Your Website Is a Mirror: AI Can’t Recommend What It Can’t See

The problem nobody warned you about

Somebody is asking ChatGPT who they should hire.

They are describing their problem in their own words. Not a keyword. Not “roofing contractor Des Moines.” They are typing something like my roof is leaking and I don’t know who to trust.

And ChatGPT is answering them. It is naming three or four businesses. It is explaining why. It is telling them what to watch out for.

You are not in the room for that conversation.

Here is what makes it worse. You have been doing this work for twenty years. You have a process nobody else has. You have a guarantee that beats everybody in your market. You know exactly how to handle the situation that person is describing, because you have handled it four hundred times.

And none of that is helping you, because ChatGPT never learned any of it.

I have run AI visibility reviews for more than thirty businesses now. Almost every single one is a good company doing excellent work. And almost every single one has the same problem.

Your website is a mirror

I did not come up with this analogy on purpose. It came out of a conversation where somebody was frustrated with me.

I was running an AI search review for another agency, on behalf of one of their clients. The report came back and listed a set of gaps — things the site was missing, things AI could not find.

The agency owner pushed back, and fairly. Those gaps were things they had already discussed with their client. In meetings. In detail. With plans in place. Some of them the client was genuinely, actively doing.

So I said the thing that has become the center of how I explain all of this:

It can only measure what it can see. It cannot read your mind. It does not know the conversations you have had. It does not know the plans sitting in your project folder. It does not know everything you already do for your customers every single week.

It reads your website. That is all it gets.

Which means an AI score is not a grade on your business. A mediocre company with a higher score is not a better company. It means their website is more aligned with what the customer is asking, and easier for AI to read. That is the whole difference.

I want to be clear about that, because a low score can feel like an insult. It is not. It is a measurement of one specific thing: the distance between how good you are and how visible that is.

AI doesn’t leave the space empty

Now here is the part that changes the stakes, and it is the single most important idea in this post.

Under the old rules, if your website did not mention something, you simply were not found for it. Silence made you invisible. That was the whole penalty. Annoying, but passive.

AI does not work that way.

And in my experience, those assumptions lean skeptical. They tend toward the worst case. And the worst case is rarely flattering to the business.

Let me show you two real examples.

The free valuation

One of my clients offers a free business valuation as the first step with any new prospect. It is genuinely free and genuinely honest.

I asked AI about it. AI warned the prospect to be careful — reasoning that a free valuation is probably inflated to flatter the owner and win the engagement.

The truth is almost the opposite. That valuation is honest, and it frequently comes in lower than the owner hoped. Some people get offended and walk away. A year or two later a fair number of them come back and say, you were right.

But the website never explained how the valuation is produced or why it is free. So AI supplied a motive. It picked the cynical one.

The roofing guarantee

Advantage Roofing has a guarantee that covers interior damage if one of their roofs ever leaks. Across thousands of roofs, occasionally one will. When it has happened, they have covered the full repair — drywall, paint, all of it.

I asked AI about that guarantee. AI assumed there was a cap on the coverage. Because in that industry, there usually is.

Nobody wrote that. Nobody claimed it. The site simply had not spelled out the terms, so AI filled the blank with the industry-standard assumption and quietly told a potential customer about a limitation that, as far as I can find, does not exist.

Think about what that means.

How AI search actually works

Before I go further, let me say something that might surprise you coming from somebody who sells AI search services.

Traditional SEO is not dead.

Traditional keyword search still carries the large majority of search volume. seoClarity, which tracks more than 300 client analytics accounts alongside clickstream data, puts LLM-based search at roughly 10 to 15 percent of all searches. Most search still runs the old way.

But that number is misleading if you stop there, because it only counts searches that happen inside an AI tool. It misses what has already happened to Google.

Google says its AI Overviews now reach about 2.5 billion people a month. Google’s conversational AI Mode passed one billion monthly users within about a year of launching — that is Google’s own figure. And Google says total search volume is at record highs, because people are asking longer, more complicated questions than they used to type.

So “traditional search” is not as traditional as it was a year ago. And the questions being asked are fuller, messier and more human — which is exactly the kind of question your website has to be able to answer.

The old way: keywords and a checklist

There are billions of web pages. Google needed a way to sort them, and the way they did it was keywords.

You write content. You identify the keyword. You optimize the page. In effect you raise your hand and say: if somebody searches this, we have something good at this link.

But a thousand other pages target that same keyword. So Google sorts them with what amounts to a checklist — the more boxes you tick, the higher you go. Then it hands the searcher a list, page after page after page. Everybody fights for page one, because nobody digs to page twenty-four.

It is also slow. In my experience you see some movement around three months, another bump around six, and you often do not see the full effect of today’s changes for ten months.

The new way: content, then trust

AI does not care about keywords the same way. It is reading content.

Your page does not need to be built around a phrase. It needs to genuinely address what the person is asking. Which means everything they are feeling, everything they are trying to solve, and everything you actually do to help them needs to be somewhere on your website.

Then AI does something the ten blue links never did.

It checks up on you.

It is one thing for you to say you do good work. It is another for AI to decide you are worth recommending. So it goes looking for proof. If you claim you solve a problem, it wants examples — case studies, reviews, specifics. Then it runs a wider search to see whether the rest of the internet agrees with your website: Google Business Profile, Yelp, the Better Business Bureau, the “best of” roundup lists.

So there are two steps, and you need both:

Your website gets you considered. No relevant content, no candidacy. You are not in the conversation at all.

Your off-site footprint decides where you land among the businesses it found.

What that looks like on a real screen

For one roofing client, I searched the way an actual homeowner would: I need my roof replaced in my city. Who would you recommend?

They were named. Fourth or fifth, out of five or six.

The AI tools show their sources, and the sources were Google Business Profile, Yelp, the Better Business Bureau, and a handful of sites I had never heard of — aggregator lists that were themselves pulling from Google and Yelp reviews.

That is the entire mechanism in a single search. The website got them considered. The footprint decided the order.

Yes, fewer people are clicking

I should name this, because you may already be seeing it in your analytics.

When the answer is written out on the page, a lot of people never click anything. Click-through rates on AI Overviews fall sharply compared to traditional results, and I have watched overall click volume drop on sites that are doing nothing wrong.

But here is the part that matters if you run a service business.

Somebody with water coming through the ceiling is going to click. Somebody with a clogged sink is going to click. Reading a summary does not fix the sink. The zero-click behavior is heaviest on informational questions — what is this, how does that work. When the question is who do I call, right now, people still go to the website.

So the goal shifts rather than disappears. On informational searches, being quoted accurately is the win — the citation is the impression, and it makes you familiar when the urgent moment arrives. On urgent searches, the click still belongs to whoever AI recommends.

Either way, the words on your page are doing the work. They just have to be right, because sometimes they are all anyone will ever see.

Why it starts with clarity

Here is where people expect me to start talking about code. Not yet.

An AI search begins with a person who has a problem. They describe it in their own words, the way they actually think about it in their head. For your website to be the answer, the words on your pages have to match the words in their head.

You cannot fake your way to those words. You have to have done the work of knowing exactly who you serve and what they are afraid of.

This is why every engagement at StoryWorks starts with a StoryBrand BrandScript. It is seven questions — eight if you count aspirational identity — and it forces you to dig into the core of what your business actually does. Ninety minutes later there is a written, detailed picture of the ideal customer. No guessing.

Without that, it is like building a house with no foundation. It just does not work.

Two patterns show up constantly on sites that skipped this:

They lead with authority and skip empathy. We have been in business since 1987. We are certified in this. We are members of that. All true, all fine — and the customer is still sitting there wondering do you actually understand my problem? Without empathy, there is no buy-in.

They never defined the plan. What are the simple steps to become a customer? A lot of companies have never thought it through. So the website confuses people instead of guiding them. And a confusing website does not get picked up.

Here is what it looks like when you do the work.

Advantage Roofing’s line is: we make roof replacement simple and stress-free.

Those words came straight out of BrandScript work. A roof replacement is overwhelming. It is one of the largest financial commitments a homeowner ever makes. There are trust issues. There are a dozen steps nobody explains. So “simple and stress-free” is not a slogan. It is a direct answer to how the customer actually feels.

And once you know that, you know what to write. You are not guessing at blog topics. You are relieving specific fears you already identified.

But clarity alone will not get you found

I used to believe clarity was enough. Get the message right, get the words on the site, done.

It is half true. If someone comes to your website and it makes sense, it works.

But if nobody knows about you, and Google does not know about you, and nobody can find the site in the first place, then the right words on the page do nothing.

It is like planning a wonderful event. Great band. Great food. Everything top quality. And then never inviting anyone.

I wrote more about where clarity ends in the honest case for and against StoryBrand.

The process, start to finish

Now that you know the problem, the solution is going to sound obvious. That is on purpose. It should sound obvious.

If AI can only see your website, and your website does not show everything you do, then the job is simply this: write down everything you actually do, compare it to your website, and publish the difference.

That is the whole idea. Here is how we run it.

1. Get the baseline score

We look at your site the way AI does and produce a number — a score across eight categories, which maps to one of four tiers.

Before you can improve anything — a company, a sports team, anything in life — you have to know where you are starting from. That is the whole job of the baseline.

Here is what those tiers mean from your customer’s side:

Invisible (0–39). In your customer’s eyes, you do not exist. You are George Bailey in It’s a Wonderful Life, in the version of town where he was never born. They see your competitors instead.

Findable (40–59). AI knows you exist. You are in the group. You might get a passing mention at the end — fifth or sixth.

Citable (60–79). AI is quoting your website directly. Think about that from the customer’s side. They described their problem, and the answer came back with your words in it. Of course they click. They already know you understand them.

Authoritative (80–100). You own something nobody else can claim.

Advantage Roofing is the clearest example I have. Citable is AI quoting their warranty. Authoritative is the Advantage Five — five promises only they make — and the guarantee standing behind it. Nobody else in their market can say it.

2. Build the client archive

This is where we learn everything we can about you. Your BrandScript. Transcripts from our calls. Questionnaires. Onboarding documents. The things you do instinctively for every customer and have never once written down.

It all goes into one file.

And that file is yours. We give it to you, because once it exists you can hand it to any AI tool you use — social posts, blog drafts, email — and it already knows exactly who you are and what you do.

3. Score the site against what we now know

Now we run the review again. Except this time we are not measuring your website against an ideal. We are measuring it against you.

This is where the gaps surface, and they fall into four buckets.

Gap 1 — You do it, but your site never says it. The most common one by far. The work is real. The page is silent.

Gap 2 — Your site doesn’t answer the questions people are actually asking. Picture a scale of one to ten, where one is the most basic beginner question and ten is “we solved your problem together.” Most companies start the conversation at a five or six. They assume everyone knows what they know.

AI needs the ones and the twos.

If you work in digital transformation, your site is written at a college level and it almost certainly does not contain a plain answer to what is digital transformation? — because writing that feels condescending to the people you talk to every day. But that basic answer is the floor everything else stands on.

Gap 3 — You never spelled out your process, so AI filled it in. The valuation. The warranty cap. Covered above, and it is the one that costs you most quietly.

Gap 4 — The technical work is missing. This is not a failure of effort. Schema.org now publishes 826 types in its vocabulary. There is markup for reviews, FAQs, local business, services. Code that tells AI what it is looking at and why it matters. There is also llms.txt, an emerging convention: a plain summary of who you are and who you help, in a form AI can read quickly. Good content with no markup around it is hard for AI to use.

4. Turn the gaps into interview questions

Every gap becomes a question. You get them ahead of time so you can prepare.

Most of them are simply about what you already do every week. The skill is in knowing which areas to get you talking about.

5. The interview

This is the step that makes everything else work, and it is the one I cannot do without you.

I could write the copy myself. It would be competent. It would not carry the weight of the owner saying it, citing real examples, telling you how they actually helped somebody. That is what E-E-A-T asks for — experience, expertise, authoritativeness and trust — and it has to come from the person who lived it.

Let me show you what I mean.

Advantage Roofing does a 19-point inspection at the end of every project. I knew that. What I did not know was where it came from.

During our one-hour interview I asked Tim Mallory about it, and he told me.

He was standing around a campfire one night with a few friends, having a beer. One of the men was new to the group, so Tim started asking him questions to get to know him. He turned out to be an airline pilot, and Tim remarked that it must be a job with a lot of responsibility — if a mistake gets made, people’s lives are at stake.

The pilot told him: here is what we do. The mechanics do the work. Then an independent inspector comes through and verifies it.

Tim thought about that and decided that is how he wanted to close out every roofing project. An independent inspector. A fixed checklist. Every single time. That checklist — plus a one-year follow-up where they run it again and fix anything that moved — is what makes his guarantee possible.

I would never have gotten that from the website.

And it is the most persuasive thing about the entire program.

6. The transcript goes back into the archive

We transcribe the interview and fold it in. The archive gets richer every time we use it.

A year from now, when there is a new service line or a new page, we do not start over. We start from where we left off.

That compounding is real. With Advantage Roofing we are on a second round right now. I ran a comparison against their strongest competitors to find what those companies talk about that we do not — and wherever the answer was we actually do that too, we got Tim on record about it. That became the next ninety days of content.

7. Build the 30/60/90 roadmap

Then the plan. First thirty days are the quick wins. Second thirty build on it. Last thirty finish the picture.

Why phased? Because it is a lot of content, and handed over in one pile it is overwhelming. Split across ninety days, it is a list you can actually work.

What you do with it is up to you. Some clients execute it themselves. Some hand it to the agency or guide they already work with — and now those writers know exactly what to build. And if you would rather we did it, we build the content too.

The first step is knowing. Then it comes down to doing.

8. Score it again

Once the content is live and AI has re-read the site, we run the same score the same way, so you can see what moved.

How fast it moves

This is where AI genuinely differs from what you are used to, and it is the part that surprises people.

With Advantage Roofing, my initial searches put them fifth or sixth.

After the content was updated, I ran the same searches again — history cleared, asking each tool to answer as though I were a first-time user, which matters more than you would think. They had moved to the number one spot in three of the four AI tools for roofing warranty questions in their market.

The one that lagged was Perplexity, because Perplexity leans hard on Yelp and the Better Business Bureau, and those profiles had not gotten much attention. So we spent a few weeks focused on earning reviews there.

About a month later, they were first across all of them.

Two honest notes. Some of what changes quickly is AI retrieving pages live rather than anything being re-indexed. And AI answers vary run to run — ask the same question twice and you will not always get the same answer. That is why the test has to be run clean, and why we re-score deliberately instead of refreshing the page and celebrating.

But the pattern holds. Traditional SEO is slowest. Local SEO moves faster — real change inside one to three months. AI moves very quickly.

And here is the one that made me a believer. Lonna Gibson at Seven & Associates got her 30/60/90 roadmap and did the first two items on the 30-day list herself.

Four days later she emailed me. They had signed one of their ideal clients.

What you can do this afternoon

You do not need to hire anybody to start. You need an hour and a willingness to look.

Read your own website as your customer. Not as the owner. As the person with the problem. Then ask two questions: Is everything you do for your clients actually represented here? Does the site answer the most basic questions people ask — the ones that feel too simple to write down?

Whatever sits in that gap is your list.

Then run the test on yourself. Open ChatGPT. Tell it to disregard any previous conversations and answer as a first-time user — otherwise your own history skews the result and you will get a flattering answer that no stranger would ever see.

Ask it the question your best customer would ask.

If you are not in the answer, ask it directly: what is preventing you from recommending my business?

It will tell you. In detail. Which sources it checked, what it found, what it could not verify. That conversation is free, and more than anything else, it is what taught me how this actually works.

One objection, answered honestly

The pushback I get is rarely about AI search. It is about AI in general.

In marketing, some people have embraced it and some treat it as the enemy. There are real reasons to hesitate. I have had a client raise the environmental cost of AI data centers and take a principled position on it. That is a fair thing to weigh, and I am not going to pretend otherwise.

What I keep coming back to is the speed. I have not seen anything in this industry grow like this.

AI search is a smaller share of search today. It is only going to get larger.

So the case I would make is simple: get in on the ground floor. Be one of the businesses that is already there, already trusted, already quoted. That position is easier to take now than it will be in two years.

General Eric Shinseki said it better than I can:

Take the step today to avoid irrelevance tomorrow.

Common questions