When AI gives you generic, off-brand output, it isn't broken. It's just missing context. The fix is to treat AI like a new hire and feed it the same five things you would give an employee: how to use your tools, your tone of voice, your brand guidelines, your agency values, and your templates. This piece walks through a six-level reliability pyramid, from a one-line prompt up to multi-file skills with code, and explains why most advisors should aim for level four: a saved skill that combines detailed instructions with letting the AI interview you, so you stop re-explaining yourself every time and your output finally sounds like your agency.
Sign in to view this.
Sign inSign in to view this.
Sign inYou've all seen this. You ask AI to create an itinerary for a five day Italy trip and it hallucinates a cooking class that was never in the brief. The slide title comes back as the dates instead of the destination. The voice reads like every other AI generated text on the internet, nothing like how your agency actually speaks. Or worse, it just refuses to do the thing.
The number one question I get from members of the community is "how do I get AI to actually give me reliable results?" That's what this is all about.
When AI gives you generic output, it's not broken. It's working exactly as designed. AI is a prediction machine, and when you don't tell it what you want, it predicts the most likely answer based on everything it's read on the internet. That gravitates to the average.
Safe colors. Classic AI voice. A Tuscany itinerary that reads like every other Tuscany itinerary online. For anything where your brand, your tone, your style matters, which is basically everything you do as an advisor, the average is exactly what you don't want.
The fix is context. AI only knows what you tell it.
Think about hiring a senior travel advisor with ten years of experience. If you handed them a two sentence brief and said "figure it out," even with all that experience, they'd have to make assumptions on every single decision. Detail level, styling, which room category goes on the itinerary, how transfer times get handled. They'd guess, and a lot of those guesses would be wrong.
AI is the same. It's not super-intelligence. It's not smarter than us. It needs the same context an employee would need.
There are five categories of context that matter: how to use your tools (Canva, PowerPoint, your CRM), your tone of voice, your brand guidelines, what your agency values, and your templates. The more of this you feed your AI, the more reliable it becomes.
And here's where AI actually beats training employees. You do this once and it's permanent. The same context can power every agent you ever use, you can update it instantly, and you can share it across your whole team so everyone gets the same quality output without you re-explaining yourself a hundred times.
There are six levels of how much context you give your AI, and each level is a step up in reliability. You don't need to climb the whole thing for everything you do. For simple tasks, level two is fine. For complex workflows that really need to match your brand, you'll want to get to level four or beyond.
"Make me an itinerary for a five day Italy trip." That's the whole instruction. Zero context.
This is where most people get stuck. They try AI once like this, don't see useful results, give up, and decide AI doesn't work. Of course it doesn't. You wouldn't expect a brand new hire to nail it from a two sentence brief either.
Same task, but you actually tell the AI everything it needs. Your role, the project context, your brand rules, what the output should look like.
A trick that helps a lot: dictate the instructions instead of typing them. You naturally give way more detail when you talk, and it's faster too. The output will be a bit unstructured, but that's fine. There are prompts that take messy voice notes and clean them up into proper AI instructions. Happy to share the one I use, just ping me.
This is already a huge jump from level one. For a lot of tasks, this is enough.
This is where it gets interesting. Instead of trying to figure out what context to give the AI, you flip it: tell the agent to interview you about anything it needs to get the task right.
Same employee analogy. If you gave a vague brief to a senior hire, the first thing they'd do is ask you a bunch of questions. AI has the same depth of knowledge, it's just too eager to help. It'll plow into the task without asking unless you tell it not to.
Once you do, the questions it asks are kind of a revelation. "When the brief lists multiple room categories, which one goes on the client itinerary?" "If transfer times aren't confirmed, do you estimate or leave it out?" Things you make decisions about every day but never wrote down. Now they're written down.
You don't want to write these massive detailed prompts every time you do a task. Same way you don't want to re-onboard an employee every Monday morning.
A skill is a saved file with all your context for a specific job. Your itinerary builder skill has your template references, your brand and style notes, your tone of voice, day structure, the gotchas you've learned. You write it once, and every time you ask the AI to build an itinerary, it loads that file first.
The beautiful thing is you don't have to write the skill yourself. Combine level two and level three. Give the AI detailed instructions, let it interview you, and then tell it "don't do the task right now, save all of this as a skill so we can reuse it." The AI helps you write the context file. It's just a text file. Anyone can write one, or have the AI write it with you.
This is probably the biggest jump in reliability you'll make. If you do nothing else, get to level four.
Even with a great skill, you'll notice subtle things the AI gets wrong after a few uses. You under-specified something. The AI interpreted a word slightly differently than you meant. Edge cases you didn't think about.
The fix is a loop. You go back to the AI and say "let's update our skill, this part is good, this part isn't." The AI starts self-experimenting with the wording. It compares the old version to the new version, judges itself on the problems you flagged, tests until the issues are gone, and then shows you the before and after.
You can keep giving feedback. "No, the old version was better here." "Yes, that's right." It's like a training session with an employee. This is what quality looks like, steer more this way, here's how we use our tools, this is what our brand needs to look like.
The more context you give it, the better. At level six, your skill becomes a folder instead of a single file.
One root file at the top with references to different scenarios in different sub-files. Maybe cruise itineraries follow a different process than land itineraries. Your color palette lives in one file, brand guidelines in another, a cruise template separate from a land template. The AI pulls in the right context when it needs it without loading everything at once.
The second piece is code. For working with tools like PowerPoint and Canva, code is the most reliable mechanism. You want slide layouts, date formatting, and data extraction to behave the same way every single time. Small scripts handle that, deterministically. And you don't write the code yourself. You describe what should happen, the AI builds it.
Three things at this level: organize your context into a folder so the AI gets the right info at the right time, document all your edge cases, and let the AI write code for the parts that need to be exact.
If you've read this far, you're probably looking at the pyramid wondering where to start. My answer is always the same: aim for level four.
Combine detailed instructions with the interview process and let the AI help you create a skill for the workflow that costs you the most time right now. Itinerary builds, client briefs, supplier outreach, whatever it is. Build that skill, use it for a few weeks, learn where it falls short.
Once you've experimented with that and you want to push a workflow even further, level six is there. Happy to help anyone in the community get there when you're ready.
The takeaway is simple. Getting reliable AI output is like onboarding a new employee. Invest the time in giving it context, and it pays off every single time after that.
The natural next step from skills is scheduling them to run on their own, even when you're not at your desk.
