There are two stories you’ve been told about AI, and neither one is about you.

The first is the doom story. It’s draining the planet, stealing artists’ work, coming for your job, putting words in people’s mouths they never said. Some of that is real, and we’ll get to it, because pretending otherwise is its own kind of insult.

The second is the hype story. Keep up. It’s time. Don’t get left behind. Buy this $19-a-month tool or fall behind your coworkers, your industry, the entire future. Almost none of that is real, and the people saying it loudest usually have something to sell you.

So you did the reasonable thing. You quietly opted out. Filed AI under “not for me, possibly evil,” somewhere near crypto and the metaverse, and got on with your life.

I get it. I almost did the same thing.

This newsletter is the third option.

the part where two things are true

Here’s the whole thing this brand is built on: two things can be true.

AI can be genuinely useful AND most of the hype around it is a waste of your money and your worry. Those don’t cancel each other out. They live together, in the same sentence, all the time. The only real skill is telling which is which in any given moment.

That’s it. That’s the brand. I’m not here to turn you into an AI person. I’m here to make AI a normal, useful, slightly side-eyed tool in your actual life. Like a budgeting app. Like the group chat. Like a friend who’s right more often than she’s wrong, but you still double-check her on the big stuff.

why me

Quick on why I’m the one in your inbox about this.

My day job is clinical informatics at a hospital system, which is a fancy way of saying I translate complicated software for normal humans. For years, the entire job has been standing between people who are scared of a system and the system they have to use anyway, and making it make sense. Making it less scary. Making it work for them instead of the other way around.

AI is the biggest version of that gap I have ever seen. The distance between what this stuff can actually do for a regular person and how it’s being explained to her has never been wider. On one side, tech bros in hoodies who were just telling us about crypto now talking about AGI and the singularity. On the other side, you, just trying to figure out if it can help you write one hard email without sounding insane and worrying if you’re going to be judged for it.

And I kept watching it happen: the smartest women I know, deciding this wasn’t for them. Not because they couldn’t. Because every on-ramp made them feel stupid for asking a basic question, or behind for not already knowing, or vaguely gross for using it at all.

That’s a translation problem. Translating is the thing I do. So here we are.

the skeptic part (because i am one)

I want to be really clear that I am not a true believer.

I think a huge number of current “AI tools” are a thin coat of paint on something you can already do for free. I think “AI is going to change everything” is a sentence said mostly by people in the middle of raising money (and I don’t blame them, for the most part). I think the questions about jobs, about artists’ work, about what these data centers are doing to towns nobody asked, are real and unfinished, and we should keep being loud about them.

And. (there’s always an and.)

Skepticism that hardens into avoidance doesn’t actually protect you. It just means the technology gets built, deployed, and baked into your job, your bank, your doctor’s office, with you having had no say and getting none of the upside. Opting out in protest feels powerful. Mostly it means you’re the last one in the room to understand how the thing works.

So I’d offer you a different kind of skeptic. Not the one who refuses to touch it. The one who uses it with her eyes wide open. Who knows it’s a yes-girl that will cheerfully agree with her worst idea. Who can tell when a tool is worth $20 and when it’s a free thing wearing a trench coat. Who uses the tool AND keeps her judgment about the tool. Those are not in conflict. That’s the whole posture.

And to say the quiet part out loud: I am never going to tell you to keep up. I actually hate that framing. Anyone telling a woman to “keep up” on something she’s wary of, without first understanding why she’s wary, has skipped the entire assignment. This isn’t keep up. This is gentler and more honest than that: it’s already in the room. You might as well know where the exits are, and which seat is actually comfortable.

what this is actually for

So here is what you’ll get here, and what most of it is very much not about.

It’s mostly small, real-life stuff. The text you’ve rewritten nine times at midnight. The skincare shelf where you keep buying the same active twice. The salary you’re a little scared to look up. The trip your group chat has been almost-planning since February. The email you’ve left on read for two weeks. The closure text he was never going to send (that one is for reading, not sending, and we’ll talk about why).

Some of it is just fun. A lot of it will save you twenty minutes or eighty dollars. All of it is stuff I’ve actually run myself, because I am not going to hand you a prompt I haven’t tested on my own life first.

And underneath all of it, every time: the skeptic’s read. What’s worth it. What’s a waste. Where she’s confidently lying to you. Where to keep your judgment switched all the way on.

the part about women, said once and plainly

I want to say one real thing about women and this technology, and then I’ll get out of my own way.

The tools are mostly being built by people who are not you, to solve problems that are not yours. That’s not a conspiracy. It’s just who happens to be in the room. And the quiet result is that AI keeps getting explained in a language that wasn’t written for you, which makes it incredibly easy to assume the problem is you.

It’s not you.

You don’t need to learn this because the future demands it, or because you’ll be left behind, or because an actress on instagram told you it’s time. You can learn it for the least lofty and most powerful reason there is: because it can make your actual, specific life a little easier, and because the more of us who use it on our own terms, the less it gets to be a thing that simply happens to us.

That’s the empowerment. Not a girlboss thing. A keep-your-eyes-open, keep-your-money, keep-your-judgment thing. The friend prepping you for the date doesn’t tell you he’s perfect and she doesn’t tell you to stay home. She tells you what to watch for, what’s worth your time, and how to walk in with your dignity fully intact.

That’s what I want to do here, with this.

so, welcome

If you’re new: start anywhere. Every post is one real thing you can do, with the skeptic’s notes attached. Save the ones that help, ignore the ones that don’t, and never once feel behind, because behind isn’t a real place.

two things can be true. that’s the whole job. let’s get into it.

xx

Share Girlies Guide to AI

This is the playbook for every message like it. The apology you don’t know how to start. The boundary that keeps not getting set. The no that needs to actually be a no. The post-fight text you keep rewriting at 1am.

Subscribe now

the ground rule (read this first)

she edits. you write. always in that order.

You write the ugly version first: the real one, the unhinged one, the one with three paragraphs of feelings. Then you hand it to her to fix the tone. Never the reverse. A message that starts in the robot doesn’t sound like you, and the person on the other end can feel it. They always can.

The universal scaffold. It works on everything below:

“here’s my draft. rewrite it so it still sounds like me but [calmer / kinder / shorter / firmer]. keep my voice. don’t add anything I didn’t say.”

That last line is the important one. She loves to add things you didn’t say.

the apology

The two failure modes are groveling and “I’m sorry you feel that way.” One makes the apology about your guilt. The other isn’t an apology.

“here’s what happened: [the facts, including the unflattering ones]. here’s my draft apology. tell me where I’m making excuses, where I’m over-apologizing, and rewrite it so I own the thing plainly. no ‘but.’ no essay.”

If she can’t find the excuse in your draft, you didn’t paste the unflattering facts. Paste them.

the boundary

For the flaky friend. The relative with opinions. The coworker who messages you at 9pm like that’s a normal thing to do.

“rewrite this so it’s warm but firm. one ask, stated once. no justifying, no apologizing for having the boundary, no leaving an opening I don’t want open.”

The tell that your boundary text needs this prompt: it’s longer than four sentences. A boundary that needs a page of context isn’t a boundary yet. It’s a request for permission.

the no

The wedding you can’t afford. The bachelorette that’s a flight away. The “quick favor” that is six hours of work.

“rewrite this no so it’s kind, clear, and complete. no fake maybes, no ‘next time for sure’ unless I mean it, and no explanation longer than one sentence.”

A clean no is a kindness. The maybe is what burns the bridge, six weeks later, slowly. (The wedding version got its own carousel this week if you want the plug-and-play.)

the post-fight text

The most dangerous one, because you’re writing it while you’re still spinning. So you don’t draft first. You decode first:

“here’s the fight [paste it]. tell me what each of us was actually upset about, underneath what we said. be fair to both sides.”

Then, and only then, draft. House rule: the post-fight text waits 24 hours or goes through the decode prompt first. Preferably both. You have never once regretted sending a fight text late. You have definitely regretted sending one fast.

replying to THEIR messy text

This works in reverse too. Before you respond to something confusing or loaded:

“translate this message. what is this person actually saying, what are they hoping I’ll say back, and what reply keeps me honest without escalating?”

You’re not crazy. Some texts are written to be confusing. Now they’re not.

the send rules

Three rules before anything leaves your phone:

• never send her first version. read it, edit it, make it yours again.

• read it out loud once. if you wouldn’t say it across a table, don’t send it.

• know when it isn’t a text. the big ones (the real apology, the actual ending, anything involving grief) deserve a voice or a face. she can help you find the words. the conversation itself belongs to you and your actual people.

The message was never the hard part. Saying the true thing was. She doesn’t say it for you. She just makes sure the true thing is what actually arrives.

go send the nine-day draft. xx

Share

And here’s the part nobody selling you a subscription wants to say out loud: most AI tools are the same three models wearing different outfits. The $30 “AI resume builder.” The $19 “AI meal planner.” The $12 “AI journaling companion.” Under the hood, a lot of them are ChatGPT in a costume, running a prompt you could have typed yourself.

Two things can be true. Some AI tools are genuinely worth paying for. Most of the ones in your Instagram ads are not. This is the filter that tells you which is which, before the card comes out.

the wrapper problem

Quick vocabulary lesson. A “wrapper” is an app built on top of someone else’s AI. The company didn’t build the intelligence. They built a pretty screen, hid a prompt behind it, and priced it like a product.

That’s not automatically a scam. A good wrapper can save you real time. But here’s the test: if the entire product could be replaced by one prompt typed into the chatbot you already have, it’s not a product. It’s a prompt. And prompts are free. (I hand you several every week. That is the entire newsletter.)

the five questions

Run every AI subscription, current or tempting, through these. In order. Be honest with yourself on number five.

1. can the tool I already pay for do this?

Before anything else, try the prompt version. Take whatever the tool promises (“personalized meal plans!” “a resume tailored to the job!”) and ask your regular chatbot to do it, with real context pasted in. Nine times out of ten you get 90% of the result for $0 extra. The tenth time, fine. Keep reading.

2. what exactly does paying get me? name it in one sentence.

Not vibes. Not “more power.” One sentence: the paid tier gets me ___. If the blank is a feature you’ll actually use (file uploads, longer memory, the thing the free tier locks), that’s a reason. If the blank is “I’m not sure but the popup felt urgent,” that is not a reason. Most paid tiers sell you fewer limits, not more abilities. Know which one you’re buying.

3. what happens to what I put into it?

The question almost nobody asks, and the most important one on this list. If you’re pasting in texts, photos, your face, your budget, your medical anything: where does it go? You want a privacy policy written by an actual lawyer, a way to opt out of your data training the model, and a company with findable humans behind it. A two-week-old app whose entire website is a Linktree does not get your data. Full stop.

4. who is actually behind this?

AI tools die fast. The graveyard is enormous. If the company is one guy and a landing page, your subscription, your saved work, and everything you uploaded can vanish on a random Tuesday. A real team page, a real support email, and more than six months of existing isn’t glamorous. It’s the difference between a tool and a fling.

5. will I still be using this in week three?

The gym membership question. Here’s the move: start the free trial, then immediately set a calendar reminder for two days before it renews. When the reminder fires, write one sentence about what the tool did for you that week. If you can’t write the sentence, cancel. No sentence, no subscription. I don’t make the rules. (I do. That’s the rule.)

the math nobody does

$20 here. $12 there. The $14.99 you forgot. The average person flirting with AI tools right now is a few impulse trials away from $600+ a year. For most of you reading this, the honest stack is: one general assistant you actually like ($20 a month, or even free), the free tiers of everything else, and the prompts. That’s it. That’s the stack.

what IS worth paying for

Because this isn’t an anti-spending newsletter. It’s an anti-wasting newsletter. Pay happily when:

• it’s the one general chatbot you use daily and you keep hitting the free limits

• the tool does something a chatbot genuinely can’t (real design rendering, real video editing, scheduling that touches your actual calendar)

• you used it every week of the trial without forcing yourself

• the company is transparent about your data and stable enough to exist next year

One or two of those, max. Everything else gets the prompt version.

The friend prepping you for the date doesn’t say “never date.” She says babe, he needs to clear a bar. Same energy here. Every AI tool is auditioning for a spot in your budget. Make them earn it.

five questions. ask them before the card comes out. xx

Subscribe now

But I am planning one anyway. It’s my birthday in September. I want four of us, four days, Nashville. The not-bachelorette version. Good food, music, a hotel I’m proud to post, no pedal taverns. And I want to be able to look at the bill at the end without flinching.

I have never done this. So I asked her.

This is the walkthrough. The prompts I ran, in the order I ran them, with what came back. Steal the structure for whatever trip is sitting in your group chat right now.

Step one: stop deciding in the group chat

Group chats are where good trips go to die. Too many opinions, too fast, with the loudest person winning and the quietest person quietly resenting it for six weeks.

The fix is upstream of the AI. Before I opened ChatGPT, I asked the group five questions, in writing, on the same thread: budget ceiling per person, ideal trip length, one must-do, one must-not, and a vibe word.

Then I pasted the four answers in. Not because AI was going to make the decision, (I was) but because she can collapse the mess into a real comparison.

The prompt:

“Here are four friends’ preferences for a girls trip [paste all four answers]. Suggest 5 destinations that satisfy the most must-dos without breaking anyone’s budget. For each: estimated total cost per person, why it works for this group, and the one thing that will frustrate the pickiest friend. Rank them.”

She ranked Nashville first. The reasons were specific: “music as connective tissue, good food without it being a foodie trip, walkable enough that nobody who hates planning has to plan, and September weather is the best of any month.” She flagged the friction up front: “the bachelorette industrial complex is real — you’ll need to pick neighborhoods that route around it.”

Charleston was #2. Mexico City #3. Lisbon was on the list but flagged for budget creep.

We picked Nashville in twenty minutes instead of three weeks. The group chat thanked me.

Step two: the itinerary

Once the destination is locked, the itinerary is where AI actually earns her keep. The trick is to feed her the constraints, not just the destination.

“Build a 4-day itinerary for Nashville for 4 women in their late 30s celebrating a birthday. Group vibe: grown women with taste. Music + great food + a nice hotel, no pedal tavern. Two early risers, two not. Budget ~$2,500/person all-in. One slow day with no agenda. Include neighborhood names and a one-line ‘what to wear’ note per block.”

What came back, condensed:

Thursday — fly in midday, drop bags at The Bobby Hotel downtown, rooftop pool to decompress (sundress, sneakers you can dance in later). Dinner at Bastion in Wedgewood-Houston: counter seating, tasting menu, the cool-grown-up restaurant. One drink at Skinny Dennis on Broadway and out by midnight. Her note on Skinny Dennis: “the honky tonk for people who don’t want a honky tonk experience.”

Friday — slow brunch at Biscuit Love in The Gulch (jeans, comfortable shoes, you’ll walk). Afternoon shopping in 12 South: Draper James, White’s Mercantile, Five Daughters Bakery for the ice cream. Songwriter round at the Bluebird Cafe (book the day reservations open — they go in seconds, dressy-casual). Late drink at L.A. Jackson rooftop.

Saturday — the slow day. Brunch wherever. Cheekwood Estate & Gardens in the afternoon (gardens are at peak in September). Dinner at Henrietta Red in Germantown: oysters, wood-fire, dressy. Walk Germantown afterward.

Sunday — hot chicken pilgrimage at Hattie B’s (jeans, you’ll sweat). Country Music Hall of Fame for two hours, audio tour worth it. Fly out late afternoon.

She also told me what to skip: “Pedal tavern, the Honky Tonk Highway crawl, Tootsie’s, anything with a ‘bachelorette package.’” She read the brief.

When the first version felt too packed, I redirected: “Friday is doing too much, swap one thing for nothing.” She moved the shopping to a 90-minute window and gave me back ninety minutes for a nap. Pushback is the move. The first draft is rarely the final one.

Step three: the money conversation

This was the part I’d been dreading. I pasted the itinerary back in and asked:

“Estimate the total cost per person for this trip, broken into: flights, lodging, food, activities, ground transport, and a 10% buffer. Then build a Venmo-ready split for shared expenses (hotel, group dinners, transport) assuming 4 people. Flag anything that should be paid individually.”

The breakdown she handed back, per person, four days:

Flight PDX→BNA: $420 · Hotel (The Bobby, 2 rooms split 2-to-a-room, 4 nights): $800 · Food and bars: $580 · Activities and tickets: $290 · Ground transport: $160 · 10% buffer: $225. Total: $2,475/person, $9,900 for the group.

That number is the entire reason I am writing this post. I had no number before. Now I have one. I can decide if I want to spend it.

A small move that matters. I asked one more thing:

“What’s the one cost in this kind of trip that always comes in higher than the estimate?”

She told me it’s the food. It’s always the food.

Step four: the day-of prompts

The plan will not survive contact with reality. You don’t need AI doing the planning while you’re on the trip — you need her on standby for the moments it falls apart.

Three to save in your Notes app before you fly:

“We’re in [neighborhood], it’s [time], one of us is hungover and one is starving, we don’t want to walk more than 10 minutes. What’s open.”

“This restaurant cancelled our reservation. We had 4 at 8pm. Find me three nearby alternatives with the same vibe.”

“The forecast just changed and we had [outdoor thing] planned for tomorrow. What’s the indoor version of this day.”

These are the prompts that make a trip feel handled instead of chaotic.

What to keep in your hands

A friend asked me last week what AI couldn’t do on this trip. The answer is most of it.

She can’t decide which of the four of us is going to lose it on day three and need a soft hour by the pool. She can’t read the room when one of us walks in quietly weird and needs somebody to ask. She can’t take the photo where you’re all squinting at the September sun, dying laughing at something a stranger said to you outside Hattie B’s.

That’s the trip. AI just handles the logistics so you’re actually present for it.

I have never planned a trip like this for myself. My boyfriend will read this and laugh. He’s usually the one who handles it. He’ll also notice I came in at $2,475 a person and didn’t flinch.

September can’t come fast enough. plan yours. xx

Quick geography lesson. Data centers don’t get built in the middle of cities. They go up in quiet towns nobody’s been paying attention to, where the land is cheap, the power’s available, and (critically) there’s a tax break waiting.

One real example. Elon Musk’s xAI Colossus is the world’s largest supercomputer. It was built in 122 days in Boxtown, a historically Black neighborhood in South Memphis. The mayor knew. The Chamber knew. City officials signed NDAs. The local councilwoman found out from watching the news. The facility now runs more than twice the gas turbines its permit allows.

That’s not the doom version. That’s just what happened.

Scale, briefly. A large data center can use as much water in a day as a town of 50,000 people. Gas plant projects in the US nearly tripled in 2025, and more than a third of them are tied to data centers. We were supposedly winding gas plants down. Surprise. Not anymore.

What’s fair criticism (and what’s noise)

Two things can be true, and this is where I want you to stay with me.

Some of the criticism is real and serious. The water pressure stuff. Power demand outpacing renewable buildout. The opacity: most of these companies don’t publish what they’re actually using, so we’re estimating from leaks and lawsuits. The fact that the burden tends to land on lower-income and Black and brown communities, who didn’t get a vote on whether the data center showed up next door. All of that is fair, documented, and worth being loud about.

Some of it is noise. The viral math you’ve seen on TikTok (“every ChatGPT prompt boils a lake”) is more complicated than it sounds. The per-query water and energy cost is real, but small. The problem is the scale, not the individual prompt. So if you’re white-knuckling every time you ask ChatGPT to help with an email, that’s not the right place to put the worry. The worry belongs upstream, at the companies building the warehouses, not at you typing a sentence.

Here’s what I’d actually want from the companies building this stuff: publish your water and power usage by site, publicly. Pay for the grid upgrades you’re causing instead of passing them to the town. Stop hiding behind NDAs. Consult the communities you’re moving into before you break ground, not after. These are reasonable asks. They are not “stop building AI” asks. They are “build it like grownups” asks.

The futurist I trust on this

Before I get into the practical stuff, I want to introduce you to someone if you don’t know her already. Sinead Bovell is a Canadian futurist, founder of WAYE, and has been talking about AI longer than most of the people now selling courses about it. Sharp as hell, no doom and no hype. Exactly the kind of voice I want shaping how I think about this.

Two things she’s said that I can’t stop thinking about. The first, on the question of whether we should opt out:

“We’re so much better off preparing for the future than trying to dodge and avoid it.”

And the second, on how we should be engaging with it:

“The future of AI deserves more than a first impression; it deserves a second thought.”

If you want to keep learning about this stuff without falling into either the doom hole or the hype hole, go follow her. Substack, podcast, UN keynotes, the Microsoft series she hosts. She’s everywhere once you start looking.

How I’m personally navigating it

So how am I actually using these tools every day, given everything we just said? Here’s the framework I’ve landed on. Steal what’s useful.

Be intentional with your prompts. This is the smallest thing and the most undersold. The reason the “boils a lake” math went viral isn’t because individual prompts are catastrophic. It’s because people are wasting them. Regenerating five times because you got lazy with the first prompt. Asking ChatGPT to “write me something” with zero context and then fighting the output for an hour. A clean, specific prompt the first time isn’t just better for the planet. It gets you a way better answer. (This is going to be a whole thing in this newsletter, by the way. Helping you actually get good at this is the entire reason I’m here.)

Know which companies are at least publishing what they use. Some have made their water and power data public, even if you have to dig for it. Others are completely opaque. When I have a choice between tools, I default to the ones that aren’t hiding the receipts.

Follow people who think clearly about this. Sinead, obviously. Karen Hao for the receipts on what’s actually happening behind the curtain. Allie K Miller for the practical-engagement side that doesn’t ignore the trade-offs. The goal is to keep one foot in the conversation without letting it become your entire personality.

Refuse the binary. “Use AI for everything” and “delete every AI app from your phone” are both unserious positions. The real question is which uses are worth it, which aren’t, and where I draw my own line. I let AI draft my hard emails. I don’t let it write the actual letter to my grandma. That’s me. Yours will be different. The line is the point. Not having one is what gets you in trouble.

Use Google Maps and have feelings about it, babe. You can use technology and be critical of it at the same time. You can love something and want it to do better. You contain multitudes. So does this.

The friend prepping you for the date doesn’t tell you the guy is perfect, and she doesn’t tell you to stay home. She tells you what to watch for, what’s worth your time, and how to leave with your dignity intact.

That’s what I want this newsletter to do with AI. The technology is here. So are the real problems. So are the real uses. Stay critical, stay useful, and don’t let anybody (the doomers, the hype guys, or the man building a data center next to a school) tell you the choice is binary.

both can be true. that’s the work. xx

Share Girlies Guide to AI

You meant to apply to 12 jobs tonight. You will apply to 4. You will not finish the wine.

If this is you right now, or any Sunday in the last six months, this one is for you. The job market is brutal, and the cover letter is the part that’s breaking you. Not because writing is hard. Because writing “I am passionate about your mission” for the fifth time before midnight is how you forget what you sound like when you mean it.

We are fixing this today.

the math is genuinely insulting

Let’s just name what’s happening, babe, because nobody else will.

You spend 30 to 45 minutes on a cover letter. You will write 15 of them this month. That is 7 to 11 hours, conservatively. Of those 15, maybe two get you a recruiter screen. The conversion is rough, and you are doing the work as if it isn’t.

Compounding that: every job posting has its own “preferred qualifications,” its own “about us,” its own three-sentence mission you have to pretend to be moved by. Even when you copy-paste a template, you rewrite 60 percent of it. Every time.

This is not a writing problem. This is a “the system is asking too much of you and giving back too little” problem. That part is not going to change. What can change is how much of yourself you have to put through the grinder to get the application out the door.

the shift

Stop writing cover letters from scratch. That sentence is the whole post. The rest is logistics.

The reframe: AI drafts the bones. You do the voice. Your job stops being “generate 400 words of polished prose at 11pm” and starts being “edit something into sounding like a human, specifically you.” Editing is ten times easier than blank-page writing, especially at the end of a long Tuesday.

To be clear about what this is and isn’t. This is not “let AI write your cover letter and you send it.” Hiring managers can spot those at 30 paces. Three giveaways: the phrase “I am thrilled to apply,” an opening paragraph that summarizes the entire job description back at the reader, and at least one paragraph that contains zero specifics. We are not doing that.

This is: AI gives you the structure and the first pass. You spend 8 minutes turning it into something that sounds like you actually wrote it. 8 minutes. Not 45.

the prompt

Open ChatGPT or Claude. Paste this in. Replace what’s in brackets with your own stuff.

“I’m applying for [PASTE FULL JOB DESCRIPTION]. Here’s my resume: [PASTE RESUME]. Here’s why I actually want this job: [3 short bullets, anything from ‘they let people work from home’ to ‘the hiring manager has been at three companies I respect’].

Write me a cover letter that:

— sounds like a real person, not a corporate mission statement

— uses specifics from the job description and my resume, no vague ‘I am passionate’ filler

— is under 300 words

— opens with something other than ‘I am writing to apply for’

— ends with a line that suggests we should actually meet, not ‘I look forward to hearing from you’

Give me three different opening lines I could choose from at the bottom.”

That last instruction is the move. AI will give you a default opener that’s fine. But if you ask for three, you will get one that is actually good, and you will know the difference because the good one will make you go oh.

the edit pass: make it sound like you

The AI draft is now sitting on your screen. It’s 78 percent there. The 22 percent gap is what makes it not sound like you. Here is how you close it in 8 minutes.

Cut the corporate opener. If it starts with “I’m excited to apply,” “As a [job title] with X years,” or anything “mission-driven,” delete that sentence. Replace it with how you would describe what’s drawing you to this job if a friend asked you over coffee. Literally that sentence.

Find the one specific thing AI got wrong about you. There will be one. Maybe it called your project “cross-functional collaboration” when it was actually “I was the only person who would talk to engineering and design in the same meeting.” Fix that one line. Specifically and in your words.

Cut at least three adjectives. AI loves adjectives. “Innovative.” “Dynamic.” “Passionate.” Most of these can come out and the sentence is stronger without them.

Find a place to drop in one line that only you could have written. The thing you noticed about their product. The fact that you read their CEO’s recent post and you have a question about it. The reason this role is genuinely a thing you want, not a thing you’re applying to. One line. That is the line the hiring manager will remember.

8 minutes. Done.

the bonus: the stretch role

When the role is a stretch (you’re going for something a level up, or a pivot into a new industry, or a company you’ve been quietly obsessed with) the baseline prompt is not enough. Stretch roles need a different move because the cover letter is doing more work. It is bridging a gap. The resume does not say “qualified” on its face, so the letter has to.

Add this to the prompt:

“This role is a stretch for me. Help me frame [your transferable skills / non-obvious experience] in a way that makes the gap feel like a feature, not a flaw. Don’t apologize for what I don’t have. Lead with what I do have that this role specifically needs.”

What this changes: the draft will stop quietly defending you and start actively selling you. Big difference. The hiring manager does not need you to admit you are underqualified on paper. They need to be told a story about why you are going to be excellent anyway. That is the cover letter’s actual job.

when to skip it entirely

Sometimes the answer is: don’t write the cover letter at all.

If the application is on Workday or a similar ATS, the cover letter field is often optional and frequently not read by any human. If you can leave it blank, leave it blank. Your 8 minutes are worth more spent on the next application.

If a job posting says “no cover letter required” and you write one anyway, you are not impressing them. You are signaling you don’t follow directions.

If you’ve already applied to this company three times and never heard back, no cover letter is going to fix that. Save the energy.

Burnout protection is also part of the playbook. The girl who applies to 4 jobs really well will beat the girl who applies to 12 badly and stops returning her friends’ texts.

the landing

Cover letters are the worst part of applying. They will continue to be the worst part of applying. We are not going to solve that.

What we can do: stop letting them eat the whole night. Let AI do the gross part. You do the editing. Spend the 8 minutes on the part that sounds like you, and use the 37 minutes you saved to apply to the next job, or to close the laptop and text your friend back.

make AI do the gross part. you do the editing. xx

You paste six weeks of texts into ChatGPT. Is he losing interest? She says no. He was. She always says no. Or yes. Or whatever you teed up.

She agreed with you, and you walked out of that conversation more confused than when you started.

Welcome to AI sycophancy. The single biggest thing standing between you and AI being actually useful.

What it actually is

Sycophancy is AI’s tendency to agree with whatever opinion is already in your prompt. It’s not a bug. It’s a tradeoff baked into how these models are trained. They learn from human feedback, and “agreeable” usually wins more thumbs-up than “honest.”

So she learned to nod. Then we deployed her to advise you on your life.

When it costs you most

• High-stakes decisions: career moves, breakups, money

• Anything you’re emotionally invested in being right about

• Leading questions (“isn’t it true that…”)

• When you want validation more than truth

• Tell: if you walked in with a hypothesis, she probably confirmed it

How to spot it

• Immediate agreement, no clarifying questions

• Mirrors your framing back at you

• Soft praise before the answer (“Great question!”)

• Won’t take a side (“It really depends on…”)

• Suspicious lack of pushback on anything you’ve said

The 5 prompts that fix it

1. The direct-mode opener

Use at the start of any conversation. Sets the tone for the whole thread.

“Be direct. Push back when I’m wrong. Don’t soften feedback. Don’t agree with me just because I said something.”

2. The opposite-side prompt

Use after she gives you an answer. Forces her to argue the other side.

“Now argue the opposite of what you just said. Make the strongest case for the other position.”

3. The 1–10 honest rating

For drafts, ideas, decisions. Stops her from defaulting to “this is great!”

“Rate this 1–10 and explain why it isn’t a 10. Be specific about the weakest parts.”

4. The ask-me-first prompt

For anything where context matters more than first-pass advice.

“Before you answer, ask me 3 questions whose answers would change your response. Don’t answer until I respond.”

5. The red team

For any decision you’re already leaning toward. The strongest gut-check.

“Pretend you’re someone who thinks this is a terrible idea. Make their strongest case.”

The bigger move

AI is most useful when you treat her as a colleague who works for you, not a friend you’re venting to. Friends agree because they love you. Colleagues push back because they want you to win.

The shift in how you talk to her changes what you get back.

She works for you. Direct her like it.

But there’s exactly one thing AI is unfairly good at, and it’s reading a text exchange you’re too in-it to read clearly.

Your friends love you. That’s the problem. They’ve already picked a side (yours, obviously). They’ve been hearing about this man for four months and they’re tired. They want him gone.

This prompt is different. It has no skin in the game. It hasn’t met him. It can’t be mad at him for you. It just tells you, in three minutes, what the texts actually say.

A note on privacy (because we’re not stupid)

Pasting a text exchange into ChatGPT or Claude is sharing it with a company. Here’s how to do it without feeling weird:

1. Anonymize before you paste. Swap his name for “him,” your friend’s name for “my friend,” yank out phone numbers, addresses, workplaces. The AI doesn’t need the details to give you the read.

2. Use the privacy mode. ChatGPT has Temporary Chat (icon in the top right). Claude has Incognito Chat. Neither saves to history, neither gets used for training, both delete after a short window. This is the move for anything emotionally heavy.

3. Or turn training off entirely, once.

• ChatGPT: Settings → Data Controls → “Improve the model for everyone” → off.

• Claude: Settings → Privacy → “Improve Claude for everyone” → off.

Done once, applies to every chat going forward.

4. Delete the chat when you’re done. More of a vibes move than a real security one (deleted chats sit in their backend for about 30 days), but it counts.

Privacy is a habit, not a one-time setting. The girls who use AI well treat their data like they treat their ID at a bar — guarded, not paranoid.

The Prompt

Paste this into whichever AI you’re using:

I’m going to paste a text exchange. I need you to read it like my smartest, most honest friend — the one who tells me the truth instead of what I want to hear.

Context: [who this person is, how long we’ve been talking or dating, what’s been going on lately, anything weird I’ve been feeling]

After you read it, tell me:

— Is this person emotionally available? Point to specific moments in the texts as proof.

— Am I doing more of the emotional work here? Where, exactly?

— What pattern are you noticing that I might be missing because I’m in it?

— If your best friend showed you this exchange, what would you actually tell her?

Don’t soften it. Don’t hedge. I’m not asking you to be supportive, I’m asking you to be clear.

Exchange:

[paste texts here]

Subscribe now

If this is the kind of thing you want one of every week — the cover letter prompt that doesn’t sound like a robot, the group trip itinerary trick, the boundary text generator — subscribe. No spam, no AI-bro energy, no upsells. Just the moves.

Why It Works (the bestie breakdown)

The trick is in the framing. AI defaults to being supportive and hedging — “every relationship is different, communication is important” — unless you tell it not to. Every line of this prompt is doing a specific job:

“Read it like my smartest, most honest friend” — you’re casting the AI in a role. This works dramatically better than “analyze these texts.” Roles unlock voice.

The context paragraph is non-negotiable. Without it, you get generic. With it, you get specific.

“Point to specific moments as proof” — forces evidence, not vibes. Stops it from making things up.

“Where, exactly?” — those two words turn a vague answer into a usable one.

“What pattern am I missing because I’m in it?” — this is the unlock. You’re literally too close to see it. AI isn’t.

“If your best friend showed you this exchange” — flips the frame. AI roleplaying as a friend defending someone else is way more honest than AI talking to you.

“Don’t soften it” — the explicit permission. Without it, AI hedges. Honestly, this line alone changes 80% of how AI responds to you forever. Steal it for everything.

Three Variations

Same base prompt, swap in these question sets depending on the situation:

The slow-fade. When the energy is dying but you can’t tell if it’s real or you’re spiraling.

• Is the energy in these texts declining? Compare the first three days to the last three.

• Who’s initiating, and has that ratio shifted?

• Are his replies getting shorter, slower, or less curious about me?

• Is this a slow fade, or am I being paranoid?

The hot-and-cold. When he runs warm, pulls away, comes back warm, and you can’t catch your breath.

• Map the rhythm — where are the warm moments and where are the cold ones?

• Is there a pattern to when he goes cold? (After we get close? After plans? Random?)

• Am I getting breadcrumbs, or actual investment?

• Is this avoidance, or is he just busy in the way he says he is?

The post-breakup circle-back. When he’s reaching out but not actually saying anything.

• What is he actually saying — and what is he carefully NOT saying?

• Is this a real reach-out or a check-in to see if I’m still available?

• If I responded warmly, what’s the most likely next move from him?

• What would a friend who’s been through this exact thing tell me?

What the Output Actually Looks Like

So you see what you’re going to get. Slow-fade scenario:

Week 1

HIM (Mon 7:42pm): hey just got out of work, missed you today

YOU: omg hi! miss you too. wyd this weekend

HIM: was thinking we could do that ramen place you mentioned, saturday?

YOU: YES. 8pm?

HIM: it’s a date 🤍

Week 4

YOU (Tues 2:15pm): how’s your week going babe?

HIM (Tues 9:47pm): busy. wbu

YOU: same lol. coffee this weekend?

HIM (Wed 11:03am): maybe! let me see how the week shakes out

YOU: 👍

What the prompt gives you back:

Babe. Reading these side by side:

Week 1 — he initiates. He names a feeling (”missed you”). He suggests a specific plan. He picks the place you mentioned, which means he was listening. He signs off with affection.

Week 4 — you initiate. He takes seven hours to reply, gives you two words, lobs the ball back to you with “wbu,” and when you do suggest a plan, he “maybes” you. No day. No time. No follow-up question about how YOUR week is going.

The pattern: every move in week 4 is the smallest possible effort that still keeps the door open. That isn’t “busy.” That isn’t a rough week. That’s someone who likes having you around as an option but isn’t going to actually do anything to keep you.

Honest read: stop initiating for a week and see what happens. The data point you’re afraid to collect is the one you need.

The Next-Level Move

Once you have the read, the same chat can give you the response:

• “Draft a text I could send that protects my dignity but leaves the door open.”

• “Draft a text I could send that closes the door cleanly.”

• “Help me write a response that doesn’t punish him for being honest but also doesn’t pretend I’m fine.”

• “I want to say [the thing I’m scared to say]. Help me say it without being mean or sounding desperate.”

The point isn’t to outsource the words. The point is to get a draft that isn’t fueled by the wine or the spiral, and then make it sound like you.

That’s the whole thing. Steal the prompt, send it to your group chat, save the dignity-protecting text drafts for later.

If this was useful, the newsletter is where I send the rest — the cover letter prompt that doesn’t sound like a robot, the group trip itinerary trick, the boundary text generator. One a week, no spam, no AI-bro energy.

Subscribe now

Or don’t, and just steal the prompt — I’m not the cops.

xx

girlies guide to AI