I Asked Claude the Question I Could Never Ask My Boss

Was this newsletter forwarded to you? Sign up to get it in your inbox.

Help us scale the only subscription you need to stay at the edge of AI. Explore open roles at Every.

I operate on the baseline assumption that I’m about to be fired at all times. It doesn’t matter how many managers tell me I’m doing great or how many positive performance reviews I receive, every piece of feedback gets filtered through my self-doubt.

I had asked AI about my career before, but never about my job performance. So when I finally did, I didn’t expect to believe its answer anymore than I believed when a human said I was doing okay.

It started as routine year-end planning at Every in mid-December. Kate Lee, our editor in chief, asked me to put together 2026 goals based on the performance of my articles. My entire career, I’d been told to use numbers to show results—in performance reviews, in job interviews—but I’d never had the data fluency to do that.

Yet when I fed the fourth quarter numbers on my articles to Claude and ChatGPT and started to see proof that my work was making an impact in terms of driving traffic and sheer output, some dormant part of my brain activated like a sleeper agent, and suddenly I was three hours deep in spreadsheets.

AI helped me do something I’d never managed on my own: believe I’m good at my job. If AI can democratize business intelligence for someone with my particular brand of professional self-loathing, it can probably help you understand your own value, too. Here’s how I ran the analysis.

Yearly review gave me proof of my performance

Step one of my AI analysis: manually exporting data from Every’s content management system into Google Sheets like it was 2009. Not exactly the future we were promised, but a necessary evil.

I uploaded the spreadsheets to Claude with this prompt: “I’ve given you four different sets of data from the Every newsletter: overall performance for Q4, as well as performance for two columns I am involved with... I’d like us to conduct a thorough retrospective on my contributions to the Every ecosystem that we can use as a basis for 2026 planning.”

I discovered that I was driving a third of the fourth quarter’s traffic with a fifth of the content. My Working Overtime column was running 13 points above Every’s average satisfaction rating as measured by the ratings that readers can give our articles at the end of every post.

A normal person might have called it there. But at this point I was obsessed with seeing my own performance reflected back to me in the numbers. So I pulled the full year’s data and gave Claude a new role: “Act as an editorial analyst and strategist. Tell me everything you can about what this data tells us.” If the fourth quarter was a fluke, I thought, the full year would expose it.

It didn’t. In 2025, I wrote 54 articles for Every—15 percent of everything we published. Those pieces drove 25 to 27 percent of our subscription trials and web views. In the final quarter alone, I contributed 18.8 percent of content output and 29.3 percent of views. Every way I sliced it, I was punching above my weight by a factor of 1.5 to two.

But the more interesting finding was how different types of work drove different kinds of value. Each column I contribute to at Every moves a different needle.

Vibe Checks, our day-zero reviews of new models, drive trials to Every. When someone’s searching for information about a new AI tool and lands on our coverage, they’re primed to subscribe. Pieces about what our engineers are building pull in traffic from people interested in the frontier of software engineering. My own column about AI’s impact on my work experience drives something harder to quantify: reader satisfaction, as reflected by the ratings at the bottom of each Every essay. According to the numbers, I outperformed in all three.

I didn’t believe it. LLMs have a reputation for sycophancy—telling you what you want to hear—so I wanted to test on another model to make sure that Claude wasn’t just buttering me up. But ChatGPT gave me the same verdict based on the same data, and pushed back when I asked whether my performance had been a virtue of writing about popular topics or producing work that anyone could have produced.

“You’re pattern-matching to an explanation that protects you from having to believe the data,” it told me.

Did the data mean I was good at my job?

I typed it before I could talk myself out of it: “Does this mean I’m good at my job?”

ChatGPT’s answer was immediate and unequivocal: Yes. It walked through the evidence point by point: the output-to-impact ratio, the pattern of high performers, the fact that pieces with more vulnerability to them—like the one about bipolar disorder or the piece about using AI as a career coach after I was fired—-consistently outperformed content that was pure analysis without the context of my experience. ChatGPT had a counterargument whenever I tried to prove it wrong.

“What about the fact that I had good source material to work with?”

ChatGPT replied that it was a credit to my editorial judgment, not a discount on the work.

“But anyone could have written those Vibe Checks—”

ChatGPT reminded me that I wrote 18 percent of content in the fourth quarter and drove 29 percent of views.

I went back to Claude and asked the same question. Claude is more willing to push back and more likely to tell you when you’re fooling yourself. I figured if anyone was going to poke holes in this, it would be Claude. It returned the same answer.

At one point, the AI asked me point blank: Why are you asking this? What is it about it that’s so hard to believe? That’s when I said the thing I’d never said to a manager or a therapist or even, really, to myself:

“I don’t know. I guess just pathological self-hatred and an inability to trust that I do good work.”

I could say it to a chatbot because there was no social cost, and I didn’t have to be worried about burdening someone on the other end with my baggage. If I believed the data, the next step was to believe myself.

Claude proved that being me paid off

I went back to Claude with a new prompt: “Let’s hold up the Working Overtime performance overall with the other content I was involved with in Q4 and think hard about what drives the most value for Every.”

Before giving me an answer, Claude asked me questions I hadn’t thought to ask myself: What’s Every’s current constraint—new subscriber acquisition or retention? What’s your role supposed to be? What can you realistically sustain? High-vulnerability writing is more draining than tool reviews. What’s the right mix for your energy?

Then it came back with scenarios:

1. Option A: You’re a conversion engine. Focus on Vibe Checks and Source Code [our columns about what Every is building]. Maximize trials. Working Overtime becomes a monthly luxury. The business case is clear: more Vibe Checks equals more subscribers.
2. Option B: You’re a brand builder. Focus on Working Overtime. Build the most devoted audience segment. Accept lower trials but create readers who will follow Every anywhere. The bet: Satisfaction compounds into long-term loyalty.

Kate had already told me that we want to publish more of my column in 2026. But I wasn’t prepared to accept it. Coming from a person, I would have brushed it off as encouragement, the kind of thing a good manager says to keep you motivated. Coming from a system that had just walked me through the data and played out scenarios for what each choice would mean, it landed differently.

The data backed this up. When I looked at which pieces performed best—earned high ratings and sparked reader responses—the pieces where I was most myself, most willing to be weird or vulnerable or uncertain, consistently outperformed the ones where I played it safe.

I’d convinced myself that the personal stuff was indulgent—a risk I was getting away with, not a strategy that was working. The data said otherwise.

Build your own case

The doubt is still there, lurking, ready to flare up the next time a piece underperforms or an edit comes back marked up with a hundred comments. But my relationship to the doubt has shifted because now I have the data.

I can look at the numbers instead of believing the narrative that my mind tells me. Instead of panicking, I can spend time thinking about where I should be spending my time and where I can do even more of my best work.

This also solves the problem with most performance reviews that act as lagging indicators of performance—summaries of human impressions, filtered through whatever your manager happened to remember. This does not have to be the case with AI.

You don’t need to be a data analyst. You don’t need special access or technical skills. You need a dataset—whatever metrics you can get your hands on, so don’t be afraid to ask—and a willingness to ask the questions you’ve been avoiding.

Don’t wait for someone else to tell you your value. Build the case yourself.

Katie Parrott is a staff writer and AI editorial lead at Every. You can read more of her work in her newsletter.

To read more essays like this, subscribe to Every, and follow us on X at @every and on LinkedIn.

We build AI tools for readers like you. Write brilliantly with Spiral. Organize files automatically with Sparkle. Deliver yourself from email with Cora. Dictate effortlessly with Monologue.

We also do AI training, adoption, and innovation for companies. Work with us to bring AI into your organization.

Get paid for sharing Every with your friends. Join our referral program.

For sponsorship opportunities, reach out to [email protected].