterly campaign, not an educator writing to people whose kids she sees at drop-off every morning.
Any family reading V1 would have known I didn’t write it.
V2 — stripped the buzzwords
Round two I added one line:
Tone: casual, warm, like I’m talking to friends. No corporate buzzwords. No phrases like “esteemed families” or “remarkable little ones”. Australian English.
V2 opened with “Hi everyone” which was progress. But the body still felt like a brochure:
The children embarked on an exciting journey of discovery this week. The worm farm became a hub of fascination and inquiry.
No teacher I know has ever called a worm farm a “hub of fascination.” Also: nobody embarks on anything in a toddler room. That word went on the banned list along with several others.
The sentences were still too symmetric. Every paragraph the same length, no fragments, no breath. It was the kind of writing nobody objects to and therefore nobody reads.
V3 — short sentences, named things
I went specific with the don’ts:
Don’t use these words: embark, journey, hub, foster, nurture, cultivate, holistic, enriching. Use short sentences. It’s OK to have one-line paragraphs. Mention specific things kids did or said.
V3 was closer. Sentences got shorter. The banned words were gone. The newsletter was readable as English, not marketing copy.
But it still wasn’t me. It wrote “We had so much fun!” twice, and I never write that. I don’t use exclamation marks in newsletters. Not because I’m dour. Because performative cheer reads as the opposite of warmth, especially from somebody who works with kids. The families know me. They know whether I had fun. I don’t need to tell them at high volume.
V3 read like a generic friendly educator. The generic part was the problem.
V4 — paste in three of my old newsletters
V3 to V4 was the change that actually worked. I scrolled through OWNA, copied three of my past newsletters into the chat (different weeks, different topics, all mine), and added:
Here are 3 newsletters I wrote myself in previous weeks. Match my voice exactly — sentence length, paragraph rhythm, what I do and don’t say. I never use exclamation marks. I always name 2-3 specific things kids said or did. I never say “fun” — I describe what made it interesting.
V4 clicked. Short sentences. No exclamation marks. A concrete moment in the middle, from a Wednesday: one of the boys had told me the worms “tickled his fingers,” and the model kept that line. The ending didn’t try to wring sentiment out of a worm farm. It just said what we’d do next week.
I read it twice, changed two words, and sent it.
V1 sounded like a campaign from a national childcare brand. V4 sounded like the note I’d have written if I weren’t tired.
The numbers from this exercise
- First attempt total time, including the four-prompt iteration: about 35 minutes. So that first newsletter saved me roughly 10 minutes net.
- The next week I reused the V4 prompt with only the content changed, and the whole thing took 12-15 minutes. That’s the real save.
- About 7 to 8 message turns between V1 and V4.
- I kept roughly 85% of V4 and rewrote about 15%, usually two or three word choices and one sentence I wanted phrased differently.
- I’ve used the V4 prompt five times now for the weekly newsletter.
- I also tried it three times for individual learning observations. All three drafts I deleted. More on that next.
The 30 minutes I save on Fridays now goes into next week’s curriculum prep. That part is hard to measure but it’s the part that matters.
The thing I won’t use AI for: individual observations
Most of the ECTs I know take one of two positions on AI. Either it’s lazy and educators who use it are cutting corners on documentation, or it’s a productivity tool you’d be silly not to use. I don’t sit on either side.
The newsletter is room-level. Eighteen families. The job is to aggregate information in a warm voice. Both of those things the AI can do at least as well as me on a tired Friday.
The individual observation is something else. It’s my attention on one specific child in one specific moment. How they held the paint brush. Who they shared their banana with. What they said to me on the sleep mat. the AI doesn’t know any of that, because the AI wasn’t in the room. When I asked it to “write an observation aligned to EYLF Outcome 5 about a three-year-old learning to share,” it generated something that sounded like an observation but said nothing of substance: “demonstrated emerging skills in social interaction through cooperative play,” and so on. The old textbook phrases that pass a documentation audit under the NQF but tell the family nothing they couldn’t have guessed.
That kind of fill-in-the-blank documentation has been killing learning stories for years. It’s not the AI’s fault. Educators have been writing observations like that for decades, on paper, in Word, in OWNA. The difference is that the AI can produce 18 of them in 4 minutes, and the temptation to ship that and reclaim a Saturday is real. I deleted all three drafts because they would have been worse than no observation at all.
So here is my position, and I expect to argue this with colleagues:
AI can replace communication tasks. AI cannot replace attention tasks. The time saved on the first should be spent on the second.
Newsletters, room updates, event announcements — all communication. Use the AI. Save the time.
Individual observations, learning stories, family conversations — all attention. Don’t use the AI. Do them properly with the time you saved.
I’m aware Queensland’s recent position on AI in ECE takes a careful middle line on this and I broadly agree. It’s not the tool that’s the problem. It’s what you stop noticing when you let the tool do the noticing for you.
Three things I’m keeping
- Show, don’t describe. The voice prompt that worked was built on three of my past newsletters, not a description of my voice. Showing is more efficient than telling, with LLMs as much as with kids.
- Banned-words lists move faster than tone instructions. “Don’t say embark” got me closer than “be casual” ever did.
- Some tasks should not be optimised away. The clue is whether the doing is the value, or the output is. Newsletters: the output is the value. Observations: the doing is the value.
The next post in this series will be the individual learning observations problem, long-form. I’ll show what I tried, what didn’t work, and why I’m not going back to the AI for that one.
TL;DR
- Tried the AI on the weekly newsletter at the end of a long Friday.
- V1 was a corporate brand campaign with “esteemed families.” Useless.
- V4 worked once I pasted three of my own past newsletters and told it to match.
- About 30 minutes saved per week after the first time.
- Tried the AI on individual observations too. Deleted all three. Won’t do it again.
- AI can replace communication tasks. It cannot replace attention tasks.