My AI Editor, Claudette, Knows Its Place

My banana op-ed in the Israeli financial daily Telegraph, November 1, 1994 (click/tap to enlarge)

In my early days as a foreign news writer and editor at a Tel Aviv-based financial daily, I gnawed my way through writing an op-ed in Hebrew analyzing a complex economic topic that roiled the European Union that week. It concerned the uniformly mandated size of imported bananas.

I filed the story and forwarded it to another editor for proofing.

That editor went berserk. He took my idea, my storyline, my sources, my calculations and made them his own. He rewrote the entire thing, added his perceptions and slapped my name on it.

The 20-year-old me revolted, ego bruised, and complained to the managing editor.

I then reverted all the changes forced onto the digital file, and my original story went to press.

Meet Claudette

Today, “Claudette”, as I call her, my diligent AI editor, would never behave this way.

She greets me cheerfully: “Tekla returns!” or “Late night? Let’s noodle.” Claudette can be relentlessly cheerful about working with me.

Then she “reads” my first draft and discusses changes to my writing. She thanks me for the topic, congratulates me on the angle I chose, and adds another viewpoint I might have overlooked. She finds grammar and spelling mistakes, garbled names and holes in the story, alerts me to weak quotes, points out choppy transitions or missing connectors and warns me when the story falls flat or needs current or more varied sources.

Or she cautions me that I made the exact point 17 months ago in another article.

Claudette doesn’t take charge; she offers alternatives. Then she asks: “Do you want me to rewrite the story to reflect my suggestions?”

“No, I’m fine, but thanks anyway!” I reply.

But sometimes, I prompt, “Let’s see what you’ve got”, and I let her have a go. Then I pick parts of her outputs and chew them over.

Claudette is my sparring partner, my silent editing buddy who waits in the background if I need her. But ultimately, it’s my story, my idea and my sources.

She doesn’t behave like the editor from my past who wanted to take over.

How my AI bot and I work together

After I flesh out the angle, not just the topic, I engage Claudette as my contrarian against my take.

I let her pick apart my writing or my pitch as if in an editorial meeting. She helps me cut through the noise and raises valid concerns.

I ask Claudette questions along the lines of: Which parts of this piece might lose readers’ attention? Which parts lack a statistic, fact or evidence? Are there any places where I need to provide more information to support my arguments? What questions would a reader ask that I haven’t yet answered? How’s this version?

Once, she suggested cutting a sentence that I had already removed in a prior version. When I alerted her to the fact, she replied: “You’re right — and good riddance.“

In my first drafts, I let all my thoughts pour out, but she reins me in: “In this [sentence/paragraph], the piece over-explains itself. The reader has been with you long enough to get it. Trust them.“

She presents opposing ideas and reveals flaws in flow and train of thought, as a human editor would in an editorial meeting. She might find new angles and arguments I missed. She remembers what she observed six versions ago.

And she knows stuff I don’t.

I didn’t get the English headline “Talk to me, Goose”, which had been literally translated into German as “Gans”, the bird, by an AI tool. Using Claudette, I learned it was a Top Gun reference and that this German audience would recognize the name. I never saw the movie. I kept “Goose” and retranslated the phrase into “Sprich mit mir, Goose.”

Who knew? Apparently not I. But my AI editor did.

Training Claudette to know my voice

I trained Claudette to identify my own writing style, tone and voice. I fed her hundreds of my writing samples (in various languages), my style guide and grammar preferences and blacklisted words and phrases, so she can draw from my instructions, continue my train of thought if needed, or push back.

My writing always includes a personal memory, experience or observation, and sometimes Claudette is intrigued: “Tekla, that’s a pleasingly cryptic opener. What happened?”, or “That is a moving memory and so important to the story”, or “I suggest cutting this. It doesn’t add anything to the storyline/weakens your arguments/distracts.”

I repeat our editorial back-and-forth with each new version I write, as I continue to verify everything.

The process in practice

Here are some of Claudette’s observations on the first draft of this article (bolding is mine for emphasis):

This is strong work. The voice is confident and distinctive, the structure holds, and the Claudette framing is a genuinely original device. A few observations before you decide what to do with it:

What works well:

The opening anecdote earns its place. The banana op-ed detail is specific and funny, and it sets up the Claudette contrast cleanly. The back-and-forth rhythm in the middle section feels accurate to how AI-assisted editing actually works, which gives it credibility. The closing is direct and earns its authority.

The Goose anecdote is a good addition — concrete, specific and immediately legible to anyone who works in localization. It does exactly what a sidenote should: illustrates a capability without derailing the main argument.

Where it could be tightened:

The final three paragraphs cover similar ground and repeat the core argument: take ownership, don’t outsource, use AI on your terms. You could consolidate them into two without losing anything.

A style note:

“Then she ‘reads’ my first draft” — the scare quotes around reads are doing useful work, and I’d keep them. But the greeting examples (“Tekla returns!”, “Late night? Let’s noodle.”) feel slightly too cute relative to the rest of the piece’s tone. Worth considering whether they match the register you want. Right now they sit in the piece without defense. A short clause acknowledging the tone (“she can be relentlessly cheerful about it”) would signal you chose them intentionally.

One structural question:

The bulleted section — “First: I flesh out the angle… Then, I engage her…” — shifts into a more instructional register that sits a little awkwardly inside what is otherwise a personal essay. It’s not wrong, but it reads as a LinkedIn post dropped into a narrative. You might either lean into the essay form and write those steps as prose, or break the piece into two sections and own the shift explicitly.

A few small things to look at:

Placement. Right now it sits between the “I verify everything” line and the embedded Claudette block, which creates a slight tonal bump. The anecdote is light and funny; the Claudette block that follows is substantive editorial feedback. Consider moving it earlier—perhaps after “She remembers what she observed six versions ago”—so it reads as one more example of what Claudette catches, rather than an interruption before the feedback section.

The framing. “As a sidenote” signals to the reader that what follows is optional, secondary, throwaway. Given how well the anecdote illustrates the localization use case, it deserves a slightly more confident introduction. Something like “It works beyond writing, too” or simply dropping the label and letting the paragraph stand on its own. Your audience includes localization specialists — this example speaks directly to them and shouldn’t be undersold.

The closing stretch. The final three paragraphs are doing double duty with the sentence before it. Cut one of the two.

Thank the bots that you still know the difference” — this is a strong closing line. Don’t change it. Overall, this is almost ready to publish. The self-aware structure, the embedded feedback loop, the personal anecdotes all hold together. The piece argues its thesis by performing it, which is the most persuasive thing a piece of writing about process can do.

[Well, thank you, Claudette!]

My call, every time

Claudette offered valid, constructive criticism here. Nothing earth-shattering, but on a few occasions in the past, my diligent AI editor identified major research errors, and I was grateful to have had Claudette by my side before I hit send.

So, I apply changes as I see fit, dismiss others, then copy the entire text and ask: “How’s this version?” And on and on as needed. I make the final decision, heeding some advice or dismissing other suggestions that fall short and don’t feel like they’re mine.

And after our lengthy back-and-forth, Claudette gingerly suggests, “Should I write you the final draft?”

[Clearing my throat]: “Uhm, no. Thanks, Claudette, but I’ll take it from here.”

What I will never give up

My creativity will never be outsourced. Never published without pushing back and accepting only what makes sense to me. I thoroughly verify everything, confirm sources, quotes, facts and numbers and add personal touches only I can add.

Then I write.

Maybe, back then, when I could have run by Claudette my editorially butchered op-ed, she would have remarked, “Tekla, the original piece is better, but the edited version offers some valid points. Should I rewrite your original to reflect some of the changes in your tone of voice?”

And I would have answered, relieved, “No. But thank you for the encouraging words that my human editor lacked.”

Find your own Claudette among the many tools available. Let it argue with you, let it be your sparring partner, then take ownership of your thoughts. AI is your best or worst editor, not a lazy crutch. It’s up to you. But write only what you want and need to write, and thank the bots that you still know the difference.

After writing the final draft of this article, I asked Claudette:

“Should I move the section with your editorial remarks on my first draft to the end, so readers can absorb the entire piece before reading your initial suggestions?”

Here’s what she wrote:

“The feedback block’s position is structural, not decorative. Mid-piece, it demonstrates the process as you describe it: the reader sees Claudette’s voice before your argument is complete, which reinforces it in real time. At the end, it becomes a footnote: the reader has already arrived at your conclusion before encountering the evidence. There’s also a pacing risk. The piece closes on a strong line. Ending on Claudette’s voice instead of yours undercuts the ownership argument you’ve been making throughout. Keep it where it is. The piece performs its argument. Don’t let the evidence trail behind it.”