What this post covers: The cost of ‘digging’ for information is high. By implementing five specific, AI-friendly documentation changes, you can transform your Staff Handbook into an searchable knowledge base, saving staff time and eliminating friction.

The Friction of Searching Codified Sources

In the study of information seeking behavior, information is found in one of three places

• Internalized (in your head)
• Codified (Written down somewhere)
• Social (ask someone about it)

When information is not internalized, people look to a codified or social source.

Tom is looking for the form to fill out to get his student a Mifi device. He starts digging into the written, codified, sources for the directions. Tom’s email. The staff handbook. The Schoololgy course. Tom checks everywhere he can and can’t find it.

Tom runs into friction. He starts to wonder. Maybe this is a waste of time. Are the resources up to date? Are they complete? How long will it take him to search? Is it even there!

Tom’s brain, when faced with multiple ways to figure where this form might be, will choose the path which uses the least perceived resources. Those resources might be time, effort, or cognitive load, amongst others. Researchers call this the Principal of Least Effort.

The Principal of Least Effort causes Tom to stop looking through codified resources. Tom needs to protect his limited resources (time/cognitive load) to find an answer. So he pivots. He turns to a social solution and walks over to Dave’s desk.

Tom still spends additional time by interupting Dave. And, now Dave is spending extra resources, answering a question he has already answered elsewhere.

And, let’s be real. We have all been Tom. And we have all been Dave.

Let the Robot Answer The Question

We can reduce the friction by putting our codified sources into AI. By feeding the AI our Staff Handbook (the codified source), staff can essentially “ask a question” as though it were a colleague (the social source). This shift allows them to ask AI Buddy first. However, realizing this potential required rewriting the handbook; it needed to be structured not just for human readability, but for AI readability as well.

Here are the 5 big changes I had to make to the Staff Handbook.

1. AI Readability
   • The Text Only Rule: Text Over Images
   • The Context Rule: Use Descriptive Hyperlinks, never “Click Here”
   • The Visible URL Rule: Make Hyperlinks Visible
2. System Design
   • The Knowledge Expansion Rule: Include Additional Resources
   • The Continuous Feedback Rule: Have a Feedback Loop

The Text Only Rule: Text Over Images

Problem: Gemini Gems can’t read images.

Here is a screenshot that was in the Handbook.

Buddy AI can’t read the screenshot. Buddy AI does not know know the MIFI Request Form is there. So when asked about the MiFi Request Form, it does not have any information.

Solution: Use Text instead

Now the section is readable by AI Buddy and people. I’ve also added more context about where to find the forms.

The Context Rule: Use Descriptive Hyperlinks, never “Click Here”

Problem: Gemini Gems does not know the context around a link.

Web-readability people will tell you not to tell someone to click here. It is not elegant, but humans can read and understand where to click. For Example:

The Calendar for the SY25-26 quarterly Progress Report Marks and Interims is posted HERE.

AI Buddy has trouble knowing that the report card calendar is in that link. When you ask it where the interim calendar is, it will not be able to tell you consistently.

Solution: Add more context to your links.

The Calendar for the SY25-26 quarterly Progress Report Marks and Interims is posted in this Google doc.

This is still very readable by people. AI will know where to find the document.

The Visible URL Rule: Make Hyperlinks visible

Problem: Google Gemini Can Not Read Embedded Links

This seems like it might be in conflict with “Descriptive Hyperlinks,” but stay with me.

In the link below, I’ve embedded a link to https://www.redroverk12.com/OLPS

Go To Red Rover to request a sub

That is very person friendly. And it is also a very descriptive link. Both people and AI Buddy can use that text lead people to Read Rover

Google Gemini can read the plain text, but it cannot read the embedded link. So it can not see that the hyperlink is to give the user.

Solution: Create Text-to-URL Mapping.

Since Buddy can only “see” the visible text, we need a way to connect the descriptive phrase to the URL:

 Go to Read Rover to request a sub =  https://www.redroverk12.com/OLPS

We can include that Text-to-URL Mapping in a few places

• At the bottom of the document
• In another column, next to the text
• In another document, also included in the Google Gemini Knowledge base

Creating the Text-to-URL Mapping is the most challenging part of this process. l’ll dedicate an entire follow-up post to strategies for Text-to-URL Mapping.

The Knowledge Expansion Rule: Include Additional Resources

Problem: Your document is not expansive enough.

We have a Staff Handbook. But, there is no way to keep everything we need in the staff handbook. The document would be too big. And, once it is too big, you less people would use it.

You would need a “handbook” for each department. The Finance department would need one. The Tech Team would need one. The Instructional team would want one. And, that adds friction as no-one wants to read through mulitple documents.

Solution: Have mulitple souces of information in the knowledge base.

Every department that gets asked the same question multiple times SHOULD have a handbook. We can link all those handbooks and Google Gemini will find an answer.

I’ve also started experimenting with an “Admin Update” file. When an admin sends out an update, I add that to the Admin Update google doc. That way Google Gemini has access to those as well.

The Continuous Improvement Rule: Have a Feedback Loop

Problem: You will not know if someone does not find what they are looking for.

If a person asks Buddy AI a question, and Buddy AI does not the the answer, there is no way to find out what wa smissing.

Solution: In the Gemini Gems directions, automatically generate a link to submit issues.

In the instruction section in the AI Buddy’s Google Gem, I added this instruction.

If there is not an answer in the Knowledge please say, "I'm sorry, I do not see this in the knowledge base.  If you think it should be added submit this form.  http://example.com/add"

The form emails me directly. I can reach out to whomever might be the appropriate resource and ask for more clarification. I can also reach out to the questioner and say the question is out of scope for AI Buddy. I’ll add more on AI Buddy’s limits below.

---

Know Buddy’s Limits

At the end of the day, we want to solve the problem Tom faced in his office. The time he spent searching for the MiFi form is time stolen from making his classroom a better place. By building our handbook so that both humans and AI can read it, we create a system that serves our staff and lets them focus on supporting students.

Keep in mind, AI Buddy is here for the paperwork, not the pedagogy.

Updates

Ok, 20 weeks until I turn 50.

A few updates from this point, back at Week 46.

One thing I’ve been dealing with is panic attacks. They were always infrequent, but over the summer they started happening more regularly. After seeing several medical professionals, we tried a few changes. I tried some medications and lifestyle changes. The thing that really seemed to help the most was exercise.

I’ve never felt the importance of exercise. The idea of exercise I understood, but not actually feeling it in my body. The cardio exercise from the Meta Quest Headset was wonderful for my mental health.

So my panic attack episodes have lessened, but weight was not changed. So I kept the idea of exercise, but focused on something else.

I’ve been working on progressive weight training for the last 11 weeks. Next week I do a follow up DEXA scan. I’ve paid no attention to my weight or eating. My weight has gone up. But, my mental health is better (only one panic attack, and it was the week I did not exercise). I feel stronger.

If the DEXA scan shows I gained muscle, I will continue with the exercising (I’ll do a future write-up on the details) and focus on the eating side of the equation.

I signed up to play Dungeons and Dragons.

Over the pandemic I got really into Critical Role and Dimension 20. And, growing up nerdy it always seemed like a think I might enjoy.

I started playing two different games online, which I found through Roll 20. One fell apart by the third session. One of the players was consistently a poor team player and all the other players slowly quit. The other has been going on for 10 weeks. I think we are probably 70% through the adventure.

I don’t feel as if I am “good” at the game of D&D. I pay attention. I have a goal of what to do on my turn. I take notes. I don’t step on player agency. That feels like it gets me to “not bad.”

I’ve been watching a lot of videos. Here is my focus for moving towards good.

Make Stronger Choices

Emily Axford (of Dimension 20 Fame) talks about making strong choices. The DM creates a world for people to inhabit. Making strong choices to fully engage in the world shows your team-mates¹ you care.

I think it has to be strong choices that move the story forward or your group together. In my game that fell apart, the problem player made strong choices. But his choices did not move the game forward. For example, he kept trying to pick-pocket people. He tried to capture a bat and make it a pet. None of his choices were honored of the other players or the story.

Think About Improv

I don’t mean this in the way of “I want to be funny.” I have some trouble interacting with my character. I feel very clunky. I wonder if learning about improv would help give me strategies to help in those moments..

I keep hearing that improv has rules and structures to help. The whole idea of “yes, and” is one. So I purchased a few books on improv to help understand how to be better in those moments.

Painting Better Pictures

In combat I like to talk through my decision. “My character sees that the princess is in trouble. So I am going to move over and attack the evil troll and try to make a trip attack.” I like the idea of describing the action more. So maybe “my character sees the the princess is in trouble. I run over, drawing my sword, the smell of the troll washes over me. I aim for the trolls feet to knock him down.”

Some of that is deciding to do it. Some of it is pre-planning and figuring out a structure to help start. What is going on. Describe the action. Add in feelings or senses. Eventually that might become natural.

Support Fellow PCs

This might just be a sub-section of the first two. Support your fellow players. Interact with them in a way that is fun, but does not take away their agency. Support their goals and find out ways to make them shine.

Here is a follow up to 60 to 50.

Exercise

Burpees were mixed. My cardio strength got better. But my ankle started to hurt. I don’t know if it was causal, but I developed Plantar Fasciitis. And, it was not fun. I did not look forward to it.

So I pivoted. I purchased a Meta Quest 2 Headset and every day I do exercise apps for 30+ minutes. Still too early to see the results, but I do look forward to it. I’m still figuring out how to include burpees on a regular basis, as I think it did me a lot of good.

Meditating

I do meditate a few times a week. I go back and forth between the Insight Timer App and one on the Quest called Tripp.

Counting Calories

I’m working on the MIND diet. I’ve reduced the amount of red meat. I eat beans and greens every day. I have not upped the fish as much as they want.

Journaling

I am not sure journaling is the way to solve my eating and anger issues. I was listening to a podcast on a book called Burnout. It resonated a bunch. I ordered the book.

I’ll update in a few weeks.

Next year I turn 50. I sounds so old. If I am lucky, I’ll get maybe 30 years? Less, being an obese man.

On one level it seems like so little time. I’m maybe two thirds done with my life. But, it could be filled with so much possibility.

I’ve grown and changed so much over the last 30 years. I’m a better person then I was then. Twenty year old Kevin never thought he would be married. Or have a child. The last thirty years have been better than I would have guessed. So I am hoping the next thirty will be better still.

The only way to be better is to try to make it better. I can’t tell what the universe has in store. But, I can commit to growth and change.

In 60 weeks I will be 50 years old.

I want to be healthier.  I don’t want to be obese.  I want to be strong.

So, 60 weeks.  In my head, I am thinking of 10 week blocks.

First block, start exercising, meditating, counting calories, and journaling.

Exercise

I’m trying burpees.  Specifically, this program. I would spend more time driving to the gym each week.  And I still can’t make it a full twenty minutes. So this is a good start.

Meditating

I downloaded two meditation apps.  Waking Up and the Insight Timer.

Counting Calories

I have the lose it app.  Everything I read says this is vital to do. My eating focus will be the MIND diet.

Journaling

This is more about self reflection.  I was eating and I realized I was eating and angry at my spouse. I need a way to reflect on this. 

So that is the first plan.  I’ll give it an update in 10 weeks. 

I’m just angry today.

I’m angry at my son’s medical issues he is going through. I’m angry with his nurse. I’m angry with my job. I’m angry with my weight gain. I’m angry at my small apartment and the constant stress of moving into our new house, which seems so far away and yet stupidly close. I’m spiraling with anger.

I opened up Twitter, a site that in a post Elon Musk purchase, seems level set-on making everyone angry. I started scrolling and rage builds. People screaming about a movie casting they disagree with. A pop star being anti-sematic. Angry politics. I started writing an angry tweet of my own.

I stopped.

I wrote here instead. This is my thoughtful little place. No angry messages without nuance. No virtue signals, especially if that virtue is anger and rage and hate.

I keep going back to a story about two albums. I think it helps me reflect on what to do with my anger. Musician Trent Reznor was in a bad place. He was depressed and he wrote that depression. He gave it voice and it became the Downward Spiral. Reznor says that

“The self-destruct button was pushed when I first started writing,” he says. “There was a sense that I couldn’t fit in anywhere, I couldn’t relate to people; I felt alone, I felt angry about it. And part of me is still that. I felt like I was heading down into something that wasn’t going to have a good ending. That ended up being addiction: its claws were in me but it hadn’t fully revealed itself.”

The Guardian Story about Trent Reznor

He started drinking and using drugs and ended up overdosing on heroin half a decade later.

And another musician, Beck. I remember this story, but I can’t find it online. So if it turns out to be false, I am sorry, but it helps me get through the days. Beck, also in a bad place, started writing out his pain. But he realized what it was doing, how it was hurting him and stopped. He instead to head the opposite direction and write himself out of his pain. Odelay was the album that came out of that experience.

The truth is I do not know what to do with the anger. The only thing I’ve found useful is to move away from what the anger is encouraging me to do. It is a fuel that burns and leaves you choking on fumes.

On the bad days I try to stop before I go to far. Sometimes I remember to listed to Beck and make something that will feed love and calm and beauty, not anger.

I talked here about how a major falling of ChatGPT is taking an article and successfully figuring out the author’s main point. My guess is, because it is built on writing via predictive text, it does not do well if the outcome would varies from what is generally accepted,

But, what if we did the opposite. What if we gave the AI a list of facts for the article. Then we told the AI the main point we wanted. Then we ask the AI to write the article.

For example, I want an article on the founding fathers, maybe George Washington. I could include all the details I wanted and come up with a main idea.

Then, I could have the AI generate a bunch of different version. I could have a version for kids who like soccer. I could have a version for kids who like Pokemon. I could have a version with easier English. I bet I could specify certain key words that would need to be included in any version. I bet I could make a Spanish version as well.

Bookriot.com has an article titled: LIBRARIES ARE MORE POPULAR THAN EVER AND LIBRARY WORKERS DON’T EARN LIVABLE WAGES: THE STATE OF U.S. PUBLIC LIBRARIES

The first paragraph talks about how in-person usage is down, library use due to digital access is up, and libraries are having trouble with funding, so workers are being paid less.

I prompted CHatGPT

“what is the main point of the linked article.”

ChatGPT responded

The main point of the linked article is that libraries in the United States are more popular than ever due to their adaptation and evolution in response to the rise of technology and digital resources. They continue to serve as vital community hubs, offering various services such as free Wi-Fi, computer access, educational resources, book clubs, and community events.

That is not the point of the article. The article is about how more and more people want to use libraries, but no-one wants to pay for them. The average library employee is paid BELOW a living wage for a family of three.

I tried upping the summary word count to 200, it still missed the point. On a list of key words, ChatGPT lists librarian salary as last.

I finally said,

“can you write a 100 word summary of this article and give me a table of key words? https://bookriot.com/libraries-are-more-popular-than-ever/ Focus on the fact that no-one wants to pay for libraries

Chat GPT responded:

This Book Riot article reveals that libraries in the United States are more popular than ever, with approximately 144,600 libraries adapting to technology and digital resources. However, despite their growing popularity and expanding services, libraries face the challenge of insufficient funding as many people are reluctant to pay for their maintenance and development.

Table of Keywords:

Libraries
Popularity
United States
Technology
Digital resources
Funding challenges
Insufficient funding
Reluctance to pay
Maintenance
Development

ChatGPT could not give me a accurate summary of the main points of the article, unless I told it the main point of the article.

I am not great with Google Scripts. I heard that you could use ChatGPT to write Google Scripts. I had an idea of something that could make my life easier with ChatGPT.

So after a few hours asking ChatGPT questions and testing thins out, I got a workable solution. But, it also was done in a way that might actually cause me more work in the end.

Someone with a better understanding of Google Scripts might have been able to craft a better question. Or, could have told me the methodology I was suggesting ChatGPT to do was not the way to go forward.

ChatGPT can suggest the code, but can’t tell me if the code makes sense.