I Used AI to Review a Reading Session That Went Sideways. Here’s What It Helped Me See.

A few weeks ago, I wrote about using AI to help build a reading intervention for my kid.

Not to replace the teacher. Not to outsource parenting. Not to pretend I had become a reading specialist because I had a chatbot and good intentions.

The point was almost the opposite.

AI was only useful once I built guardrails around it. And I feel that even more strongly when it comes to my children.

I had to start with what the teacher was actually seeing. I had to learn enough about reading to ask better questions. I had to stop treating Lexile growth as the whole story. I had to build a system that focused on the real gaps: cause and effect, tracking a full text, and using evidence from the story instead of answering from memory.

That was the first lesson.

AI can help, but only if it's connected to something real.

This is the second lesson.

AI may be even more useful after a session goes badly.

Because when you're sitting there with your child, watching a reading session come apart in real time, you don't always understand what you're seeing.

I sure didn't.

My kid recently asked for a preview of where we're trying to get by the end of summer. They're finishing elementary school and heading toward middle school. Their latest STAR assessment came back at 835L and roughly at grade level, which is a major jump from where they were earlier in the year.

But the summer goal isn't just to make the number go up.

It's to help them handle more complex text with more independence. The kind they'll see in middle school. The kind that doesn't just ask whether they can read the words, but whether they can hold the whole thing together.

So when they asked to see what the next grade up might feel like, I let them.

If I'm honest, part of me hoped they'd nail it.

My kid has always been a great reader in the way most people mean that phrase. Can read the words. Can sound fluent. Can enjoy the story.

But processing what they read in a deeper way is different. Cause and effect. Evidence. Mechanism. Why one thing led to another.

That's the part we've been building.

I had AI write a customized Harry Potter-style story at about 1050L. It was explicitly a preview. Not the new normal. Not the next required level.

A glimpse of the target.

The story was called The Chandelier and the Dragon.

Dobby had added candles to a chandelier in the Great Hall and accidentally wrapped the hoisting rope around the cleat in the wrong direction. A correctly wrapped cleat holds by friction. A cleat wrapped the wrong way only holds while the tension stays steady.

Ginny notices the chandelier tilting and pulls the rope gently to level it.

The rope slips.

The chandelier drops.

Fred and George try to help by hauling harder.

That makes it worse.

The rope releases, the chandelier drops farther, a crate tips over, and a dragon-shaped firework erupts into the Great Hall.

It was a great story.

Maybe too great.

For a while, I thought they were doing really well.

My kid was decoding. Was engaged. Liked the dragon. Noticed vocabulary gaps in real time, and asked for "ignited" to be added to the vocabulary list. Connected "listing" to "tilting." Remembered "simultaneously" from an earlier story.

That's part of what made the session hard to read as a parent.

They were struggling and doing well at the same time.

And when your child is doing well, you start to hope.

I did.

I started to imagine the clean version of the story. The proud dad version. They read the preview. They answer the questions. We celebrate the breakthrough. I tell them, "See? You're closer than you thought."

I had already written the ending in my head.

Then we got to the question that asked what Luna identified, what Hermione did, and how we knew it worked.

That was the turn.

At first, it looked like they were getting stuck on page layout. They said part of the paper needed to be somewhere else so they could see both things at once.

I thought they were getting distracted.

I thought they were focusing on the format instead of the question.

And, honestly, part of me wanted to say the very Gen X thing:

Just deal with it.

Middle school isn't going to adjust for you.

The paper is the paper.

You have to work with what you're given.

Some of that came out.

Not my worst parenting moment.

Not my best either.

Then their voice got sharper. Tears started to well up. They shut down for a while.

A long while.

It felt very long.

I sat there thinking: What just happened?

A minute earlier, they were reading about a dragon-shaped firework banking through the Great Hall. They had nailed parts of the reading. They were in the story. They were doing the thing.

Now we were stuck in frustration, silence, and page layout.

In real time, I could see the behavior.

I couldn't see the structure underneath it.

That's where AI helped.

After the session, I gave AI the transcript from the recording and asked it to assess what had happened.

Its answer made the session make sense.

This wasn't a skill failure.

It wasn't laziness.

It wasn't defiance.

It was working memory overload.

That sounds technical, but the practical meaning was simple:

They were trying to hold too much at once. (Hey I know a guy who writes about that very thing)

They had to remember the question.

They had to find the right part of the story.

Track Luna's diagnosis.

Track Hermione's two-step fix.

Understand friction versus tension.

Explain why the wrong fix made the problem worse.

Then identify the proof sentence that showed the real fix worked.

And they had to do all of that while navigating a page layout that suddenly mattered more than I understood.

The page complaint was real.

But it was also carrying something bigger.

What they were really saying, in the only language available to them in that moment, was:

I can't hold all of this in my head and work the page at the same time.

They actually said the key sentence out loud.

"I keep forgetting what the question is."

I heard it.

I didn't understand it.

Not fully.

AI helped me hear it after the fact.

That matters because the parent interpretation determines the parent response.

If I think the problem is focus, I push.

If I think the problem is avoidance, I push engagement.

If I think the problem is attitude, I tighten up.

But if I understand the problem as overload, I scaffold.

Those are not the same response.

And they don't produce the same child.

Eventually, we got through it together.

I remembered what I'd told them at the beginning.

This is a preview.

Expect struggle.

It's okay.

I'll help you when you get stuck.

I had to practice what I keep telling them to practice.

Take a breath.

Not a dramatic one. Not some perfect parenting reset. Just enough of one to stop making the moment worse.

Then I stopped treating the answer like something they needed to produce independently and started working beside them.

I helped them find the lines.

I helped them connect the mechanism.

I helped them see that the chandelier rose because Hermione rewound the rope correctly and it was finally holding by friction instead of unstable tension.

Once I joined them inside the work instead of pushing from outside it, they came back.

That part was me being their dad.

AI will never be that.

But AI did help me see what I almost missed.

They didn't refuse the work.

They didn't decide reading was too hard.

They didn't leave the session defeated.

At the end, they still wanted the next story.

When I told them we were going back to our normal level, they pushed back.

"That's not middle school."

That sentence mattered.

The hard session didn't make them want to quit.

It showed them the gap.

And they still wanted to close it.

That's not something a test score can show you.

The AI review helped me see something else too: the story design had violated one of my own rules.

The cause is supposed to be visible and pointable.

This one was partly visible, but the real mechanism was more abstract than usual. Friction versus tension on a wrongly wrapped cleat is harder to picture than a handle turned the wrong way or a latch blocking a door.

I could explain it with my hands.

I could draw it.

But inside a dense preview, after 200-plus words of harder text, it was too much to ask them to hold the entire mechanism independently.

That doesn't mean the session failed.

It means the session gave us data.

The Lexile was higher.

The vocabulary was more dense.

The mechanism was more abstract.

The question required synthesis.

The page layout added friction at the worst possible moment.

That wasn't one hard thing.

That was five hard things stacked together.

And I didn't see that clearly while I was sitting at the table.

I saw the paper complaint.

I saw the tears.

I saw the shutdown.

I saw my own hope for the session slipping away, which probably made me less patient than I wanted to be.

AI gave me the replay.

Not the parenting. The replay.

It helped me review the transcript after the emotion had passed. It helped separate the visible complaint from the hidden load. It helped me see that the problem wasn't that they couldn't do the work.

The problem was that the design asked too many kinds of work from them at once.

That gives me something useful for the next session.

Not a lowered standard.

A better bridge.

Next time, if the text level goes up, the mechanism needs to be more concrete.

If the mechanism is abstract, the page layout needs to be easier.

If the question requires synthesis, I may need to preview the structure before asking them to produce it.

And if they say, "I can't look at both of them," I need to hear that as instructional data before I hear it as resistance.

That may be the most important lesson.

A child's frustration is not always the enemy of learning.

Sometimes it's data.

Not clean data.

Not polite data.

Not easy-to-interpret data.

But data.

Something is too heavy.

Something is unclear.

Something is stacked wrong.

Something needs a scaffold the child can't name yet.

That's what AI helped me see.

Not instead of parenting.

So I could understand the moment better and parent the next one with more precision.