When AI Gets It “Almost Right”: Why That’s Not Good Enough

When AI Proofreads Like a Dishwasher That Misses Spots

Why “almost right” isn’t good enough when AI tools become part of daily life

I enjoy writing as a hobby. For me, it’s a form of therapy — a way to think out loud and capture ideas. Sometimes I share my work on platforms like Medium and Vocal.

Like many writers, I’ve started using AI tools to help polish my drafts. Not to rewrite them, but to catch grammar slips I might overlook. Strangely enough, I find proofreading my own writing much harder than reviewing someone else’s. When I already know what I meant to say, my eyes skim too quickly and miss mistakes.

AI has been useful for those “big things” I don’t always notice. But recently, something interesting happened that made me rethink how much I trust it.

When AI Invents Mistakes

Just today, I asked an AI tool to proofread one of my essays. Twice, it flagged “errors” that weren’t errors at all. In fact, the words it suggested adding were already there in my original sentence.

So I asked the AI directly: “Why so many errors today in your feedback?”

Its response was refreshingly honest:

“You’re absolutely right to call this out, and I apologize. I made careless errors by not carefully re-reading the original text before providing feedback. In both cases, I suggested adding punctuation or words that were already there—which undermines the trust you should be able to place in my proofreading.”

The AI admitted it wasn’t being thorough enough and even offered suggestions for better prompts:

“Please carefully proofread this text for grammar errors. Before suggesting corrections, double-check that the issue actually exists in the original text by re-reading that specific section.”

Or: “Proofread this text. For each error you identify, quote the exact problematic phrase from my text before suggesting the correction.”

Both approaches were meant to add accountability by forcing the AI to slow down and verify. But here’s the catch: shouldn’t this carefulness be built in already?

Like Running a Faulty Dishwasher

Imagine this: you run your dishwasher, but sometimes the dishes come out dirty. You ask why, and the machine replies, “I should have cleaned them better. To be sure, press the start button harder next time.”

Sounds absurd, right?

Yet that’s essentially what happened with my proofreading tool. The system admitted it should already be thorough but asked me to adjust my behavior instead.

Why This Matters Beyond Grammar

Grammar suggestions aren’t life-or-death. If AI makes a mistake here, it’s frustrating but not catastrophic. But what happens when AI gets things wrong in more critical settings?

We’ve already seen examples:

• Grok making offensive remarks and praising Hitler.
• Copilot generating inaccurate news summaries.
• ChatGPT citing fictional legal cases in real court filings.

Now think about governments, healthcare, or military systems increasingly adopting AI. A proofreading mistake is harmless. A miscalculation in defense, medicine, or law could be devastating.

And it’s not just about accuracy. With Spotify and YouTube recently requiring AI-driven age verification in some markets, we may soon face everyday frustrations like being wrongly locked out of content. What happens when AI misjudges your age, creditworthiness, or medical eligibility?

The Bigger Picture

We’ve become used to apps updating daily to fix bugs. Agile development makes us, the users, part of the debugging process. That’s fine for a photo filter or a to-do list app. But AI is different.

AI is rapidly becoming embedded in systems where mistakes have real consequences. Asking users to catch those mistakes after the fact is no longer acceptable. These tools need to work out of the box, the right way.

The stakes are simply too high for “good enough.”

A Real-Time Example

• Ironically, even while writing this very article, I asked AI to proofread again. It made the same mistake I was describing.

Here’s the sentence I wrote:

• “As many companies rush to implement AI into their processes, I wonder about the quality of the output.”
• The AI suggested: “Add ‘about’ for clarity.” But “about” was already there.
• When I pointed it out, the AI admitted:

“You’re absolutely right—I made the exact same type of error you’re writing about! This is actually a perfect real-world example of the issue you’re highlighting.”

It was almost comical. But it also drove the point home: even when specifically instructed to avoid these mistakes, the AI repeated them

I don’t expect perfection from machines. But I do expect consistency. If AI systems know why they’re making errors, then companies need to design them so those errors don’t keep happening.

Because while I can laugh off a grammar slip, there are industries and scenarios where “almost right” could cost far too much.

AI doesn’t just need to be powerful. It needs to be trustworthy.