I gave ChatGPT the same request twice. One answer I would never post, and the other sounded a lot more like me. It was the same AI and the same request, and the only difference was what it knew going in.

If the answers you are getting from AI are not where you want them to be, chances are high that it comes down to how much you are forcing it to assume. This is what context is, what it changes, and a brief you can copy and fill in.

Read the full transcript if you would rather not watch.

Jump to a section of the video

It knows a lot about the world and nothing about you

An AI model has been trained on a huge amount of general knowledge. When you open a new chat, it knows nothing about your business, your customers, or your relationships. It only knows what you give it. That is what context is: what you type, the pictures you show it, the transcripts and documents you feed it. It does not matter which model you use, because all of them need it.

Say you own a business and you ask someone to send a quote to the Johnson account. Your best employee knows the Johnsons, knows their price list, and knows what it costs to ship to them. A new hire is completely in the dark and has to guess.

AI is the new hire. Does it know your company, your pricing, your history with that customer, whether you bill them every 30 days or give them 90 days to pay? There is a long list of things it does not have. When people tell me AI is not giving them what they want, I would bet that most of the time it comes back to the context they gave it.

Test one: the same LinkedIn post, with and without context

I opened a plain ChatGPT session, not inside a profile, so it had nothing on me. I asked it to write a LinkedIn post about why business owners should think twice before signing up for another SaaS tool.

It gave me a long post that opened with "Before you sign up for another SaaS tool, ask yourself one question." It was written for business owners, and it loosely pointed in the direction of what I believe. It was also vague, there was nothing personal in it, it did not sound like a real person, and it made some things up. I would not post it.

Then I started a new session. This time I gave it my website, how I talk, what I believe, and who I am writing to, and I put the exact same request at the end. The post it wrote opened with "Not every business problem needs another SaaS tool. I know that's not what most software companies want to hear."

That one was far better. It sounded more like a person, it said what I actually believe, and it did not make anything up. I still would not post it, because I am picky about what goes on LinkedIn and it needed more context than I gave it. The point is how far the answer moved from one change.

A vague request and a specific one

The same thing shows up in everyday questions. I asked, "Can you recommend me the best steakhouse?" It had no location from me, so it went and found one on its own from my internet connection and listed places near there. I never gave it that, and if I had been on a VPN it could easily have been wrong.

Then I asked what I actually wanted: a steakhouse in Kansas City, downtown, that has been in business for more than ten years. With the city, the part of the city, and how established I wanted it to be, the answer was specific enough that I wanted to go eat there.

Test two: when it does not have your numbers

For the last test I asked what a made-up product called AcmeCRM would cost, as if it already knew the product, the pricing, and how much I would use it. It had none of that.

This time it did the right thing and asked me for more information instead of guessing. That is worth knowing, because it does not always happen. Sometimes the AI asks for what it is missing, and sometimes it just blurts out an answer and hopes you like it. You cannot count on it to tell you when it is guessing.

Why context changes the output

When the AI does not know something, it fills the gap with a guess. It does not know your business unless you tell it.

Showing works better than describing. Two real examples beat an instruction like "write casually." If you want help with your emails, give it emails you have written and liked. When I work with AI, I give it as much as is reasonable about my company, my objectives, and the problem I am facing right at the start, and I end up having to prompt much less because of it.

Every assumption is a potential error. When you hand a new employee a task without the training or the background, they make mistakes, and the same thing happens here. Sometimes the guess turns out right, and the models are getting good, but the fewer assumptions it has to make, the better. Every piece of context you give it is one less guess.

You would not want a salesperson or a vice president of sales guessing their way through the job. You would want them to have what they need to make accurate decisions. Think of the AI the same way. It is really smart and really stupid at the same time, and it needs a lot of context to give you good output in a specific area.

A context brief you can copy

Fill this in, paste it first, and put your request last.

If you are not sure what to put in it, ask the AI: "What context do you need from me to give a better answer? What assumptions are you making?"

Depending on the tool, it may or may not keep this between conversations. When I recorded this, the places to save it were projects and memory in Claude, memory in ChatGPT, and gems in Gemini.

Better answers start before the question

I know it sounds cheesy, and I believe it anyway. A prompt is just a question. If you put the work into the question and give it the context it needs, the answer you get back is in a different league.

This pairs with why AI gives you generic answers, which covers three questions that show you what the AI assumed, and with how much business owners should trust AI. If you want your whole team working this way, that is what our AI training is for.

If you have questions about any of this, or you disagree, reach out.

Full transcript

What I said in the video, start to finish. It is unscripted, so it reads like talking. Each timestamp opens that moment on YouTube.

0:00 Same AI, same request, two answers. One I'd never post and one that actually sounds like me. The only difference was what the AI knew going in. If the answers you're getting from AI aren't where you want them to be, chances are pretty high. It's because of how much you're forcing it to assume in your interaction. So let's break down what context is, why it changes everything, and I'll give you a template you can copy. Today, I'm really excited to talk to you about how you can get better answers out of your AI. So let's jump in here.

0:36 What I'm going to be talking about today is how prompting works, right? It knows a lot about the world, right? It's been trained on a ton of general knowledge, but it knows absolutely nothing about you when you come into it. So it doesn't know your business, your customers, your relationship, any of that stuff. It only knows the information that you give it, right? So that's what context is. Context is this foundational, extremely important part of getting quality output from your AI that you're dealing with.

1:11 It doesn't matter the model. All of them need context, right? So what you type, the pictures you show it, the videos, transcripts, any documents you feed it, that's what it's going to remember. But also, that's, That's the information it needs to actually accurate, the information it needs to accurately respond to what you're requesting. So here's an example, right? Let's say you own a business and you ask someone, hey, could you send a quote to the Johnson account, right? And you know, your best employees, they know the Johnsons and what they're

1:51 price list is, what it costs to ship them the product or whatever. But a new hire has to completely guess and is completely in the dark. And AI is the exact same thing here, right? So if you're asking AI to send a quote to the Johnson account, does it know your company? Does it know your pricing? Does it know your relationship with them? Does it know, you know, do you bill them every 30 days? Do they have to pay within 90 days of receiving it? Like there's this massive list of context that it doesn't have. And that's the problem.

2:25 See so much people complaining about AI sucks it's not giving me what I want nine times out of ten I I I would bet you it has to do with the context people are feeding it in their interaction how does context change the output so I wanted to show you live this so we're just going to pull up a generic ChatGPT not even in a profile right there's no context here so I ask it to write a LinkedIn post about business owners who should think twice before signing up for another SaaS tool. And look at this. It gave me a long post, right? And, you know, some people might copy and paste this.

3:10 But why this is important as an example, right? I guess let's just, let's just read through it a little bit. Before you sign up for another SaaS tool, ask yourself one question. Do we really need it? Are we just adding complexity? Business owners are constantly being sold, the next big thing, et cetera, et cetera. It just keeps going and going. There is nothing personalized about this, right? It does not sound like a real person for the most part. Sure, it's written for business owners.

3:43 Does it say what I actually believe? Overall, kind of? Right? It's a vague. It, it, it, it, it, it, it made some stuff up I definitely wouldn't post this right so I'm I'm going to take a second here I'm going to write a very similar prompt I'm going to start a new session and I'm going to add context to this here's the thing in this one I I gave it my website I gave it my tonality I told it kind of what I'm believing and the perspective of who I'm writing to with the exact same prompt over here added to the

4:18 end of it right now here is here is how it's it acts it so differently. Not every business problem needs another SaaS tool. I know that's not what most software companies want to hear, but if you're a business owner, it's worth thinking twice before adding another subscription to your monthly expense. Blah, blah, blah, blah. If you want to pause and read it, you can. Do I really need another tool? Blah, blah, blah, blah. Okay. Let's put this to the side. Now,

4:50 does it sound like a real person? Ah, a little bit better. It's still need, it still needs way more context, but this is like, I would say 15 times better. It is written for business owners. It does generally say what I believe, much better than the last one. It didn't make anything up, but I still wouldn't post it. I don't like posting stuff to LinkedIn, and I get really picky about it. But that's just kind of there to show you an example, right a similar example I could give you is um let's do this I'm going to start a new session here again and can you recommend

5:33 me the best steakhouse right that is super vague no context it's pulling my IP address so it's showing I'm in Colorado Springs but that's it being shifty right like I didn't give it that context it had to go dig that up and what if I'm using a VPN maybe it's wrong right so it's listed some but let's let's actually tell you do what I want what I actually want with context can you recommend me a steakhouse in Kansas City that is downtown and I don't know has

6:28 been in business for over 10 years right so this is more context I'm giving it the location um the city the location of the city and then how old I want it to be and it's giving me a more specific outreach, right? Or not outreach, output. It's giving me more specific output. And now I want to go to that place for steak. Anyway, the goal is here is to give you an example of when you're interacting with AI, if you're not getting the output that you want or desire, it's highly recommended to assess how much context you've given it and if you need to give it more.

7:11 Obviously, there's going to be times when it just struggles to, to answer your prompts appropriately, but context that plays a massive part here. This will show off of the assumption. So here's another example of not enough data. And this is assuming that it knows what AcmeCRM is and that it can scrape the data and it knows how much I'm going to use it and all this stuff, right so we asked how much will that cost now it's instantly this is this is cool it's growing right it's starting to ask you for more context because we didn't give it to it right and there are times when

7:53 it will do this and there are times when it won't and it'll just make assumptions right so let's go back to that it didn't it didn't even guess because it doesn't have all the context so we need to give it the price and the setup and all these all these things right this is a great example of when you're interacting with AI there are times when it will ask you for more context and there are times when it will just blurt out whatever you ask it and hope that you like the answer so keep that in mind now why why it has this impact it fills in the gaps with guesses assumptions right If it doesn't know your business, right, it's, how is it supposed to know?

8:36 You've got to tell it, right? It doesn't know your business unless you tell it. You got to show it, don't describe it. Two real examples beat "write casually", right? So like if you're wanting help with your emails, feeding it information of other emails that you've written that you like, that's going to help it exponentially. When I'm interacting with AI, I'm giving it as much information that I think is, reasonable about my company, my objectives, the problems I'm facing, because I have to prompt much less because I'm giving it an adequate amount of information right out the gate.

9:12 That's why I'm just super into this. And every assumption is a potential error, right? So when you have a new employee and you tell them to do a task and you don't give them the proper training or context, they're going to make mistakes. And that's what happens with AI. The assumptions, in my opinion, are a mistake. Sometimes they can be right. You know, it's getting pretty good, but the less assumptions AI has to make the better. So every piece of context you give it is one less guess it has to make. One less guess.

9:42 So let's go back to that example of you're a business owner, right? Do you want your employees having to guess to exist as their role, right, as a, let's say an SDR or even, I don't know, a vice president of sales, right? You don't want them making assumptions. You want them to have all the information they need to make accurate decisions and, and, yeah, reduce the assumptions they have to make in their professional life. So just think of the AI in that respect, right? It's really smart, but it's really stupid.

10:18 And you have to give it a lot of context in order to make it give you really good output in specific arenas, right? But anyway, here's some resources for you. You can screenshot this. I guess we could put this context brief in the description below. But, you know, just who are you? What do you believe? Who's your audience? How do you write? Examples of your writing. Facts, facts. facts as possible and the tasks of what you want it to do if you don't know or you don't want to copy this model you can

10:59 just ask the AI hey what context do you need to give me even a greater output what assumptions are you having to make to interact with me what load bearing questions do you have to give me a higher quality output and also like depending on the type of AI you're using it might not be saving this context. Some of them might be. So how to make it stick in Claude. They're using projects in memory. In ChatGPT, they just have memory in Gemini. Their gems, soon it'll be called skills. So the better answers start before the question.

11:33 And I actually really believe this. I know it sounds cheesy. But it's like, if you're able to curate a really good question, a prompt, right? We all know prompts are just questions. If you can curate. it and give it all that context, you're going to have an incredible answer. And in order for it to filter through all of that and give you a good output, context is king. I'm Tim from Norsoft. Thank you for watching. I appreciate you tuning in. If you have any questions, if you disagree, let me know in the comments.

12:06 I would love to hear your thoughts.