Are you helping AI find you?

Over the last couple of weeks, I reviewed 20 B2B websites for members of Exit Five.

The companies were completely different. Healthcare software, enterprise AI, video production, financial services, event marketing, access control and more.

But the questions in their submissions sounded familiar:

Does our new story actually land?

Which product should lead?

Do buyers understand why we’re different?

Are we trying to say too much?

Why do people seem to understand the offer, but still hesitate to act?

While 20 submissions is very far from statistically significant, when I put the original intake notes side by side, I see a clear pattern.

Most of these companies weren’t dealing with one bad headline or crappy copy in general. Their business had changed, expanded or become more sophisticated, but the story around it hadn’t settled yet.

6 messaging problems that kept showing up (and that plague most B2B companies)

1. The intended message didn’t reliably come through

The team knew what the company did and why it mattered, but would someone arriving without all that internal context reach the same conclusion?

Would they understand the category in the first ten seconds? Would the value proposition come through clearly? Would the business look as strategic as the work actually was, or would it resemble another generic provider?

This is one of the hardest things to judge from inside a company because you can’t unknow what you already know. You automatically fill in gaps that a new buyer has to cross alone.

2. The business changed faster than its story

Some companies had recently rewritten their messaging. Others were overhauling the website, expanding into a new market, changing their positioning or moving from a behind-the-scenes partner into a visible brand.

The old story wasn’t necessarily wrong. It was just no longer big enough—or specific enough—for where the business was now.

That creates a messy transition where old category language, new positioning and future vision can all appear at once. Each sentence may be defensible, but together they don’t always sound like one company.

3. Too many products and promises were competing for priority

Should two products receive equal attention? Should the homepage lead with the platform or the most urgent buyer problem? How much of the long-term vision belongs beside the immediate value?

When everything feels important internally, companies tend to give everything equal weight externally. The buyer then has to decide what matters, which offer fits and where to go next.

The information may all be there. The hierarchy is what’s missing.

4. Differentiation lived inside the team

Several companies had a clear reason they weren’t simply another project-management tool, data provider, consultant, design studio or production company.

But that distinction often depended on context the team carried in meetings, sales calls and internal documents. The public copy named capabilities without making the underlying point of view obvious.

A buyer could recognize the category while still wondering why this version of it deserved attention.

5. The page created understanding, but not enough confidence to act

Every submission had a commercial next step: book a demo, start a trial, contact sales, apply, run a pilot or buy a package.

But understanding the offer doesn’t automatically make that next step feel safe. Buyers still wanted to know what would happen after the click, how much commitment the action required, who would take part, what they needed to prepare and what they would receive.

In the wider teardown analysis, 18 of the 20 websites had at least one CTA that didn’t match its destination, left the next exchange unclear or did both.

The route technically worked, but not the decision around it.

6. The buyer’s real pain was buried under the product description

Across very different offers, the buyers were usually trying to escape some form of complexity, risk or operational burden.

They wanted to replace fragmented tools, stop juggling vendors, make a difficult decision with more confidence, reduce manual work, prove something would survive scrutiny or scale without exhausting the team.

Many of the common B2B pains.

Yet many pages spent more time explaining what the product contained than helping the buyer recognize the situation that made those features valuable.

That is the difference between describing your product and helping someone buy it.

What these companies were really trying to achieve

Once I looked past the individual copy questions, the motivations were remarkably consistent.

They wanted to:

• see the company through a buyer’s eyes;
• check whether the intended message worked live on the page;
• decide what should lead and what should support it;
• make a changed business easier to explain;
• separate the company from superficially similar alternatives;
• give buyers enough confidence to take the next step;
• create a story the wider team could reuse consistently.

In other words, they didn’t only want better copy. They wanted confidence that the company communicated the same message wherever a buyer encountered it.

Turns out, those same contradictions that make a new buyer hesitate also make the company harder for an AI system to describe. Old and new positioning carry equal weight. Several products appear without a clear priority. Generic category language obscures the difference. Claims and proof point in slightly different directions.

The system can’t see the internal debate or know which messaging version the team now considers current. It can only work with the story those public sources create together.

AI can only work with the story it can find

When a grounded AI system goes looking for your company, it may encounter your homepage, product pages, company profiles, directories, press coverage, documentation and comparison pages.

If those sources contain three generations of positioning, give every product equal weight, use vague category language or make claims that the nearby proof doesn’t support, the system has to assemble the meaning itself.

Sometimes it will get the story right, sometimes it will repeat the old one, and sometimes it will flatten the company into the nearest familiar category because that is the most consistent explanation available.

AI just exposes the fact that the company’s own sources never aligned.

This is why I don’t think improving how AI describes your company should begin with chasing individual answers. You can correct one result today and get a different version tomorrow. You need to fix the story upstream, then keep checking how it spreads.

The correction loop

For me, the correction loop starts before you monitor anything:

1. Approve the story. Define the buyer, problem, primary promise, meaningful difference and next decision (aka your messaging framework). A monitoring agent shouldn’t invent these for you.
2. Inspect the sources. Find where your website, profiles, directories, proof and supporting pages contradict, dilute or omit that story.
3. Correct upstream. Fix the authoritative sources rather than trying to patch one AI answer at a time.
4. Test retrieval. Ask the same branded, category, comparison and buyer-scenario questions across grounded AI systems.
5. Trace the answer. Inspect which sources support the inaccurate, generic or conflicting description.
6. Repeat when the business changes. New products, markets, campaigns and positioning decisions can reopen the gap.

You can automate parts of this, including monitoring answers, comparing them with an approved story, tracing citations and identifying likely source conflicts. But that doesn’t remove your judgment. Someone still needs to decide what the company should stand for, which evidence to use throughout assets and channels, and whether a correction makes the story more accurate and aligned rather than merely more flattering.

That’s also what I learned from the teardown work. Before rewriting another isolated headline, follow the whole buying journey. Check whether the promise, proof, important answers and next step still form one argument.

I wrote a deeper breakdown of that research in The UX of copy: your buyer should not have to assemble your argument.

And if you want me to check on one important buying journey for your company, I’ve turned the same process into a focused Message-Market Fit Teardown.

I review up to three connected public pages, diagnose where the story loses clarity or confidence, and challenge my reading with 18 structured synthetic buyer pressure tests.

You get the main message-market fit diagnosis, an annotated journey review, a claim–proof–action map, one illustrative rewrite or restructuring direction and a prioritized plan for what to fix first.

Before you automate how AI describes your company, make sure a buyer can follow the same story from promise to proof to action.

​Get your Message-Market Fit Teardown​

Discovery

The UX of copy

The hero gets blamed for almost every messaging problem. After reviewing the 20 Exit Five websites, I found the bigger problems further down the journey: claims separated from the proof that should support them, important answers arriving too late, and CTAs that asked for commitment without explaining what happened next.

I pulled the full findings, method and five practical fixes into The UX of copy: your buyer should not have to assemble your argument. It’s the deeper companion to today’s piece and a useful way to audit one important website path before rewriting anything.

LLMs reward expertise

Sean Goedecke makes a strong argument that the most important prompting skill is expertise in the domain you’re prompting for. His example is mathematician Terence Tao using ChatGPT: the useful difference isn’t a bag of clever prompt tricks. It’s Tao knowing which part of an answer matters, where something looks strange and which direction is worth testing next.

That matches my experience with messaging work. Expertise helps you get to a better answer faster, but more importantly it helps you recognize when a fluent answer is wrong, shallow or solving the wrong problem. Read “LLMs reward expertise.”​

Research systems, not slop opinions

This week I shared how John Horton and the Expected Parrot team made research surveys version-controlled. I use Expected Parrot for the teardowns and in my work. Every question, logic change and alternative version has a visible history, so marketers can compare approaches without losing the original research or rationale.

The wider point is that an agent is only as useful as the system you give it. Give it a homepage and ask for improvements and you’ll get plausible opinions. Give it structured buyer evidence, controlled methods, versioned inputs and clear evaluation criteria, and it can help you investigate what your ICPs notice, trust and question before a human makes the final call. Read the LinkedIn post.​

Resonance