Hands On AI Training

09/04/2026

Start paying attention to what you fix by hand. Every one of those little corrections is an instruction you've never actually saved.

You know how this goes. A draft comes back from the AI. You fix the greeting. You cut the last paragraph because it's laying it on thick. You change "utilize" to "use," because "who talks like that"? You add the missing phone number, which it leaves off every time. Four edits, maybe two minutes, and you don't give it a second thought.

Then next month the same job comes around and you make those same four edits again. And the month after that.

Here's the thing. Each one of those is a standard you already have in your head and have never once said out loud or stored it so it can be reused. Say it a single time and you quit paying for it. Never open with a rhetorical question. Don't close with a sales pitch. Always include the phone number. Save it so you can copy the prompt the next time you generate the chat.

That's the whole habit. Next time you clean up an AI draft, stop when you're done and ask yourself what you just changed. Write that sentence down somewhere. Stick it in the request next time.

Do that three or four times and the drafts start landing close to finished, because you taught it your standards instead of putting them back in by hand every time.

This works the same in Claude, ChatGPT, and Gemini, no matter the tool.

handsonaitraining.com

09/02/2026

If you work for a Minnesota city, here's something worth knowing. What you type into an AI tool might be government data. So might what it sends back to you.

The League of Minnesota Cities laid this out in the July/August issue of their magazine. Their staff attorney, McKaia Dykema, walks through how the Government Data Practices Act applies to AI. Government data means anything collected, created, received, or maintained by a government entity, in any format at all. That can reach what you type into an AI system and what it produces for you.

Two things follow from that. Putting nonpublic data into an outside AI tool could count as a breach, and she names Social Security numbers, personnel records, and privileged communications specifically. And AI records, including your input history, might have to be handed over in response to a data request.

I want to be clear that I'm not an attorney and this isn't legal advice. The League published the article and sample policy language at lmc.org, and your city attorney is who you want for the classification questions.

But there is a habit that comes out of this, and habits are my end of it. Before you paste something in, know what it is. Public or not public. That's a thirty second pause before you hit the keys, and it works the same whether your city is in Copilot, Claude, ChatGPT, or Gemini.

handsonaitraining.com

https://www.lmc.org/news-publications/magazine/july-august-2026/letter-of-the-law-july-2026/

As artificial intelligence (AI) tools become more common, Minnesota cities are exploring how these technologies can improve efficiency, customer service, and daily operations.

08/31/2026

Small businesses have already made up their minds that AI works. What they say they're missing is training.

The Global Technology Industry Association put out a study on August 27, based on 520 small business and microbusiness decision makers. Eighty four percent of them said the AI tools they use are producing real business results. So the argument about whether this stuff is worth the money is pretty well over for that group.

Then the study asked what kind of help they actually want. The top two answers were both at 38 percent; employee training and help figuring out which tools to use. Both of those came in ahead of system integration, cybersecurity, and writing policy.

I'll say up front that GTIA is a trade association for IT service companies, so the study is built to tell its members what to go sell. Take the numbers as a picture of what small businesses say they need, not a prediction about anything.

The picture is one you've probably seen. An owner buys the subscription. Somebody on staff watches a video about it. Nobody ever sits down with the real work, on the real computer, and stays there until it runs right.

One more number. Forty four percent have a written acceptable use policy, which means more than half of these businesses are using AI with nothing on paper about how. This remains a critical missing piece of policy.

Training and picking the right tool is what they said they wanted, and neither one is a big project.

handsonaitraining.com

08/28/2026

There's a conversation you've been putting off. Here's something that helps more than I expected it to.

Rehearse it first. AI will play the other person, and it does it well enough to be worth your time.

Tell it who you're dealing with and what the conversation is about. The resident who's angry about a water bill. The board member who thinks you got it wrong. The customer who wants money back that you can't give back. Then tell it to be that person, to push back the way they really would, and not to go easy on you the second you say something sensible.

Then just have the conversation. Type your side.

Here's what happens. You'll hit the one sentence you can't get out cleanly. Everybody's got one. It's almost always the part where you have to say the thing they don't want to hear, and it comes out either too soft to land or a lot harder than you meant it. Way better to run into that at your kitchen table than standing across a counter from somebody.

When you're finished, ask it where you were weakest, and what it would have said if it were trying to knock you off balance.

You're not writing a script. Scripts fall apart the second the real person says something you didn't plan on. You're finding the words for the hard part so you're not making them up under pressure.

Same in Claude, ChatGPT, and Gemini. This works great to rehearse for an interview too! Load up the job description and your resume and start an interview with the AI.

handsonaitraining.com

08/26/2026

Eighty six percent of people who use AI at work say they treat what it gives them as a starting point, not a finished answer, and that they stay responsible for the thinking. That's been my line since I started this, so I'll admit it was nice to see the number.

Microsoft's 2026 Work Trend Index surveyed 20,000 people across ten countries who use AI at work. When they looked at what actually separates the people getting real value from the people who aren't, it wasn't mostly about the person. The conditions around them mattered more than twice as much as their own mindset and effort. Does the culture let you try something and have it not work. Does your manager use it out in the open. Has anybody ever said what good work looks like around here when AI helped make it.

Ten percent landed in a category the report calls blocked agency. Capable people stuck in organizations that haven't caught up to them.

One should take fair warning because of the source. Microsoft says right in the report that these are associations and not causes, and they sell AI software, so read it with that in mind. It still matches what I'd expect. One person teaching themselves in a small office hits a ceiling that has nothing to do with how good they are.

If that's somebody on your team, the question isn't whether to send them to training. It's what happens when they come back.

handsonaitraining.com

08/24/2026

Next time AI hands you a first draft that's just okay, don't start rewriting it. Ask it one question first. What would you need to know from me to do this better?

Then answer whatever it asks you.

Almost everybody reads a bland first draft as the tool doing the best it can. Nine times out of ten it's the tool doing the best it can with what you gave it. You asked for a newsletter piece, and it has no idea who reads your newsletter, or when it goes out, what actually got said at the meeting, or that the board already shot down the first version of this idea back in April. So you get something that could have come from any organization anywhere.

Ask what it needs and it'll just tell you. Usually four or five questions. You'll know the answer to every single one, because they're questions about your work, not questions about AI.

Answer them however they come out. Half sentences are fine. Nobody's grading your typing. Then tell it to go ahead and try again.

That second draft is the one actually worth your editing time.

This works equally well in Claude, ChatGPT, and Gemini, because it isn't a feature. It's a question. There's nothing to turn on, nothing to buy, just two minutes on the front of a job you had to do anyway.

handsonaitraining.com

08/21/2026

Before you let an AI assistant loose in your web browser, here's something worth knowing. A webpage can give it instructions that you never see.

It's called prompt injection. Somebody hides text in a page, or a document, or an email, and that text tells the AI to do something other than what you asked for. You get back a nice summary. Meanwhile it quietly did something else along the way, something you don't want it to do!

Anthropic published real numbers on this back in May, and I'll give them credit, because most AI companies didn't. Testing across 129 browser environments, a determined attacker got through 31.5 percent of the time with the safety systems switched off, and half a percent with them on. Their own write-up plainly says that no browser agent is immune. OpenAI has said the same thing, and that it may never be fully solved for browser agents. Google hasn't published a number yet.

None of that means stay away from this stuff. It means know which mode you're in.

Sitting and typing with Claude, ChatGPT, or Gemini in a chat window has none of this risk. Turning one loose to browse the web, read your email, and act on your behalf is a different animal. Do that on purpose, on jobs where you can look at what it did afterward.

handsonaitraining.com

08/19/2026

Here's a quick one for anybody who runs a business. Open ChatGPT, Claude, and Gemini and ask each one to recommend a business like yours in your town. Then read what you get back. There's a decent chance you're not in it.

The whole thing takes a minute, and almost nobody has done it. It's worth doing because how people go looking for a plumber in Winona or a bookkeeper in La Crosse is shifting, and the answer they see gets built out of whatever those systems can find about you and line up.

Ask three ways. Name what you do and the town you do it in. Ask who it would recommend and why it picked them. Then ask it flat out what it knows about your business by name.

That third one is the one that pays. You find out in a hurry whether your hours, your address, your phone number, and a plain description of the work all match up across your own website, your Google listing, and wherever else you show up. When those don't agree, you don't get recommended. You get a hedge. You can watch it happen right there in the answer.

I'm not an SEO shop and this isn't a ranking strategy. It's a check you can run yourself, today, for nothing.

Run it, write down what's wrong or missing, fix what's yours to fix, and run it again in a month to make sure the fix took hold.

handsonaitraining.com

08/17/2026

Ninety two percent of nonprofits use AI. Only seven percent say it's actually made a major difference in what they can do. That gap isn't about the tools they're using.

Those numbers come from the 2026 Nonprofit AI Adoption Report, a survey of 346 organizations put out by Virtuous and Fundraising.AI. I'll be upfront that Virtuous sells fundraising software to nonprofits, so I'd treat this as a picture of how people are working rather than a scorecard. The behavior part is what's interesting anyway.

Eighty one percent said their AI use is individual and ad hoc. Four percent said they have workflows written down that somebody else could follow. So the value is real, and it's showing up one person at a time, quietly, and none of it survives that person leaving to take another job.

You've probably watched this happen. The development director has figured out how to draft an appeal that actually sounds like your organization. Nobody else has the first idea what she types to get there. Somebody else summarizes reports in Gemini, somebody else drafts in Claude, and those two have never once compared notes. The fix isn't another tool. It's writing down what already works so it belongs to the organization instead of living in one person's head.

Pick one task your team does every single month. Get the instructions on paper. That's your first repeatable workflow, and most nonprofits don't have one yet.

handsonaitraining.com

08/14/2026

Take the thing your organization publishes that nobody can understand and have AI write the version people can actually read. It's about a half hour of work. Very few organizations actually get around to do it.

Paste in the ordinance, the personnel policy, the program eligibility rules, the form instructions, whatever it is that people call and ask you about, into a fresh AI chat. Then ask for a plain language version at an eighth grade reading level that keeps every requirement exactly as it is and doesn't change a single deadline, dollar amount, or name. Ask it to tell you what it wasn't sure how to simplify.

Then sit down and read the two side by side, because that part is on you. You're looking for anything that got softened. An exception that disappeared. A requirement that quietly turned into a suggestion. The AI rewrites fast and has no idea which sentence or idea your attorney thinks is most important.

What you get out of it is a web page or a handout or an email that cuts down the calls from people asking what the rules mean. That's time back for whoever answers the phone, and clarity for the reader in the first place.

This works the same in Claude, ChatGPT, or Gemini. Pick the document that generates the most confused phone calls and give it a try.

handsonaitraining.com