The AI Test That Shows Whether Your Vision Actually Explains Your Roadmap

Your vision is not in the room

I have never heard a company announce out loud that its roadmap contradicts its vision. It does not happen in meetings. It happens in the backlog, where nobody is reading both documents at the same time.

The vision says you exist to make one specific thing possible for one specific person. The roadmap says you ship a dashboard in Q3. Both statements are true. Neither one explains the other.

You will not catch it by reading them back to back. You wrote them both, so you know what you meant.

You need something with no reason to be polite. That is what AI is for here. Not to write your strategy. To interrogate it.

The framework: WHY, HOW, WHAT

Every piece of content you put into the world does one of three jobs. In order: Know, Like, Trust. In the framework's language: WHY, HOW, WHAT.

WHY, HOW, WHAT: the organic content engine that moves a stranger through Know, then Like, then Trust. WHY earns Know. HOW earns Like. WHAT earns Trust. Most of what you publish is WHY.

Here is each one doing its actual work.

WHY is awareness (Know, 70%). It carries the values and the philosophical side of the brand. One pain point per piece, no more. You tie that pain to one of three basic human motivations: relationships, status, or money. Then you explain why the pain matters and why it should be solved at all. The call to action is deliberately low-commitment. You are not selling. You are asking the reader to raise their hand, to answer one short question, to identify themselves as someone who has this problem. This is how a stranger comes to know you exist and to know what you stand for.

HOW is the Like (15%). This is where people see how you work. Your method, your thinking, the way you approach the problem. WHY tells them what you believe. HOW lets them watch you believe it. This is the content that makes a stranger decide they like the way your mind moves, long before they have any reason to trust what you sell. Most companies skip straight past it, because "how we do the work" feels less urgent than "what we have for sale."

WHAT is the Trust (15%). This is your proof. The product, the offer, the result, the promise you can stand behind. It is where the roadmap lives. It is also the easiest of the three to talk about, which is exactly the trap.

The split by volume: 70% WHY, 15% HOW, 15% WHAT. Most of what you publish should be awareness. You earn the right to ask.

Read that split again, then look at your last month of publishing.

Most companies run it backwards. They publish WHAT almost exclusively, sprinkle in some HOW, and call the occasional values post their WHY. The exact inverse of the framework. They describe the offer to people who have no reason yet to know them, like them, or trust them.

Your roadmap is WHAT. The product, the features, the quarter-by-quarter plan. By the framework's own math, it is the smallest slice of your voice. It is also the easiest slice to talk about, which is why it is the one you keep talking about.

Which means the roadmap is exactly where a vision goes missing without anyone noticing.

The test

The test is not an analogy to the framework. It is the framework applied. Your vision is your WHY. Your roadmap is your WHAT. The test asks one question: can your WHAT be explained by your WHY?

Two passes. The first one is the whole diagnosis.

Pass one: the explanation pass. Put your vision and your roadmap into the same context. The vision is your WHY: who it is for, what pain it names, why it matters. The roadmap is your WHAT: every item you are actually building this year.

Then ask the AI to explain each roadmap item in terms of the vision. One sentence each. No reusing any word from the item itself.

Read the output twice.

Once for the answers. Once for where it had to reach.

Items that are wired will come back clean. Items that are not will come back in alignment language. "Supports our commitment to." "Aligns with our goal of." "Helps us deliver value." That is not the model failing. That is the model doing exactly what you asked with the material you gave it. Where it has nothing, it fills.

Now the cost, and sit with it. That hedge is information, and it may tell you an item you already announced has no line back to why the company exists. You will feel the pull to explain it away. Do not. When it happens, you have two honest moves. Kill the item, in front of the team that built it. Or admit the vision was decoration and write one that isn't. There is a third move, which is to decide the test was flawed and go back to the backlog. That one is the most popular and the most expensive.

There is a smaller fear inside this same pass: that the model hands you a bridge that sounds good and you believe it. That is a real risk, because the model is agreeable. So do not read the test for the answers. Read it for the places it had to reach. A clean explanation is cheap. A hedge is the signal.

And there is a larger one under both. For a founder, the vision is not a document. It is the reason to get up. Putting it in front of something that will not flatter you feels like being tested, personally. It is. That is the point.

There is a scheduling version of this too. We will run it after the launch. The launch is the test.

Pass two: the sameness pass. Take the same roadmap and ask the AI to write the WHY for it as if it belonged to your closest competitor. If the output fits them as well as it fits you, the vision is not doing work. It is decoration.

The cost here is starker than pass one. A fit-either-way answer means you have been running on a vision that could belong to anyone, which means it has been shaping nothing. Not the roadmap. Not the content. Not why anyone chooses you. You can sit with that, or you can fix it.

The work was never missing

A pharmacy owner of seventeen years opening a wellness clinic sat in one of my live AI workshops. She had already done the hard part. She had written her business brain document, the file that holds who she is, who she serves, and where she is going.

She came because the AI kept producing work that did not sound like her business. She had decided the problem was her prompting. Her words: "I had my brain, but I'm having trouble making sure that all of my projects know my brain."

The work was not missing. The wiring was. Her projects had never been pointed at the brain file, so every single project started from nothing. We attached the folder and named the file in the instructions. In the last hour of that workshop she produced a vetting guide, a legal-coach skill, and a launch checklist.

Notice the shape of it. The expected ending is that she needed better prompts. The actual ending is that she stopped starting from zero and got three finished assets inside an hour.

But there is another ending, and it was the more likely one. If we had not caught it, she would have concluded the brain document was wasted effort. She would have gone back to prompting from scratch. And she would have blamed the model for output that ignored a business it was never shown.

Her fix was not a better prompt. It was a check. Before you rewrite the prompt, confirm the project is actually pointed at the brain folder and that the instructions name the brain file explicitly.

Run the same check on your roadmap. Before you rewrite the roadmap, confirm the vision was actually in the room when it was built. If your projects never named the vision, every item in them started from nothing.

And it will sound like it.

AI does not reveal you. It reveals your inputs.

We send every applicant for a role the same interview questions to answer on video before any live conversation, so the team can hear the person before meeting them.

The videos came back nearly identical to each other. The candidates had fed the questions to AI first. One written answer still had the AI's own closing line pasted at the bottom, offering to craft more follow-up answers.

The easy read is that we caught people cheating. The real read is worse and far more useful: nobody spots it, and the whole screening stage quietly compares one AI's answers to another AI's answers while believing it is reading people.

Same input, same output. That is not a flaw in the candidates. It is a law. Hand the model a job description and it hands the job description back in a nicer font. It cannot tell you about a person it was never shown.

Our fix was a timer. The question is seen for the first time when it must be answered.

Your fix runs the other direction. You want the AI to have more of you, not less. But the principle is identical. The output is only ever as specific as the input. If your roadmap went into the model without your vision attached, what came back was your industry's roadmap.

Which is exactly what your customers are already getting from everyone else.

A claim in a category is not arguable. Put it on a scale.

There is a move I use when a company tells me they are AI-forward. I do not argue with the claim. I put it on a scale. The thing most companies call being AI-forward is the first two steps of a much longer distance, and calling it arrival is what stops them going further.

A claim in a category is a yes or a no, and yes or no ends the conversation. A claim on a scale is arguable. Arguable is useful.

Do the same with "we have a vision."

In a category, that question gets a yes and the conversation is over. On a scale, the question becomes how far your roadmap actually carries it, item by item. Now you have something you can work with.

Two places to leave this alone, and what it costs to ignore that

I am not going to hand you a clean process and pretend it is free. Running this test in the wrong place costs you more than skipping it.

Do not run it against a vision you inherited and nobody believes. You would be testing your roadmap against a placeholder, and the test will tell you the roadmap is the problem when the problem is the sentence. Fix the sentence first. The cost of ignoring this is that you kill a working roadmap and keep a dead vision.

Do not run it on plumbing. Some roadmap items are infrastructure, compliance, or the thing that keeps the lights on. They do not need a why, and pretending they do is a different failure. The cost of ignoring this is that you force meaning onto every line in the backlog and produce the exact generic language you were trying to escape.

Monday

Before you rewrite a single prompt, a single page, or a single roadmap item, do two things.

The cheap one first. Put the split on your own publishing. 70% WHY, 15% HOW, 15% WHAT. Then count what you actually published last month. If WHAT is carrying your calendar, you have been describing the offer to people who have no reason yet to know you, like you, or trust you. Reallocate. Pick the next three weeks and publish nothing but WHY.

Now the one with a bill attached.

Run pass one on your roadmap. Vision and roadmap in the same context, one sentence per item. And when the model hedges, do not reach for a better prompt. Reach for the two honest moves. Kill the announced item, or rewrite the vision. Both cost something real. One of them is the price of a roadmap that carries anything at all.

You already suspect which items will come back hedged. Running the test does not create that problem. It makes you say it out loud, before you spend another year talking about WHAT to people who never got the WHY.