Why do our AI marketing agents sound like everyone else?
By Greg Rosner
Founder of PitchKitchen · Author of StoryCraft for Disruptors
· 8 min read
TL;DR
AI marketing agents sound like everyone else because they start every job in a Context Vacuum, with nothing specific about your company to read, so they fill the gaps with the average of the internet. Marketing engineering, the role Greg Isenberg describes as wiring AI agents into one working machine, fixes this by starting with the brain: documented context on who buys, how you sound, and what your founder believes. The brain matters more in B2B, because a company selling to a few hundred accounts can't test its way to a message. Build the brain first. Then run agents that read it before they write, and keep every one of them draft-only.
Your AI marketing agents sound like everyone else because they start every job with nothing of yours to read. Every company rents the same models. The one thing a competitor can't rent is a documented brain that tells each agent who buys from you, how you sound, and what your founder believes. Build that first, then wire the agents to read it before they write a word.
That discipline has a name now: marketing engineering. In B2B, the brain matters even more than the agents.
We didn't invent the category. We run our own company on it, and we've started running it for clients.
What is marketing engineering?
On October 5, 2026, Greg Isenberg published a letter arguing that the most valuable new hire in marketing is the marketing engineer. That person is part marketer and part builder, and their job is wiring AI agents into one working machine. He walks through the same idea in his podcast episode on marketing engineers. If you run a B2B company, it's worth your time.
His history lesson is the part I'd tape to the wall. The story of marketing is the story of an experiment getting cheaper. AI took the cost of a new headline, page or campaign close to zero. When a test costs nothing, the person who can wire up fifty of them a week outruns the team that needs three meetings to approve one.
His instruction for where to begin is four words: "start with the brain." He means a folder of context every agent reads before it touches anything. Three files carry most of the weight: who buys, voice examples you loved and rejected with the reason for each, and the founder's opinions.
On top of that brain he runs six agents: customer language, buying trigger, search, creative, closed-lost and performance. The guardrails live inside the tools themselves. Email agents can only draft. Ads launch paused.
Why do AI marketing agents produce generic work?
The model fills every gap in its context with the most common answer on the internet. We call that the Context Vacuum, the place an AI falls when your company hasn't written down who it is.
An agent in a vacuum still ships plenty. Look closely and you'll find the same subject line, the same "unlock efficiency" landing page and the same LinkedIn post your closest competitors' agents wrote this morning. Does that sound familiar?
Picture a stormtrooper helmet sitting on a bag of empty armor. Every piece is there. Nobody's home.
That's a crew of agents without a brain. Give your clones the right spirit, otherwise it's a box of bones.
Most founders respond by swapping models or rewriting prompts. We call that Model Theater, and it doesn't help, because every model falls into the same vacuum. Would you hand a new sales hire a tagline and tell them to go close?
Does marketing engineering work for B2B companies?
Yes, with one big adjustment. Isenberg's playbook leans toward high-volume businesses that can launch dozens of ads a day, let the market vote and keep the winners. Volume does a lot of the thinking.
Now picture a $20M B2B company selling into 300 named accounts, each with a buying committee of a handful of people. You'll never get enough clicks from that room to let a test choose your position for you. By the time an A/B test reaches significance, the deal has already closed.
In B2B, you can't brute-force the message. The brain has to be right before the first agent writes a word.
AI brought the cost of content to zero, so volume stopped being the moat. Perspective took its place. A B2B team gets more from one accurate page about who buys and why than from a hundred agent-generated ad variants built on a guess.
What goes in the brain for a B2B company?
At PitchKitchen, the brain is the Magnetic Messaging Framework, a 35-section document we build with you in the 90-Day Magnetic Messaging Sprint. It maps cleanly onto Isenberg's three files, with more depth where B2B needs it.
| Isenberg's file | What it answers | Where it lives in the MMF |
|---|---|---|
| customer-fit.md | Who buys | Your ideal customer, the moment they start looking, and the accounts you serve best |
| voice.md | How you sound | Voice and tone rules, with approved and rejected examples and the reason for each |
| founder-opinions.md | What the founder believes | Your strategic narrative: the shift in your market, the villain, and the beliefs your company stands on |
The B2B extra is the buyer's own language. A buying committee reads your page through the words they already use in budget meetings. If the brain holds your vocabulary and skips theirs, every agent downstream writes fluent copy nobody searches for.
How do we know the buyer's words matter that much?
We measured our own blind spot. In a 30-day read ending August 5, 2026, PitchKitchen showed up in 0.92 percent of 651 AI conversations about brand identity. We had three live pages on the subject at the time.
The cause was vocabulary. Our pages said narrative identity, the accurate name for the work. Buyers typed branding, rebrand and brand refresh. The engines pulled the pages that used the buyer's words, and ours sat there unread.
A second read stung more. In the week ending August 2, 2026, a buyer question nearly identical to the title of one of our live posts scored zero percent visibility. The page existed. The engines didn't use it.
Which AI marketing agents should a B2B company run first?
Isenberg lists six. For a B2B company between $5M and $75M in revenue, we start with these, because each one reads the brain and each one pays off inside a long sales cycle.
- 1Customer Language. Reads your sales calls and pulls out the exact words buyers use for their problem. Those words flow back into the brain, then into your headlines, pages and sales openers.
- 2Lost-Deal Reopener. Records the real reason you lost each deal, then drafts a personal note the day that reason goes away. A budget freeze lifts. A competitor's contract comes up for renewal.
- 3AI Answer Tracking. Asks ChatGPT, Claude and Perplexity the questions your buyers ask, every week, and shows where you're missing and who shows up instead. That's the AEO work, and it's the same weekly check that surfaced both of our misses above.
From there you add from the menu as the work earns it: a buying-trigger watcher that flags accounts the week a new problem shows up, decision tools like a calculator your buyers bookmark, campaigns aimed at the people who influence the purchase, and a Monday memo where every line ends in keep, kill, change or test.
How do we keep AI agents from embarrassing us?
Put the brakes in the tools. Isenberg makes this point, and it's the one I'd underline twice. Every agent we run is draft-only. Emails land as drafts, ads launch paused, and nothing reaches a buyer until a named person on your team approves it.
A prompt that says "never send" is a polite request to a model. Removing the send permission from the tool settles the question.
Does PitchKitchen actually run on AI agents?
Yes. PitchKitchen runs on nine agents built in Claude Code, and every one of them reads our own Magnetic Messaging Framework. A few of them:
- Tara scans Reddit, X and Hacker News for what our buyers complain about.
- June checks what AI engines say about us every week and drafts posts for this blog. Nothing publishes until I approve it.
- Nora works my warm network.
- Gary runs campaigns to our email list.
- Bob watches the whole machine and flags anything that breaks.
Bob exists because agents break. A scheduled job quietly stops running. Sometimes a data source goes dark and an agent keeps working from stale numbers for weeks before anyone notices. Nothing flashes red when it happens. The work just drifts a little further each week.
That's the honest reason we run the crew for clients instead of handing over a folder of prompts and wishing you luck. A crew needs a supervisor, and most growth-stage companies don't have a spare one.
What can we check this week?
Run these before you buy another AI seat or hire anyone. They're free and they take about an hour.
- 1Open the brain. Try to hand a new agent the documents that say who buys, how you sound (with examples you rejected and why) and what your founder believes. If all you can find is a tagline and a brand-colors PDF, your agents live in a Context Vacuum.
- 2Run the swap test. Ask your agent for the opening line of a cold email to your best-fit buyer. Then ask it to write the same line for your closest competitor. If you could swap the logos, you've found the problem.
- 3Ask the engines. Type the questions your buyers ask before they call you into ChatGPT, Claude and Perplexity, and write down who gets named. For a structured read on the homepage side of that check, run the free Brand Signal Score.
Where does PitchKitchen fit?
We build the brain, then we run the crew that works from it. The brain comes out of the 90-Day Magnetic Messaging Sprint, $15K to $45K depending on company size. The crew runs inside Open Kitchen, our fractional CMO service, at $3,500 to $7,500 a month depending on the menu of deliverables you pick. Every agent drafts, and a named person on your team approves.
The full model, from the brain to the named crew, lives on our marketing engineering page.
The agents are for rent to anyone with a credit card. The brain they read has to come from you.
Questions People Ask
FAQ
What is marketing engineering for B2B companies?
Marketing engineering is the practice of wiring AI agents into one working marketing system, run by a person who is part marketer and part builder. Greg Isenberg made the case for the role in an October 2026 letter. For B2B companies, the work starts with a documented brain covering who buys, how you sound, and what the founder believes, because a company selling to a few hundred accounts can't test its way to the right message the way a high-volume consumer brand can.
Why does AI-generated marketing sound generic?
The model fills every gap in its context with the most common answer on the internet. PitchKitchen calls this the Context Vacuum. Swapping models or rewriting prompts doesn't fix it, since every model falls into the same gap. The fix is giving every agent one documented brain to read before it writes: your ideal customer, your voice with approved and rejected examples, your founder's beliefs, and your buyers' own words.
Which AI marketing agents should a B2B company run first?
Start with agents that work from evidence you already have and pay off inside long sales cycles. PitchKitchen runs three as standard. Customer Language pulls buyers' exact words from sales calls. Lost-Deal Reopener records why you lost each deal and drafts a note when that reason goes away. AI Answer Tracking asks ChatGPT, Claude and Perplexity your buyers' questions every week and shows where you're missing.
How much does it cost to have PitchKitchen run AI marketing agents?
PitchKitchen builds the brain in the 90-Day Magnetic Messaging Sprint, which runs $15,000 to $45,000 depending on company size. The agent crew runs inside Open Kitchen, PitchKitchen's fractional CMO service, at $3,500 to $7,500 a month depending on the menu of deliverables. Every agent is draft-only, and a named person on your team approves anything before it goes out.