I thought he was just a late talker — until an AI that knew my son gave me a real answer

By his second birthday, the other toddlers in our playgroup were already stringing words together. Theo wasn't. He'd babble, point at the dog, hand me a cup — but actual words? Only a handful. Everyone had a reason ready: "boys talk later," "he'll get there when he's ready," "my nephew didn't say a thing until he was three." And maybe they were right. But standing in the kitchen one night after he'd gone down, I kept circling the same question I'd already pushed away twice that week: should I be doing something, or am I overthinking this? It wasn't panic. It was that low, shapeless not-knowing that doesn't point anywhere in particular — which is exactly why it wouldn't leave.

"He'll get there" is probably true. It just isn't about your kid.

The reassurance was kind, and it might even have been right. But it was advice for every two-year-old on earth — and I wanted to know about Theo. Every checklist I found online either told me everything was fine or sent me straight to the worst case. Nothing in between. Nothing that actually looked at my own kid — where he was doing well, and the one or two places worth a closer look.

I didn't want a score. I wanted a picture — where he was doing well, and the one or two places worth watching.

A friend told me about a different way to get one. Not a "free diagnosis" — a way to let an AI actually get to know Theo. It took about five minutes, one night between bath and bed: a series of plain questions about what he can already do and what he can't yet, across all six areas of development — not just the talking. At the end I didn't get a number. The AI had learned who my son is.

So I ran the test that convinced me. I took one question and asked it in two places: "My 2-year-old barely says any words — what should I do?"

I asked a generic AI:

Every child develops at their own pace, and plenty of kids talk later without it being a problem. Give him lots of input: read to him every day, narrate what you're doing, name objects, sing songs. Keep an eye on his progress over the next few months, and if the speech still worries you, bring it up with your pediatrician.

None of that is wrong. It's good advice. But it's the same advice for any two-year-old on the planet — there's no Theo in it.

I asked firstk, which already knew Theo:

Theo is doing well on gross motor, fine motor, and cognitive — all within range for his age. The area to watch is expressive language: from what you told me, he still uses only a few words with clear intent and rarely points to ask for what he wants. For the next two weeks, focus on giving him reasons to talk. At bath time, name one thing at a time — "water," "foot," "duck" — and wait about five seconds, looking right at him, before you hand it over. At meals, offer a spoken choice — "water or milk?" — and hold the cup until he tries. The science behind this (the Denver II, one of the frameworks pediatricians actually use) treats language as a skill you draw out with opportunities, not a race he's losing.

The difference wasn't that the AI was smarter. It was that my son's name was inside the answer.

Don't believe me? Paste that same question into your own ChatGPT and compare the two.

I felt my shoulders drop in a way I hadn't realized they'd been up — not because anyone promised me everything was perfect, but because, for the first time in months, the advice had edges, and Theo was inside them. The next morning, at bath time, I held up the duck and waited. He didn't say it. I held it up again. And he looked at me like he'd just figured out the rules of a game.

And I know what the more skeptical version of me would have asked, because I did ask it: is this just one more thing trying to sell me something? It's free, it took about five minutes, and not once did it tell me Theo was "behind" or hang a label on him — it isn't a diagnosis, and it said so plainly: if something ever truly worried me, that's a conversation for his pediatrician, not an app. What it gave me instead was the one thing all those late-night searches never had — a next step with his name on it.

If you carry that same question at the end of the day, the next step is a small one: let the AI get to know your kid, and trade the not-knowing for a picture with his name in it.