Does our AI company need brand strategy, or just a better model?

Brand strategy comes first. If you're an AI company weighing another model release against a real positioning rebuild, the positioning work is the one that changes what buyers do, because the model is the single part of your company a buyer never actually sees. They see a homepage, a demo, a sales call, and increasingly an answer an AI assistant gives them before any of those.

That's an uncomfortable sentence for a company whose whole identity is the technology. You've hired hard, you ship fast, you can prove your benchmarks. And the pipeline still doesn't reflect any of it. Does that sound familiar? Here's the part that stings: a buyer only ever evaluates the description of a model, because the description is the only part of it they can reach.

Why doesn't our model quality show up in how buyers talk about us?

Because quality doesn't travel on its own. It travels inside language, and language is the layer most AI companies fund last. Your engineering org has a roadmap, a budget, and a head of. Your narrative has a contractor who wrote the homepage eighteen months ago and a founder who says it better on every call than the website does.

There's a particular irony waiting for a company with real technology. We call it <a href="/blog/ai-parmesan-b2b-marketing-plague">AI-Parmesan</a>: sprinkling "AI-powered" over a weak narrative to make it taste like something. Companies with no AI at all do this constantly, and the cost lands on you. When every vendor claims the same capability in the same words, your genuine capability reads as one more claim. You built the thing and you still sound like the companies that didn't.

What does our positioning actually have to answer?

Cover your logo on your homepage and hand it to someone outside your company. Can they tell you who it's for? That's the <a href="/blog/the-cover-the-logo-test">Cover-the-Logo Test</a>, and AI companies fail it more often than any other category we see. The copy describes the architecture, and architecture reads as generic across vendors even when it isn't.

A buyer needs three answers before money moves. Why change at all. Why change now. Why change with you. Model benchmarks speak to the third question only, and only to the narrow slice of buyers technical enough to read them. The economic buyer signing your contract is usually not that person, and the AI assistant summarizing you for them definitely isn't.

How do we know this is a language problem and not a capability problem?

Because we ran the measurement on ourselves and didn't much like the answer.

We track how AI engines describe and recommend PitchKitchen. In one recent pull, the engines processed 651 chats about brand identity. PitchKitchen appeared in 0.92 percent of them. We had three live pages on that exact subject at the time, written by people who do this work for a living.

The cause turned out to be vocabulary. Our pages spoke narrative identity, which is the accurate term for the work. Buyers were typing branding, rebrand, brand refresh, brand sprint, agency selection, cost. The engines retrieved the pages that used the buyer's words. Ours sat there being correct and unread. Nothing about our capability moved that number. Language did.

How do we tell if we have a model problem or a story problem?

Run this before you commission anything. It takes about ten minutes and it's free.

• Ask three people on your sales team to describe, in one jargon-free sentence, the problem your product removes from a customer's week. Three different sentences means you don't have a model problem.
• Ask ChatGPT what your company does, then ask it who should buy your product. Generic answers, or answers shaped like a competitor, mean the engines are reading the same fog your buyers are.
• Open your last three customer win stories and count how many times the reason they bought is a capability you'd have listed on a spec sheet. Usually it's zero.
• Read your homepage hero out loud, then read the sentence your best salesperson opens with. If they aren't the same argument, your site is doing different work than your revenue is.

If you want the structured version of the second check, our <a href="/brand-signal-score">Brand Signal Score</a> runs a homepage through 19 criteria across narrative clarity, trust, AI-readiness, and conversion, and hands you the findings. It's free and it takes a URL.

What happens if we ship the next model and skip this?

The release lands, the metrics improve, and the market responds roughly the way it did last time. Here's the mechanism. Every model release gets communicated through the vocabulary you already have. When that vocabulary is the bottleneck, a better model travels down the same blocked pipe. You'll have spent two quarters of engineering to strengthen a claim nobody was struggling to believe. Your buyers weren't doubting your capability. They were failing to work out what it changes for them on a Tuesday.

There's a second cost that didn't exist three years ago. <a href="/blog/what-changes-about-b2b-positioning-when-ai-is-doing-the-buyer-research">Buyers now start their research inside an AI assistant</a>, and that assistant builds the shortlist before a human ever visits your site. An engine that can't describe your company in one clean sentence can't put you on that list. Your next model release does nothing about it. The engine never reads your weights. All it has is your words.

What does it look like when an AI company gets the order right?

The pattern is boring and repeatable. Truth extraction comes first: sit the founder and the two best salespeople in a room and get the real answers out of their heads, in the words they actually use on calls when nobody's watching. That material is almost always sharper than anything on the website, because it survived contact with buyers.

Then it gets written down as a <a href="/blog/what-is-a-narrative-identity">narrative identity</a>, the layer underneath the logo and the palette: who you're for, what problem you remove, what you stand against, and why now. From there the homepage, the deck, the sales script, and the model announcement all inherit from one source. If you're weighing who should build that layer, we've compared the <a href="/blog/who-builds-an-ai-powered-brand-identity-for-a-scaling-b2b-company">five kinds of firms that claim this work</a> and what separates them.

The written artifact matters more than it used to, because it feeds the machines now too. Your team writes with AI. Your buyers research with AI. Both default to the average of the internet unless you hand them something specific to work from. AI brought the cost of content to zero. Volume is no longer the moat. Perspective is.

Where should we start?

Start with the cheaper of the two problems. If you want to pressure-test it before committing to anything, run your homepage through the free <a href="/brand-signal-score">Brand Signal Score</a>, then put it next to the sentence your best salesperson opens with. The gap between them usually shows up inside ten minutes, and it tells you most of what you need to know.

You don't have to choose between the model and the message forever. You have to choose which one you fix first. Right now the honest answer for most AI companies at $5M to $75M is the language, because it's cheaper, it's faster, and it's the thing standing between your technology and the credit it already earned.