Why does AI give generic answers about my business, and how do I fix it?

AI gives generic answers because it starts every conversation knowing nothing about you. It hasn’t read your emails or sat in on your client calls, so it fills the gaps with what an average business would say. A better prompt won’t close that gap, because a prompt is one sentence and your business is years of decisions. The fix is context: write down once who you are and how you work, give it a file for each client, and let it read what’s happening this week. Keep working in the same place and that context grows, so the answers get more specific every month.

Why does every AI answer sound like it was written for someone else?

You’re using ChatGPT or Claude or Gemini or whatever else. You ask it to write an email for a client or to create a proposal or a LinkedIn post, and every time it spits something back out, you find yourself editing it. AI was supposed to save you time, but the edits you’re making don’t feel like that.

You’re definitely not alone here. Everybody I talk to who’s just using AI in the Claude or ChatGPT chats has this same problem. They log in with something that they want AI to help with, they ask the question, and AI gives out a generic response that it thinks is the best solution to their problem. The reason it’s generic and not tailored to you is because AI knows nothing about you or your business.

Every new chat starts from zero, with nothing about you, so it writes the average of everyone.

For example, we can take a look at the question of writing a client follow-up email. The generic ChatGPT response will be something along the lines of:

“I hope this email finds you well. I wanted to reach out and provide you with a status update on where things currently stand with your project…”

The whole thing sounds super generic and there’s nothing specific about the client. The AI didn’t know that the client likes quick bullet points, nor did the AI know about what was actually going on in the service delivery.

The AI that knew about the business knows that the client is doing what they call “the Refresh” (a specifically named initiative in that business). It also knows that an addendum has been signed and locked in and which platforms the client’s using and anything still open from email and video call threads.

The response you get from the AI that knows is something along the lines of:

“Just a quick update on the Refresh. – Addendum: signed, returned, locked in. Scope for now is the booking system only, the website rebuild is still on for Q1, just not this quarter. – Platform: landed on the one that handles multi-person calendars cleanly, that was the actual failure point behind the double bookings. – Still open: no word back on the old logo files…”

What does the generic answer cost you each week?

Ask yourself how long it actually takes you to fix your AI drafts versus writing it yourself. Because if you’re not actually providing any information in your prompt, you’re likely going to get something you’ll have to rewrite entirely. An entire rewrite takes longer than it would have to write it yourself in the first place because the rewrite had to go through AI first. In the best case with AI, you give it some information and it gets you anywhere from 50 to 80% of the way there and you’re only writing or rewriting a few lines, which isn’t really saving you time because that probably ends up taking a very similar amount of time to writing it yourself.

If we take a look at the numbers, AI is saving workers roughly 11 hours per week through automation alone, but there are 6.4 hours a week going to botsitting. Botsitting is the time spent making AI usable. So you can do that easy math. More than half of the time is spent fixing what AI outputs. However, 53% of the people in that study say that critical information they need is not accessible through their AI systems. Which means for just over half of the people running AI, they have not given the AI the proper information it needs to do the work properly. And yes, this goes all the way down to it not properly working on your email. The fix that half of these people are facing is simply giving AI the information it needs.

This means that the 6.4 hours of rework is actually up for grabs. Right now it’s 4.6 hours of efficiency gained, but the more we improve our AI output, the closer we get to our 11 hours per week. Even if it was only one hour of rework, that’s still 10 hours a week that we gain back from AI. Over the course of a month, that 10 hours turns to 40, and 40 hours a month is a full workweek gained. All from simply giving your AI more information to do the work it needs.

Sources: Work AI Index 2026 (Glean)

Measure yours in 3 minutes: the free Task Audit scores how much of your recurring work could run without you.

Why doesn’t a better prompt fix it?

Writing a better prompt doesn’t really fix anything because it just changes where the rework happens. You might see some gains and AI gives you some better results, but what happens when you need to update your prompts?

Maybe you learned how to write better prompts and you’ve built a prompt library or a custom GPT. You’ve trained these prompts as well as possible and sure they’re giving you better outputs. But the question is, once again, are the prompts actually tailored towards what’s happening in your business right now? I’ll admit they might be, because the prompts have fill-ins where you give that information, but that’s still an extra step on you.

But once again, there are more solutions to this. Take projects as an example. Claude and ChatGPT both offer projects that allow you to save instructions/prompts and specific files to give better output. But what happens when a strategy changes or there’s new information to update or something is so outdated that it’s not giving the right information anymore? It is then up to you to go and update the prompt, the instructions, or the files to correct it. A project should suffice for something that’s simple. But the more complicated you get, the more likely any of those pieces of information will need updating.

No matter the situation or what tool you’re using, the issue will always arrive back at giving AI the proper ‘context’ it needs to do the work. To get the best results for you and your business, your AI should know about things like the offers, clients, how you do things, or what happened last week. Basically, AI should have the right context about your business.

Context just means everything your AI can see when it answers you. A blank chat has nothing. A structured prompt has some stuff, same with a project. But the real key is providing enough information to fill the context window to answer specifically about you. The context window is simply what the AI can see in the current conversation, so kind of like a working memory.

It’s a little technical, but what I just described is basically what “context engineering” is. It’s giving the window just the right amount of information for the next step. Giving AI way too much information will worsen the output, and too little gives weaker results. So giving your AI the instructions and the availability to go fetch the right context it needs is the key.

What does your AI need to know before it can answer like it works for you?

AI should be able to gather the context it needs by itself. This means you need to give it access to that. In Claude Code, this is as simple as setting up a folder on your computer and starting Claude Code within that folder.

You can create a context subfolder in that and put everything it needs in that subfolder to reference later. Or if that data lives externally, such as Fathom recordings or emails, you simply set up the connection tools required to go gather them and anytime you reference that information, Claude Code has access to go get it.

There are a few things that you need to basically tell Claude Code about your business to give it an accurate representation.

There are some core things that should go in your context folder:

• Documents on information about your business and the offer that you provide.
• A document on your avatar or ICP explaining who you sell to.
• A testimonials document.
• A voice profile to show AI how you talk. (This can be created for you.)
• And a key metrics file. This file does require an automation to be gathered for you, otherwise you would have to update it manually.

Additionally, you should create a subfolder for all of your clients. This just allows AI to help you understand who you’re serving on top of how you serve, created in the context folder. This one can include any and all information about how you’re currently serving your client, but it does need one central file that explains who the client is and how you reach them. If you don’t have the contact information, such as email, name, phone number, then AI will be unable to gather call recordings or email threads for you.

Once you have all of this, AI will understand who you are, what you’re doing, who you’re serving, and then be able to give you tailored answers like as an employee of yours. After a client meeting, it’ll be able to pull that call, understand what your offer is and understand where the client is currently sitting in the process and give you recommended next steps or create client updates like it’s actually working with the client. And that’s just one example on top of everything else it could be doing for your business.

You could give it things like marketing info and strategy and it’s able to understand what your social medias are doing and how they’re performing. You can give it your sales call strategy and it’ll become a sales coach that has access to your actual calls.

And the best thing is that every time any of this information changes, all you simply need to do is tell AI what’s changed, in which file, and it goes and updates them for you. Every time you have a new client, you simply say so. Or if you have a new piece of strategy that you want to implement, you just add it to the context folder. And instead of spending time on the rework and the info providing or the prompt switching, everything’s already there. And a prompt is simply an instruction to do a task rather than an intricate, fully laid out paragraph about what to do.

What happens when a system starts from your own task list instead of a template?

When we specifically take a look at AI for automation and getting work done, this becomes even more powerful. A client of mine has over 200 tasks per project of hers.

“I actually realized that we had over 200 tasks per project that had to be checked, and some of those had subtasks. So when that was in my face, it was a pretty big aha.”

Her issue was trying to find a project management tool that worked for everyone on her team. The generic tools were decent, but with 200 tasks, none of them could really accommodate what they were doing.

The solution we found was building her own custom task tracker. Which for 200-plus tasks seems a little difficult, but all we did was give her AI workspace a long document of all the tasks in order and how they happen. AI was able to read this as context and use it in building a project management Google Sheet with the correct scripts to create a fully customized project management tool that they’re now using every day.

They could have started and said that they wanted a custom project management tool, but doing that by hand and going back and forth with AI would have been a nightmare where simply just giving the SOP document made it ridiculously easy.

“My entire team can be in the tracker and we can each see what’s been done.”

The result of using her context to create a system that really worked for her was around 25 to 30 plus hours a month back by her own count. And she thinks that’s actually lower than what it really is.

But the point here is that she only gave it one document to make her task tracker. She definitely didn’t give it her entire Google Drive, as that wouldn’t have been helpful for AI.

What should you keep out of your AI’s context?

Context doesn’t mean dump everything into AI and expect it to be better. When explaining context engineering, Andrej Karpathy, a founding member of OpenAI, pointed out that more isn’t actually better. Too much or irrelevant context actually costs more and makes answers worse.

So you shouldn’t be feeding it everything at once in the first place, but more importantly, you shouldn’t be feeding things that are completely irrelevant.

This includes out-of-date material. You should be removing old pricing, retired offers, last year’s processes, or anything like that from your workspace. AI doesn’t know it’s the stale version unless explicitly said somewhere in that document. AI is just going to see it and assume it’s the right thing and use it for its response, giving you incorrect information.

Sensitive information should also not go in your workspace. Passwords, a client’s sensitive data or anything like that. You can turn off the data-sharing setting so consumer ChatGPT and Claude plans can’t actually train on any of this. This is more along the lines of AI accidentally reading the information and putting it in a document and you not catching it. If a password or an API key or sensitive information is put in something that it shouldn’t be and you miss it, it’s on you for the repercussions.

And as I said earlier, you don’t want AI to have everything all at once. So the best way to allow AI to only have the things it needs is by giving it specific instructions. In Claude Code or Codex, you can use the CLAUDE.md (Claude Code) or AGENTS.md (Codex) files to provide instructions for every new chat to follow. In these files, the simple instruction is to tell the AI to only use the information it needs and to tell it where that information lives so it doesn’t search over the whole workspace. The same applies to the tools you give it as well. Combine those two instructions. It only ever uses the tools it needs and gathers the information it needs, preserving the context window.

When we put the context together correctly, we get something that can write a client status update that knows who the client is, leaves out sensitive client information, and uses the tools necessary to gather email threads and call recordings to update the client file accordingly. We get AI that’s giving specific results like it works for us, leaving out the things it doesn’t need, and updating the information automatically for us so we don’t have to.

The first thing I recommend you do to set up your own context is after you’ve downloaded Claude Code or Codex and you’ve opened a new folder on your computer, have AI interview you about all the things I listed in that context section. Ask it to ask you questions until it has a great understanding of who you are, what you do, who you serve, so on and so forth until you’re satisfied.

Sources: Shopify CEO and ex-OpenAI researcher agree that context engineering beats prompt engineering (The Decoder)

And if you’d rather I just look at your business directly, I do a free discovery call.

Frequently asked questions

Why does every AI answer sound like it was written for someone else? Because AI knows nothing about you or your business. Every new chat starts from zero, so it writes the average of everyone, and you get a draft with nothing specific about your client, how they like to be updated, or what’s actually going on in the work.

What does the generic answer cost you each week? Workers say AI saves them roughly 11 hours a week, but 6.4 of those hours go to “botsitting,” the time spent making AI output usable. That leaves about 4.6 hours gained. Cut the rework to one hour and it’s 10 hours a week, around a full workweek every month.

Why doesn’t a better prompt fix it? A better prompt changes where the rework happens. Prompt libraries, custom GPTs and projects all hold some information, but when your strategy, offers or clients change, it’s on you to update them. What the AI needs is context: everything it can see when it answers you, including your offers, your clients, how you do things and what happened last week.

What does your AI need to know before it can answer like it works for you? Your business and offer, who you sell to, your testimonials, a voice profile, and a key metrics file, plus a folder for each client with who they are and how to reach them. With access to your call recordings and email, Claude Code can pull the latest call and write a client update as if it had been working with that client. When something changes, you tell it what changed and it updates the files.

What happens when a system starts from your own task list instead of a template? A client with over 200 tasks per project couldn’t find a project management tool that fit her team. Given one document listing every task in order, AI built a custom tracker in Google Sheets that her team now uses every day. By her own count it gives her 25 to 30-plus hours a month back.

What should you keep out of your AI’s context? Anything out of date, like old pricing, retired offers or last year’s processes, because AI assumes what it reads is current. Keep passwords, API keys and clients’ sensitive data out too, since AI can drop them into a document you don’t catch. A CLAUDE.md or AGENTS.md file tells it where information lives, so it only pulls what the task needs.

Photo by Maksym Kaharlytskyi on Unsplash