· Early Childhood Teacher
About half of our families are second-language English speakers, and one afternoon I watched a beautifully written daily report go home and land with nobody. The parent smiled, nodded, took the printout. From her face I could tell she had not read a word of it. That moment is why I now use AI for parent communication when families don’t read English at home, but only as a drafting step, never as the thing that reaches a family. Here is the workflow I actually use, the words AI gets wrong for our community, and the line I will not cross.
The day a daily report went home and nobody read it
Our room sits across baby, toddler, and pre-school groups, and on any given week the home languages shift. We have families who speak Arabic, Mandarin, Vietnamese, Dari, and Tongan at home. Roughly half of our enrolled families do not use English as their first language. That is not unusual for Sydney. It is the room I work in every day.
The report that went home that afternoon was fine. Warm, specific, full of the small wins a parent wants. The problem was never the writing. It was that I had written it for a reader who does not read English, and no amount of polish fixes that.
”They don’t speak English” and “they don’t read English” are two different problems
This is the bit most advice misses. A family can speak conversational English at pick-up and still not read a paragraph of it. We also have families from oral-culture backgrounds where reading any language at home is not the norm. So “translate it” is only half an answer. Sometimes the fix is a different language. Sometimes it is just shorter sentences, and sometimes it is voice, not text at all.
I started sorting families into three buckets in my own head: needs another language, needs plainer English, needs to hear it rather than read it. The tool I reach for depends on which bucket, and most of the time it is more than one.
Translation tools for parent messages: what I actually noticed
I tested the free translation route first, the kind that swaps phrase for phrase. For a one-line “pick-up at 3pm today” it was fine. Anything with warmth in it fell apart. The tone went flat and institutional, and a couple of times the meaning drifted in ways a parent would not catch but I would wince at.
Where phrase-by-phrase translation breaks
Early-childhood writing is full of soft, relational language. “She had a big day and needs an early night.” “He is settling in beautifully.” Run that through a literal translator and the warmth is the first casualty. The words arrive; the reassurance does not.
The accuracy figure that made me cautious
There is research behind my caution, not just a hunch. One study of automated translation in parent communication found accuracy hovering around three-quarters, with errors concentrated in exactly the nuanced, idiomatic phrasing we use most (Journal of Latinos and Education, via Taylor & Francis). Three-quarters. For a “fun day” note, a one-in-four wobble is survivable. For a message about a child being unwell, it is not.
A plain-language workflow I use for routine comms
Here is the practical part. For routine, non-sensitive comms (daily reports, newsletters, reminders, “bring a hat tomorrow”) I use an AI assistant as a drafting partner, and I tell it three things every time.
First, write at a Year-6 reading level. That single instruction does more for access than any translation step, because a plain English source translates far more cleanly than a flowery one. Short sentences. One idea per line. No idioms.
Second, I prepare a translation handoff prompt for the interpreter colleagues we work with. The AI helps me produce a clean, simplified English version and a note flagging which phrases are culturally loaded, so the human doing the actual translation is not guessing. I am an ECT, not a translator. The handoff is the point.
Third, and this is the catch nobody warns you about: cultural reframes do not translate. Phrases like “tummy time” or “settling in” are not universal concepts (I learned that the hard way after a parent thought “settling in” meant her son was unhappy). So I keep a running list of words AI gets wrong for our community and I pre-substitute them before the draft goes anywhere near a family or an interpreter. “Settling in” becomes “getting used to the room.” “Tummy time” becomes a plain description of the activity. That list has saved me more grief than any single tool.
Setting up a family-language profile once
I keep a simple note per family: home language, whether they read it, and preferred format. Set up once at enrolment, updated when things change. Then the AI draft gets pointed at the right output before I write a single line. The rest of how I structure my documentation is in writing child observations with AI without losing your voice.
Batching daily reports and newsletters
For newsletters I draft the English once at a Year-6 level, then run the plain version out to the handoff step. Doing it in a batch rather than family by family is where the time actually comes back. Over a fortnight of newsletters and reports I clawed back close to 40 minutes a week that used to go into rewording the same message five ways. That time goes straight back to the children.
Audio for families who don’t read at all
For families in the “needs to hear it” bucket, text translation misses the point entirely. We use short voice messages through our family-engagement platform instead, sometimes recorded by a bilingual colleague. AI helps me script a tight 30-second version in plain English first; the spoken delivery does the rest. More on the drafting side of this in how AI wrote the parent newsletter and the prompts I use to sound like me.
What I will not hand to AI
This is the section I want educators to take away even if they skim the rest. Some messages never touch an AI draft, and never go out on a machine translation. Ask your service about its own policy, but in my room the list is firm.
- Incident and injury reports. Anything where a child was hurt goes through a human, and where a family needs it in another language, it goes to a qualified interpreter, not a chatbot.
- Custody, consent, and legal or enrolment documents. The stakes and the precision are too high.
- Health and medical messages where a misread word changes what a parent does.
- Anything a certified interpreter should rightly handle.
The pattern is simple. If the AI being subtly wrong could hurt a child, or mislead a parent, or drop the service into a compliance hole, it does not get the job. A friendly Friday newsletter is fair game. An injury report is not.
My honest opinion: the risk is not bad translation, it’s false confidence
Here is where I will take a stance. The danger with these tools is not the occasional clumsy translation. You can spot a clumsy translation a mile off. The real problem? The smooth one that is quietly, confidently wrong, because a fluent message that has lost its meaning looks exactly like a fluent message that kept it.
A parent reading a polished paragraph has no way of knowing that “settling in” came out the other side as something closer to “moving house.” It reads fine. They act on it. The harm is invisible until it isn’t. That is why my workflow puts a human at the end of every message that matters, and why I treat AI as a drafting tool that earns me time, not a translator that earns my trust. Plenty of platforms now advertise one-tap home-language messaging. I use those features for low-stakes notes and I keep a human between them and anything that counts.
Where this fits EYLF and NQS expectations
None of this is really a tech project, if I’m honest. It is the everyday work of respecting the families standing in front of you. The Early Years Learning Framework names secure, respectful and reciprocal relationships with families, and respect for diversity, as core principles of practice (ACECQA, Belonging, Being and Becoming: the Early Years Learning Framework V2.0). Communicating in a way a family can actually receive is what that principle looks like on a Tuesday.
The first move is not a tool at all. It is asking each family, at enrolment, what language and what format works for them, and writing it down. Respect for diversity is practice, not a poster on the wall. The AI does not change that part. It just gets the plain-English draft to me faster, and I keep a human on the parts that count. I unpack my broader “where I won’t automate” rule in drafting EYLF learning stories with AI and the format I keep.
TL;DR / Key Takeaways
- Using AI for parent communication when families don’t read English works best as a drafting step: plain-English first, human or interpreter last.
- “Don’t speak English” and “don’t read English” are different problems; the second needs plainer text or audio, not just translation.
- Machine translation accuracy for family-facing text has been measured around 75%, with errors in exactly the warm, idiomatic phrasing educators use most.
- Cultural reframes like “tummy time” and “settling in” do not translate; keep a running substitution list for your community and fix the draft first.
- Never machine-translate incident reports, custody, consent, or health messages. Ask your service’s policy and use a qualified interpreter for anything sensitive.
Sources
- ACECQA. Belonging, Being and Becoming: The Early Years Learning Framework for Australia V2.0 (2022). https://www.acecqa.gov.au/sites/default/files/2023-01/EYLF-2022-V2.0.pdf
- Journal of Latinos and Education (Taylor & Francis). Communicating with Parents of Latino English Language Learners: Is Google Translate an Accurate Option? DOI 10.1080/15348431.2022.2104849. https://www.tandfonline.com/doi/abs/10.1080/15348431.2022.2104849
Megan is an ACECQA-registered Early Childhood Teacher (ECT) working across baby, toddler, and pre-school rooms in Sydney, with 5+ years of experience. This is a personal-practice reflection, not professional or legal advice. Always confirm your communication, translation, and privacy obligations with your own service’s policies and a qualified interpreter where required. Last reviewed: 7 June 2026.
Fact-checked 2026-06-07. Last reviewed 2026-06-07.