Three out of every four small business owners are using AI right now. Most of them say it's helping. And almost none of them have it actually built into how they run their business.
That's not a guess — it's what small business owners themselves said in a recent Goldman Sachs survey of their 10,000 Small Businesses network. Seventy-six percent are using AI. Ninety-three percent of those say it's had a positive impact. But only 14% say it's fully part of their operations. The rest are stuck somewhere in between — trying it, liking it, not quite trusting it enough to build on it.
If that's you, you're not behind. You're exactly where most people are. The question worth asking isn't “should I be using AI more” — it's “what's actually separating the 14% from everyone else?”
I know that in-between phase firsthand. About two and a half years ago, I was writing a book for my children and using AI as my content editor. Looking back, if I knew then what I know now, the end product would have been a lot better. I was in the search-bar phase without realizing it — asking, accepting, moving on, instead of actually briefing it on what I was trying to do.
I don't think that means I should have waited until I knew more. Like most new things, you jump in and get better as you go. The point isn't to skip the in-between phase. It's to know it's temporary.
Two Ways of Asking
I've watched myself do this, and I'd bet you've done it too. There are really two instincts people bring to AI, and most of us start with the first one before we grow into the second.
Instinct one: treat it like a search bar. Type a quick question, get a quick answer, move on. It's not wrong — it's just how we've used every screen in front of us for twenty years. Google trained us well.
That's exactly how I used AI when I started editing my book. I'd hand over a chapter, take whatever came back, move to the next one. It felt like progress. It wasn't really collaboration — it was just faster typing.
Instinct two: start with the problem, not the question. Before you ask anything, you lay out what you're actually trying to solve — the background, the constraints, what “done” looks like. Then you ask.
The longer I worked with it, the more that shifted. I stopped taking every suggested edit at face value. I started pushing back — asking why it wanted a change, telling it what I was actually trying to say, going back and forth until the feedback matched what I was trying to deliver, not just what read cleanest on a page.
I don't think of the first instinct as a mistake. I think of it as where everyone starts. It's the same arc as anything else you got good at in your career — nobody opens with the advanced version. You earn your way into it. The shift from search-bar thinking to problem-first thinking is just AI fluency growing up.
What This Looks Like in Practice
Here's what that shift actually looked like with the book. Early on, I'd get a suggested edit and just take it — a sentence restructured, a paragraph tightened, whatever it flagged. I figured it knew grammar and structure better than I did, so why argue.
But the book wasn't really about grammar. It was about what I wanted my kids to remember about their dad. A few chapters in, I noticed some of the “cleaner” edits had quietly flattened the parts that actually mattered — the awkward, personal, very-me phrasing that was the whole point of writing it myself instead of hiring a ghostwriter.
So I started pushing back. Instead of accepting a suggested edit, I'd ask why it made the change, and I'd tell it what I was actually trying to communicate — not “make this better,” but “this needs to sound like I'm talking directly to my daughter, not writing a memoir excerpt.” Some edits I kept. A lot I didn't. But every round of that back-and-forth made the next round sharper, because it wasn't guessing anymore. It actually knew what I was trying to deliver.
That's context in practice. Not one perfect prompt up front. A conversation — where you keep pushing back and telling it more until the output actually matches what you meant.
Good AI feedback isn't something you accept. It's something you build — by pushing back until it actually understands what you're trying to say.
The Toolbelt
There's a second layer to context that's easy to miss, and it's less about what you say and more about what you let AI use.
Think of it like a toolbelt. Left on its own, AI will reach for what's already in its head — trained knowledge, which can be outdated or just generic. That's fine for some jobs. For others, you want it reaching for something different: going out and pulling real, current information instead of guessing from memory.
That's not your job to do for it every time — it's a decision you make on purpose, the same way you'd decide which tool a job actually calls for instead of grabbing whatever's closest. I learned that lesson on a flight line long before I learned it at a keyboard: you don't reach for one universal wrench. You reach for the right tool, deliberately, because the job in front of you demands it.
When I put together research for this newsletter, that's exactly what I do. I don't just ask for “articles about X” and hope for the best. I tell it precisely what angle and audience I'm after, and I have it go out and pull real, current sources — rather than relying on whatever it already “knows,” which might be a year stale. Specific ask, plus the right tool for the job. That's context working on two levels at once, not just one.
Try It Yourself
Next time you open a chat with AI, try building your ask like this:
“We have a [process/task] that we're trying to [automate / improve / understand]. Here's the background: [context]. Here's what's been going wrong or what I'm trying to solve: [problem]. Here's what I want back: [format/output].”
Fill in your own blanks. Doesn't have to be work at all. Here's one I've actually used, dinner, not dashboards:
“My wife and I have been cooking together and neither one of us likes cleaning dishes. We have focused on sheet pan recipes. Please provide us 5 sheet pan recipes that contain a protein and vegetables and are suitable for two adults. These should be able to be prepared start to finish in 30 minutes.”
Notice what's actually in there. Who's cooking. Why sheet pans specifically. What counts as a good answer — protein, vegetables, two servings, 30 minutes, start to finish. None of that is complicated. It's just complete — same shape as the template, just filled in with something a lot lighter than editing a book.
Pick one thing and try the complete version instead of the quick-question version.
Hit reply or drop a comment with what you tried — I'll feature a few next issue.
Context isn't a one-issue topic — we'll keep coming back to it as the newsletter goes on, each time a little more specific: how to give AI context from a document, from an ongoing project, from your own numbers. This was the wide net. Future issues pull it in tighter.
— Eric
More of this on LinkedIn: linkedin.com/in/eric-skelton-mba-3079a47