AI. pencil. ChatGPT.

Why ChatGPT Keeps Using the Word “Surface” (And Why It’s So Annoying)

The pebble in the shoeThis is one of those things that feels too small to matter, until it does.Everything else can be working just fine. The system. The process. The tools. You’re in a good rhythm, you’re excited to get started, and then there’s this one tiny thing that throws you off completely. That was me, a few weeks ago, heading out for a long walk with a shoe that otherwise felt perfect… except for a single pebble. Small. Inconsequential. Impossible to ignore.That’s what the word “surface” has become for me in ChatGPT drafts lately.It’s not that I don’t understand what it means. I do. And it’s not that the drafts are unusable, they are. But the word keeps showing up in places where it feels oddly formal, slightly abstract, and just off enough to break my flow. Once I notice it, I can’t not notice it. It pulls me out of the text every time. Break my flow moment.This reflection sits alongside Episode 8, where I spiral (lovingly) about this one word, not because it’s evil or wrong, but because it reveals something about how these tools work, and how sensitive we are to tone when language actually matters.Read more and/or watch this video on our Substack: https://aiisapencil.substack.com/ This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit aipencilchatgpt.substack.com

When and Why I Ask ChatGPT for Help

This isn’t usually my tone when I talk about tools like ChatGPT.I’m generally pretty clear about where I stand, I see large language models as thinking partners, not shortcuts, and I’m far less interested in speed than I am in clarity. But I recently read something that knocked me just slightly off balance, in a good way, and I wanted to sit with it instead of brushing past it.Read more and/or watch this video on Substack: https://aiisapencil.substack.com/ This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit aipencilchatgpt.substack.com

ChatGPT as a tool for difficult work conversations

Something had happened on a project I’m working on. Someone made a decision that ignored my process and, honestly, it felt like it undermined a lot of the work I’ve been doing. It wasn’t catastrophic. There wasn’t a lot of money attached to it. But it hit that familiar nerve of wait, why wasn’t I included in this?I could feel myself drafting the message in my head as I walked. You probably know the kind.Quick.Pointed.Emotionally accurate, but not especially useful.And I knew that if I sent that message, I’d spend the rest of the day replaying it. Wondering how it landed. Wondering how they’d respond. Wondering if I’d made things worse.I didn’t want the message to be about me.I wanted it to be about the project.So instead of sending anything, I opened ChatGPT and started talking to it.Read more and/or watch this video on our Substack: https://aiisapencil.substack.com/ This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit aipencilchatgpt.substack.com

ChatGPT think with me, not for me

I had a realization recently that was… equal parts obvious and annoying.There’s a lot going on in my head when I write to someone. Context. Backstory. Examples. Reasons. Assumptions. Side notes. All neatly connected in a way that makes perfect sense to me.And then I send the message.And then comes the back-and-forth.Clarifying questions. Misunderstandings. “Oh, I didn’t realize you meant that.” Moments where I realize, oh right, nobody else is inside my brain.Wild.I’m very aware that I connect dots quickly, sometimes too quickly. I’ll assume people are following along because I’m following along. But why would they be? They don’t have the same internal Google Doc running in the background.So lately, I’ve been slowing myself down.Not by writing more. Not by over-explaining. But by getting the missing context out of my head before I hit send.That’s where ChatGPT has quietly become one of the most useful tools in my actual, everyday communication life.Not for facts. Not for research. Not for “write this for me.”But for pulling the unstated stuff out of me.ChatGPT as a context extractor (not a writer)Here’s what I’ve been doing.Instead of saying, “Write this email” or “Make this post sound better,” I’ll say something more like:Here’s the situation. Here’s what I’m trying to say. Ask me questions to make sure this makes sense the first time.And then I let it do exactly that.Read more and/or watch this video on our Substack:https://aiisapencil.substack.com/ This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit aipencilchatgpt.substack.com

I’ve Been Too Polite With ChatGPT

Three years.Three stinking years I’ve been using ChatGPT.Since the beginning. Since the “wait, this thing can finish my sentences?” era. Since the “oh no, this might actually be useful” phase. Since the “okay, calm down, it’s still just a tool” self-talk.And it took me until this week to realize something important:I’ve been way too polite with it.Not polite in a “please and thank you” way (although… yes, also that). Polite in a professional, buttoned-up, LinkedIn-approved way. Polite in a “hello, fellow knowledge worker” tone.Which is wild, because this week was the first time I gave ChatGPT a sarcastic persona.And I’m never going back.The problem isn’t writing. It’s revision.Here’s the thing: I love writing.I do not love revising.I don’t love editing. I don’t love reorganizing. I especially don’t love blog writing, which somehow manages to feel both redundant and high-stakes at the same time. I’d much rather brain-dump the ideas, list everything I know, and move on with my life.Unfortunately, that’s not how writing that “lives in the world” works.If something’s going to exist for more than five minutes: a blog post, a newsletter, a resource people might bookmark…\it needs outside eyes. It needs critique. t needs someone to say, “Yes, but… this part doesn’t land.”And critique always stings. Even when you ask for it. Even when you know you need it.So I do what a lot of us do: I use ChatGPT for feedback.Usually with a very sensible, very adult, very reasonable prompt chaining session. * Check for clarity. * Check for structure. * Check for logic gaps. * Be helpful. * Be kind. * Be constructive.Then someone casually mentioned, “Oh yeah, but that’s nothing compared to my sarcastic one.”Record scratch.Enter: Sarcastic ChatGPTI can’t share their exact prompt because it’s not mine. But the idea was simple:Same analytical rigor.Same checklist.Same goals.Completely different tone.The instructions included things like:* Be brutally honest* Roast the writing hilariously* Swear if necessary* Don’t let weak logic slideAnd wow.The feedback started with something along the lines of:“This piece starts strong and immediately gets lost in a cul-de-sac of ‘show notes are important,’ like it’s trying to convince itself.”Did it sting?Yes.Was it accurate?Also yes.And here’s the important part: instead of spiraling or shutting down, I laughed. Out loud. In a coworking space. While trying not to disturb anyone.Another gem:“Congrats. You found a way to make a good point while looking like you were typing downhill.”I mean. Come on.Painful? A little.Useful? Extremely.Memorable? Absolutely.The unexpected part: I started talking backAt some point, I typed:“Okay, smartass.”And that’s when something shifted.Once I started responding in kind with sarcastic, casual, mildly unhinged language, all the stress disappeared. I stopped bracing myself for feedback and started playing with it.I actually lost track of time during the edit. In a good way.Not because the task got easier, but because it got lighter.And that’s when it hit me: I’ve been treating ChatGPT like a very polite junior colleague, when what I really needed was a sparring partner.Not someone whose tone ends up in the final text. Just someone who can verbally jab me during the process.Being “professional” is overrated (in your prompts)I realized I’ve been too professional with my LLM.The output will be what the output is. That doesn’t change much.But my interaction with it? That can be messy. That can be sarcastic. That can include swearing, jokes, exaggeration, and personality.For some reason, I’d mentally separated “good output” from “fun interaction,” as if seriousness was required for quality.It’s not.In fact, telling ChatGPT how to be, not just what to do, unlocked something I didn’t know I was missing.When I typed curses (which I almost never do on a physical keyboard), it felt weirdly freeing. Like letting off steam without involving another human being.Maybe I’m just repressed.Maybe this is just genuinely fun.Possibly both.This doesn’t make the AI sloppy, it can make you relaxedThis is the part I want to be clear about, because it’s easy to misunderstand.Giving ChatGPT a sarcastic persona doesn’t mean you abandon structure, clarity, or critical thinking. You still need:* Clear instructions* Specific criteria* Guardrails against nonsense and hallucinationsTone doesn’t replace rigor. It coexists with it.What changes is your experience of the work.If revision feels less awful, you’re more likely to do it well.If critique feels playful instead of punishing, you’ll actually absorb it.If the process is lighter, you’ll stay with it longer.That matters.A rare challenge from meI don’t usually do this. I’m not big on weekly challenges. I don’t want to turn this into homework.But just this once:This week, give your ChatGPT a persona that’s ridiculously fun.Sarcastic.Blunt.Over-the-top.A little unhinged (within reason).Tell it exactly what to look for. Tell it exactly how to critique. Then tell it how to sound while doing it.See what happens.Even if the output isn’t wildly different, I suspect your mood will be. And how time passes while you’re working. And how much resistance you feel opening the document again.I still love writing.I probably still hate revising.But at least now, revision comes with a smartass in my corner.How on earth did it take me three years to figure this out?Go. Just go.See ya next week, StephAI disclaimer:I use ChatGPT to co-write many of my online texts. Having said that, for these Substack posts we start with the video transcript. A transcript that comes from a recording that I create alone, with zero AI assistance. I also prompt chain and edit like hell during and after the cowriting process, so I have to admit that I have no idea where my writing begins and Chatty (my affectionate name for ChatGPT) ends. So take with this as you wish. I just wanted you to know that some of the eloquence here is in fact from me but not me exactly. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit aipencilchatgpt.substack.com