Sponsored by

This is why your AI-assistant makes you sound like a flippin’ wanker.

One of my clients has gone all in on AI. Every email they send, every LinkedIn post, every thought they have is run through an LLM. I love this guy. He's smart; he's ambitious and does really good work. Somewhere along the line, though, maybe it was in third grade, writing book reports, he got the idea that he wasn't a good writer. So as soon as ChatGPT launched into the world, he was so excited because he knew he would never ever have to write anything ever again.


During one of our calls, I asked him about his content generation system that he was telling me about. He opened ChatGPT and put in a really basic prompt for a LinkedIn post waited 11 seconds, copied the output, pasted it into LinkedIn, and looked at me like he'd just split the atom.

The post was fine? But it had the same vibe and pizazz of 400,000 other posts written by 400,000 other people doing the exact same thing. You know what was in it: three em dashes and four emojis. It's not X, it's Y, and the word "compound" or "compounding." If that's how you speak, cool, but the worst thing about it was that there were no edges to it. I could have read that post 200 times, and it never ever would have stuck. We talked about it for 30 minutes, and I still can't remember what the post said.

He told me that he was doing everything this way: landing pages, client reports, prospecting, follow-up emails, the whole shebang. Every single output felt smooth and boring, without any charm, any grace, or any fucking thing that was interesting.

He was really proud that he had a prompt pack for everything that he created. He told me that he felt like he had an AI-powered content platform. No, my guy, you've got a typewriter that runs on autopilot.

About a week ago, I wrote an “Here's an IDGAF take - AI is completely fucking over your agency.” And I was bemoaning the fact that AI takes all the grit and humanness and texture out of everything. It's turned advertisers and agencies into these boring blobs of sameness. (And as a quick aside, some people responded to that post with AI-generated bullshit. Like, come on, dude, read the room.)


You want to know why that is happening?

It's because you are an AI-prompter, but you need to become an operator of AI.

Ask my kids...I have a bifurcated opinion about AI. I love it because it allows me to do stuff that I could never do before. I hate it because if you don't pay attention to it, it makes you say stuff that sounds smart but is functionally meaningless. But I decided to pay attention to it and own it as much as I possibly can. Just as an FYI, I rarely use AI to write this newsletter. I generally use it for ideation and structure...because I want you guys to get my unfettered thoughts. But, I do use AI for writing email sequences for my agency growth tools, some web pages and some social posts.


But I’ve had to wrangle AI into submission because I think my voice, tone, and swear words make a difference in the impact of my communication, and I don't want that watered down by Sam Altman or Dario Amadei or anybody else.

How Jennifer Aniston’s LolaVie brand grew sales 40% with CTV ads

The DTC beauty category is crowded. To break through, Jennifer Aniston’s brand LolaVie, worked with Roku Ads Manager to easily set up, test, and optimize CTV ad creatives. The campaign helped drive a big lift in sales and customer growth, helping LolaVie break through in the crowded beauty category.

Prompting is perilous and pernicious

This is the day in the life of a prompter:

  1. You have a thought.

  2. You type a question into Claude or ChatGPT.

  3. You read the output.

  4. You either copy it directly (gross) or you change 6 words.

  5. You post it.

  6. You move on.

  7. Tomorrow you do it again.

That's not a system. That turns you into a cut and paste monkey for AI.

Prompting has 3 major issues:

Every output starts from zero. Even when you are using all of the memory from your AI bot of choice, every session starts reasonably close to zero. If you are talking about something that doesn't already have context, AI is freaking clueless. Every new topic is Groundhog Day.

AI doesn't really know who the hell you are. AI does a pretty good job of sounding like it knows who you are. It doesn't have an actual connection between what you are asking it to do and what's on your website, the thing that created the impulse to do whatever it is you are asking the AI to do. It gives you something that references words it knows and things it has heard you say before. Because it doesn't have real context about you and what you are trying to do, it gives you very blah, blah average. When you are a marketer, average fucking sucks,\

There's no feedback loop. When something works, the system doesn't learn. When something flops, the system doesn't adjust. You are the only one paying attention, and if you are anything like me, you are so busy getting to the next thing that you don't effectively capture the impact of anything that you create.\

The Outcome: you waste a lot of time cutting and pasting, producing content that sounds smart but delivers very little impact, and you are psyched because you are more efficient.

One brand built 30+ landing pages through Viktor without a single developer.

Each page mapped to a specific ad group. All deployed within hours. Viktor wrote the code and shipped every one from a Slack message.

That same team has Viktor monitoring ad accounts across the portfolio and posting performance briefs before the day starts. One colleague. Always on. Across every account.

5,700+ teams. 3,000+ integrations.

So let's operate an actual system

If you really want to use AI in a way that amplifies you, you've got to invest the time to build the support and scaffolding that allow whatever AI you love to work with you instead of doing a pretty shitty job of completing a task it doesn't really understand.

We tend to think that artificial intelligence somehow means awesome intelligence, accelerated intelligence, or actual fucking intelligence, but AI knows very little. It's really good at figuring shit out once you give it context, a problem, and a desired outcome. It's really fucking great.

AI is an awful lot like hiring somebody in your agency who doesn't know anything about marketing. They don't have any insight or sense of what is good or bad. They don't have any discernment or taste, and they certainly can't improve anything.

If you were hiring somebody, you'd give them a brand voice, examples, and a way to understand whether their output was good or bad. You'd coach them, talk to them, and give them feedback.

If you really want to operate AI in a way that is truly helpful for you, you have to onboard it, give it context samples, and provide feedback. Otherwise, there is no connection between the output created by AI and your intent, or the usefulness and utility of that output. It is output that sounds like it ought to belong to the thing that you are doing, without any real understanding of what you are trying to accomplish.

The value isn't in what you ask AI to create today, it's in all the shit that you set up before you ask for anything. When this all started, there were prompt engineers who convinced you that the magic is in the prompt. If you want to use AI as an effective operating tool for your agency, the magic is in the context.

Once you make this shift, you get leverage like you are Archimedes and you are lifting up the earth with a lever? AI can give you leverage like that.

What This Looks Like In Practice

I'm not a master of AI, and there are bazillions of people who know more about it than I do. Let me share with you exactly how I'm using it.

I have a bunch of skills, which is an AI instruction manual for when I ask for a particular kind of work. Sometimes, in that skill, there are references, which is another document that might have examples, things to avoid, and stuff like that. It is a skill that the AI uses to create a desired outcome. There's one called tim-voice that contains the structural patterns, signature phrases, profanity calibration (yes, calibration), opening hook types, and example openings from my best content. There's one called tim-blog-voice that's tuned differently because blogs are their own thing, and they've got some SEO magic built in. There's one called tim-content-pillars that documents the six topics I actually care about and the angles I take on each. There's one for my brand colors, fonts, and visual identity. There's one for my "Perfect Proposal Framework" that I built so I'd never have to re-explain it. There's even one for the WTF Roadmap deliverable I make for coaching clients, which encodes the exact 8-section structure I use so I can spin up a client deck in 20 minutes instead of 4 hours.

When I sit down to write a newsletter (like this one), I'm not opening a blank box and typing "write me a newsletter about X.” I'm sitting down with a system where I can start talking about what I want to write about and it already knows how I write, & who I write for. it knows what I sound like when I've had my ADHD meds and my daily allotment of caffeine, what my favorite dad jokes are, what words I really fucking hate, and what I've been writing about recently. The starting point isn't zero. The starting point is me, on a good day, with a researcher and an editor already in the room.

That's the fucking difference. I'm not saying, "Give me something." I'm working with this machine that knows what I believe a good output is. I'm giving it input it can contextualize and organize for me so that when it's time to put together a newsletter, it's quick. And if you've noticed, some of these newsletters are approaching 3,000 words. That's like Stephen King level output volume.

Getting the basics down took me a day, maybe. Not a quarter, or hiring a six-figure consultant. A weekend of sitting down and writing out the things I already knew about how I work, in a format the AI could read. That's it. Now, every time I create something, I give it feedback so it gets better over time. There is a little maintenance that happens just about every day, but that is maybe 3 to 5 minutes a day. My voice skills, my attitude, and my way of thinking have never been more easily accessible to me, and in turn I hope to make it easily accessible for you.

Are you tracking agent views on your docs?

AI agents already outnumber human visitors to your docs — now you can track them.

The Three Things You Need To Build (In This Order)

If you want to stop being a prompter and start being an Operator, here's the actual build order. I'm being prescriptive on purpose because I know what happens when I leave this open-ended, you'll nod along, close the email, and do nothing.

1. Write a Voice Document. Believe it or not, I started writing my voice document by asking Claude, "Hey, I want your initial drafts to be closer to what I put out. How do we do that?” It gave me a bunch of steps. It asked for some writing. It asked me questions, and now I have working document with five sections:

  • How I communicate (sentence length, formality, energy)

  • My vocabulary (words I use, words I’d never say)

  • My opinion fingerprint (10 specific takes that define my perspective and would feel really fucking weird for someone else to use

  • My structural patterns (how I open, transition, & close)

  • My CTA style (how I actually ask people to do things in real life, not how some bullshit copywriting guru told me to).

This document is access to everything. If you skip it, nothing else works. Two hours, max. Ask your LLM how to do it. Don't be shy. Be yourself.

2.Collect Your Best Work Posts, emails, podcast transcripts , sales call recordings, voice memos, I even uploaded a stand-up comedy routine I did in 1987. Th this is the storehouse of linguistic patterns that your LLM reads every time it creates something for you. Since AI is a pattern matcher, it matches your patterns!

3. Build a layered prompt structure The three layers are:

  1. System

  2. Task

  3. Output.

The system layer is standing instructions (voice doc, ICP, offers, examples). This never changes.

The task layer is what you want today, like "write a LinkedIn post about pricing."

The output layer is exactly how you want it formatted (length, structure, tone, CTA style).

This system stack builds all three and the output is really close to what you imagined in your head.

Why This Matters Right Now (And Why Most Of You Will Still Do Nothing)

Here's the part where I might get a little ranty.

There is a window right now, and it's not going to stay open forever, where most of your competitors are in prompter mode. they are getting shitty output and publishing it. They are basically filling in templates like some sort of cold email in a box bullshit thing from 2022.

There are some people who are building these systems that are actually super useful. When you've built this machine that has your point of view, has context, has feedback, and has patterns to match against, it now becomes an amplifier, not a template form filler. Anything you need to create has to have your point of view, your insight, your edge. You can get most of the way there today, and you can continually refine it.

You already see the gap between someone who prompts and someone who operates. When you can smell an AI-created post or email ad a mile away because of the mediocrity and level of boring beige, you know that's a prompting methodology at work. you are way fucking better than that.

You are an operator.

You run a business, a department, clients, employees, strategy, and vision. Why would you let AI fuck you over by not telling it what to do? You always want to ride the horse, not be the donkey.

Our sponsorships are ads (obviously - duh). But some of the companies or services linked in our content may pay us a minimal commission if you buy something from them. Potential payment is never a factor in what we link to, and we have direct experience with every company or service where you can buy something that we mention in our content.

Keep Reading