I spent forty minutes last week getting AI to write one paragraph.
Not a keynote. One paragraph. Maybe sixty words, ad copy for a sponsor in Source of Sources. I had the whole thing in my head before I opened the laptop. I could have typed it in four minutes and gotten on my bike.
Instead I prompted. Then I read it, tweaked it, reprompted. Added “make it punchier.” Added “okay, less punchy, that’s too punchy.” Somewhere around minute twenty I stopped writing a paragraph and started playing a slot machine that paid out in adjectives.
Here’s the purpose of today’s email, and I want you to write it so you’ll remember it:
AI is very good at taking your time if you let it, and very bad at building trust on your behalf.
Both halves of that sentence will cost you money. Let’s take them one at a time.
Part one: the time thing
I have ADHD. I talk about it constantly, I’ve built a career on the idea that it’s a competitive advantage, and I stand by that. But an advantage is only an advantage if you understand the mechanism.
My brain runs on novelty and variable reward. That’s the whole engine. It’s why I’ve jumped out of planes five hundred times and why I’ve done Ironman races and why I keynote on customer experience in front of thousands of people.. Those things all work for me because they have something in common: a hard stop. You land under the parachute. The race ends. You step off the stage after the applause and it’s over.
AI has no hard stop. Ever. There is always one more prompt. There is always a slightly better version sitting behind the next regeneration, and the “slightly better version” is the exact shape of the carrot my brain chases hardest. It is the most perfectly engineered infinite loop ever handed to a distractible person, and that’s dangerous.
So it takes your time. Not in a dramatic way. In a beige, forgettable, forty-minutes-on-one-paragraph way, forty times a month, until you look up and wonder why your quarter felt busy and produced nothing.
Part two: the trust thing, which is worse
Every day, about 40,000 people open Source of Sources, and journalists on the other end get pitched. I’ve been running some version of this since HARO, when we connected north of 800,000 sources with reporters before I sold it.
I can tell you exactly what’s happened to the pitches in the last two years. The volume is way up. The quality is flat. And the reason is that a lot of people decided AI could do relationship work on their behalf.
It cannot. Not because the writing is bad. Frequently the writing is fine. It’s because trust isn’t a document. Trust is the accumulated evidence that a specific human being paid attention to you, repeatedly, when they didn’t have to.
The pitches that get used are the ones where somebody clearly read the query, understood what the reporter was actually chasing, and said something useful in three sentences. That was true in 2008 and it’s true now. The best PR has never been about volume or polish. It’s about being genuinely useful to the person telling the story. You can’t outsource being useful. The whole point of being useful is that it costs you something.
Same thing in customer experience, which is the other half of what I do for a living. Companies are racing to put AI in front of the customer, and I get it, the unit economics are gorgeous. But go look at where your customers actually decide how they feel about you. It’s not the tracking-number lookup. It’s the moment something went wrong and they were angry and scared and somebody made it right. That’s the moment that creates the person who tells forty of their friends about you. That’s the entire premise of every keynote I give, and AI hasn’t changed it. AI just made it cheaper to skip.
Skip it enough and you’ve got a company that answers in four seconds and that nobody would defend at a dinner party.
So what do you actually do?
Three rules. I use all three.
1. Give AI a timer and a job description. Before you open the window, write down what you want and how long you’re giving it. Fifteen minutes, first draft of the outline. When the timer goes, you’re done, you take what you’ve got. AI is a great intern. You would not let an intern eat your afternoon by silently handing you thirty variations of the same thing.
2. Never let AI own the first moment or the recovery moment. First contact and the moment you screwed up. Those two belong to a human being with a name. Everything in the middle is fair game. Scheduling, formatting, summarizing, drafting, research, cleanup. Give it away happily.
3. Ask one question before every task: if this works perfectly, does anyone trust me more? If yes, use the tool and use it hard. If no, the tool is not the problem, the task might be.
That paragraph I spent forty minutes on? I closed the laptop, wrote it myself in four minutes, and did a Peloton ride.
The AI wasn’t wrong. It was just never going to be the thing the sponsor was paying for.
If you run a small business and you keep hearing that AI is going to change everything but nobody has told you where to actually start, I’m running a webinar on September 16 that does exactly that. No jargon, no hype, just the tools that will save you hours this week. It’s $79 through August 1, then it goes to $129. Details at shankman.com/smallbizai.


Wow.
This is so good, Peter. Thank you. I love the timer idea.