Why AI recommends one vendor and ignores the one next to it

Two companies sell almost the same thing. Same category, comparable product, similar price. A buyer opens Perplexity and asks which vendors are worth evaluating. One company gets named. The other doesn't exist as far as the model is concerned.

The one that got named is not always the bigger brand. Often it's the smaller one. So what actually decides it?

Not domain authority. Not how much content you've published. Not whether you "did SEO." The thing that decides it is what your content is built on, and almost nobody is building on the right material.

The models are starving for one specific thing

When a language model answers a buyer's question, it's pulling from whatever it can find that's relevant and credible. Here's the problem with most B2B content: it was assembled from the same public web everyone else read. The vendor researched the category online, summarized what the category already says about itself, and published it. Competent, clean, and completely interchangeable.

A model has no reason to surface that. It says the same thing fourteen other pages say. There's no unique signal in it, nothing the model couldn't get somewhere else, so it gets averaged into a generic summary and the brand behind it disappears.

The content that gets cited carries something the model genuinely cannot find anywhere else: the actual language your customers use. The way a buyer describes the problem in their own words on a sales call. The objection they raise before they'll commit. The thing they're really trying to get done that they'd never type into a search bar.

That language is rare by definition. It lives in your calls, your demos, your support tickets, your win-loss reviews. It's not on the public web because you're the only one who has it. When you put it into your content, you hand the model something with no substitute. That's what earns the citation.

I call this layer Buyer IP. Most of it is sitting unused inside companies that already have it.

What this looked like at Pricefx

I was CMO at Pricefx, a pricing software company. When I started, organic was responsible for under 5% of qualified pipeline. The content existed. It just sounded like everyone else's pricing-software content, so it did nothing.

We stopped writing about pricing software the way the category writes about pricing software. We went into the sales calls and the customer conversations and pulled out the exact questions buyers were actually asking, including the uncomfortable ones most vendors avoid answering in public. Then we built content around those, in the buyer's language, answering the real question instead of dancing around it.

Organic pipeline went from under 5% to over 50% of qualified deals. The company scaled from roughly $20M to about $70M in ARR over that stretch. Organic stopped being a nice-to-have and became the engine.

This was before ChatGPT made the dynamic this stark. The mechanism was already true. Content grounded in real buyer language wins, content grounded in the public web gets ignored. AI just took that rule and made it the whole game, because now a model is doing the first round of vendor selection before a human ever reaches your site.

Why your team keeps producing the ignorable kind & what AI recommends

It's almost never a talent problem. Your writers are fine. The issue is where the raw material comes from.

Most content teams are briefed to "write about" a topic. So they go research the topic, which means they read the same public sources the models already ingested, and they produce a well-organized version of the consensus. By construction, that can't be unique signal. It's a remix of the average.

The expertise that would make it unique is in the building. The founder knows precisely why buyers choose you over the obvious alternative. Your reps hear the real objections every day. Your product people understand the problem at a depth no competitor's blog captures. None of it reliably makes it into the content, because the system for extracting it doesn't exist. So the team defaults to researching the web, and you ship more of the stuff models skip.

Closing that gap isn't about publishing more, it's about changing the input. Pull from your own calls and customers instead of the public web, and the same writers start producing content the models have a reason to cite.

How to tell which kind you're publishing

Open your last five published pieces and read them with one question in mind: could a competitor have published this almost word for word?

If yes, it's built on the public web, and it's getting summarized into oblivion. If a piece contains a question only your customers ask, phrased the way they actually phrase it, with an honest answer most of your category won't give, that's the kind that gets cited, and what AI recommends.

The fix starts in conversations you're already having. The next sales call, write down the exact words the buyer uses to describe their problem. Not your category's words for it. Theirs. That sentence is worth more than a month of topic research, and it's the raw material almost nobody is using.