Should You Write Your YC Application With ChatGPT? What the 'Demonstrate Human Effort' Backlash Means for F26

The most upvoted thing on Hacker News today is an essay with a blunt title: "If you are asking for human attention, demonstrate human effort." It has more than 1,400 points and hundreds of comments. A few spots below it sits another front-page piece, "Don't You Just Upload It to ChatGPT?" Together they capture a mood that has been building all year, and it has direct consequences for anyone about to submit a YC Fall 2026 application.

If a reader can tell something was generated rather than written, the effort signal collapses. The YC application is the highest-stakes short document most founders will ever write. So the question is no longer hypothetical: should you use ChatGPT to write it?

Will YC partners know if you used AI?

There is no reliable detector that flags "AI-written" text, and YC has not published one. But that is the wrong thing to worry about. Partners are not running your answers through a classifier. They are reading thousands of applications in a compressed window, and they have developed pattern recognition for the specific failure mode that AI tends to produce: fluent, confident, generic.

The tell is not a turn of phrase. It is the absence of specifics. AI writes "we are building an AI-powered platform that streamlines workflows for enterprises." A founder writes "we cut a 40-person claims team's review time from 11 minutes to 90 seconds, and two regional insurers are paying us $2k/month each." The first sentence could describe a thousand companies. The second could only describe yours. When every answer reads like the first sentence, the application gets a quick no, regardless of how it was produced.

What AI is actually good for on a YC application

Using AI on your application is not cheating, and refusing to use it is not virtue. The useful frame is division of labor: let the model do what it is good at, and keep the parts that carry signal for yourself.

AI is genuinely helpful for:

• Tightening, not generating. Paste your own rough answer and ask it to cut word count by 30% without removing specifics. This preserves your facts and your voice while killing throat-clearing.
• Pressure-testing. Ask it to act as a skeptical YC partner and list the three weakest claims in your answer. Then go fix the underlying weakness, not just the sentence.
• Format and clarity. Catching a question you answered vaguely, flagging where you said "we plan to" instead of "we did," checking that your one-line description is actually one line.

What AI cannot do is know what you built last weekend, why you of all people are working on this, or the unguarded story of how you found the problem. Those are exactly the things partners read for.

What you have to write yourself

A few sections carry almost all the weight, and they are the ones a model has no access to.

"What have you made before?" and "What have you built for this?"

This is the founder-market fit and traction core of the application. It rewards concrete nouns and numbers: the thing you shipped, the users you have, the revenue, the week-over-week growth. Look at the companies launching right now. BitBoard (YC P25) showed up on the Hacker News front page this week with a working analytics product for AI agents, not a description of one. Intuned (YC S22) is often cited as a company that got in and then changed its idea entirely, which only works because the founders demonstrated they could build. The common thread is evidence, and evidence is unfakeable by a language model that was not there.

The "why you" story

The insight behind your company usually comes from a specific, slightly weird piece of your own history. ChatGPT will smooth that into a LinkedIn summary. Keep it lumpy. The detail that feels too specific to include is usually the one that makes a partner stop scrolling.

How to keep your application sounding human

Paul Graham's old essay "Write Like You Talk" is the single best style guide for the YC application, and it predates this whole debate. The test: read your answer aloud. If you would never say it to a friend at a bar, rewrite it. AI defaults to the opposite of how you talk, which is precisely why over-edited, AI-polished answers feel off even when nothing is technically wrong.

A practical workflow that respects the "human effort" principle:

1. Write every answer yourself, fast and ugly, in your own words.
2. Use AI to compress and to surface weak spots, never to generate from a blank page.
3. Read the final version out loud. Put back any specific number, name, or story the editing sanded away.

The goal is not to hide that you used a tool. The goal is an application that could only have been written by the person who lived it.

The real risk for F26 applicants

The danger this cycle is not that you will get caught using AI. It is that the median application is getting blander as more founders paste in generated answers, and blandness is what gets rejected. That is actually good news. In a field of interchangeable, confident, specific-free applications, a concrete and a little bit awkward one stands out more than it did two years ago. Demonstrating human effort is now a competitive edge, not just a principle.

If you want a second read before you submit, that is exactly what YC Roaster is for: it connects you with founders who have been through YC and can tell you, honestly, which of your answers sound like you and which sound like everyone else. A human who has read the real thing will catch the generic answer faster than any detector, which is the whole point.

Write it yourself. Use the model to sharpen it. Then make sure the version you submit still sounds like a person who actually built something.