Thirty Days of Letting a Robot Write My Portfolio: The Interview Data Is In

I've spent the better part of two years treating my portfolio copy like a bonsai tree. Pruning adjectives. Rearranging paragraphs at 1 a.m. Rewriting my bio so many times that I could probably recite it from muscle memory.

So when AI writing tools got good enough to actually threaten my workflow, I did what any reasonable developer with too much curiosity and not enough interviews would do: I ran an experiment. For thirty days, every word on josephkhan.me was machine-generated. I tracked recruiter emails, LinkedIn InMails, and interview invites against my previous ninety days of hand-written copy.

Here's what happened.

The Setup (Because Yes, I Made a Spreadsheet)

Baseline first. In the three months before the experiment, my hand-written portfolio pulled in 11 recruiter responses and 4 first-round interviews out of roughly 60 applications and inbound contacts. Decent, not spectacular. I'm a mid-level frontend dev in Austin, so I'm competing in a market that is not exactly hurting for candidates.

For the test month, I used AI to rewrite four things: my hero bio, three project case studies, my "about" page, and the one-paragraph blurbs on my project cards. I gave the model my actual metrics, my real code snippets, and notes on what each project solved. Then I edited lightly — maybe five minutes per page — and shipped it.

What Got Better Fast

The case studies were the surprise. My hand-written versions were, in hindsight, love letters to my own process. Tons of "I decided to" and "my approach was." The AI versions led with the problem and the outcome because I prompted them to, and that reframing alone made them more readable.

Recruiter response rate on applications went up about 30% during the test month. Not life-changing, but real. Two of the recruiters who replied specifically mentioned my case studies being "clear" — which, given that a machine wrote them, is both flattering and slightly deflating.

The project card blurbs were another win. I'm bad at summarizing. Give me 100 words and I'll use 140. The AI gave me tight two-sentence descriptions that apparently scan well, because click-throughs from my homepage to project pages nearly doubled.

Where It Fell on Its Face

My about page was a disaster. The first AI draft read like a LinkedIn influencer's morning post: "I'm passionate about crafting delightful digital experiences." I have never said the word "delightful" out loud in my life. I asked three friends to read a version and all three guessed it was AI-written within seconds. One called it "resume GPT voice."

The other problem: specificity decay. The AI loved vague, confident claims — "significantly improved performance" — where my real numbers were "cut LCP from 4.1s to 1.8s." Every time I let a vague claim through, I felt the copy get weaker. Hiring managers in the US market, especially at product companies, smell hedged language instantly. They've read a thousand resumes that say "improved efficiency."

There was also one moment that genuinely embarrassed me. An interviewer asked about a line in my case study that claimed I "pioneered a design system migration." Pioneered is doing a lot of work there. I followed a Figma file and complained in Slack. Hand-written me would never have written that. But AI-written me shipped it, because it sounded good.

The Hybrid That Actually Works

After the month ended, I didn't go back to full hand-written. I landed on a split that I think is worth stealing:

Let AI write: project card blurbs, the structure of your case studies, first-draft bullet points from your raw notes, and email outreach copy.

Never let AI write: your bio, your "why I build things" section, and any sentence containing a number you didn't verify. Your bio is the one place where generic voice is fatal, because it's the one page where the reader is explicitly trying to figure out what kind of person you are.

The tell, I've decided, is adjective density. AI copy stacks pleasant, empty modifiers — robust, seamless, delightful, passionate. Human copy that's worth reading is mostly verbs and nouns. "I rewrote the checkout flow in three weeks and cart abandonment dropped 12%" beats anything with the word "seamless" in it.

So yes, the robot got me more interviews. But the ones it got me, it got by making my real work easier to read — not by inventing anything. That distinction is the whole game.