Should we write about more topics so AI recommends us, or fewer?

A CEO showed me a content calendar with nine topics on it, built to get found by AI.

Her marketing lead had a clean argument, and it deserves full credit. Buyers start in ChatGPT now. Engines pull from a wider field than Google ever did. Cover more ground, get pulled into more answers. Nine topics, three posts a month in each one, all of it adjacent to what they sell.

The instinct is right about where buyers went. It's wrong about what happens once you get there.

That calendar runs 27 posts a month. The measured evidence says the wide version earns fewer recommendations than a narrow one would, so the spend doesn't just underperform. It works against the thing she's buying it for.

Being a source and being the recommendation are two different jobs

When an engine answers your buyer's question, two separate things can happen to you. It can pull your page in as a source and link it. Or it can name you in the sentence as one of the answers. On a visibility dashboard those look alike. To the buyer reading the answer, they're nothing alike.

A link is a footnote. A mention is a shortlist.

Semrush published a study on August 3, 2026, run with Kevin Indig's Growth Memo, covering 1,094 US categories tracked monthly in ChatGPT: 283,215 citation observations and 76,493 brand-mention observations. In categories close to what a brand actually does, brands were cited as a source in 74% of appearances and named in 44%. In categories further out, 50% cited and 25% named.

Read the second row again. Step away from your core and you keep about half your ability to be a source. You lose closer to half your ability to be named.

Your hand is probably up: more categories, more total mentions, even at a worse rate per category. That's the trade the nine-topic calendar assumes. The same study found it doesn't hold up, and the reason is the next section.

Why covering more ground is worse now than it was three years ago

Breadth used to be expensive. A competent post on a new topic cost a writer most of a week, so a company covering nine topics had clearly decided to, and that decision carried information. AI took that cost to almost nothing. Now anyone can be everywhere, and being everywhere says nothing about you.

Volume stopped being a moat the moment it stopped being hard.

The study puts a number on the floor. A brand that appeared in only one of a category's five buyer prompts was associated with a drop in mention share, and that penalty didn't disappear until the brand reached at least three of the five.

One post in a category isn't a foothold. It's a footnote in somebody else's answer.

Two things to hold about that research. It reports associations rather than causation, and the authors say so plainly. And Semrush sells the visibility tool the data comes from, with Indig as a partner on it, so read it as directional rather than settled.

Run this on your own content this week

Three tests, an hour total, no tools to buy and nobody to hire.

1. 1Write the five questions a real buyer types to find what you sell. Ask each one in ChatGPT, in a fresh chat, without your company name in the prompt. Count how many answers name you. Under three of five and you're sitting in the penalty the study measured.
2. 2Take your last twenty published pieces. Next to each one, write the category it serves. Count the distinct categories. More than two and you're funding breadth you have no path to owning.
3. 3Ask an engine your core category question, then ask the same question one step out from your core. Note whether you get linked or actually named in each. The distance between those two answers is your focus problem, measured on you rather than on a sample.

What we see across 200+ B2B companies

The pattern holds often enough that it's usually callable before the analytics come up. Companies that publish widely get cited widely and recommended narrowly. The wider the calendar, the bigger the gap between how visible they feel and how often a buyer hears their name.

That isn't a volume problem. We've published more than 350 posts.

It's a concentration problem, and we had to fix it on ourselves before saying any of this to a client.

What this looked like on our own site

Last August we tracked 50 buyer prompts daily for five weeks. Nine of the ten pages engines retrieved most from our site were lists and comparisons. Our healthtech firms list averaged 2.84 citations per conversation. One long essay got pulled into 33 conversations and quoted zero times.

The essay circled a broad idea. The list answered one category's question for one kind of buyer, and it said who each option fits.

We didn't fix that by writing more. We fixed it by deciding, on paper, which category we intend to own, and then writing inside it.

What this means for you

If you're a founder between $5M and $75M with a marketer asking for a wider calendar, you don't have a content problem. You have a decision nobody has made yet.

1. 1Name one category, in the words your buyer types, not the words your team uses internally. If the name only makes sense after a paragraph of explanation, it isn't a category yet.
2. 2Point your next twelve pieces at the five questions a buyer asks inside that one category. Five question shapes covered properly beats nine topics covered once.
3. 3Write the decision down where a machine can read it: who it's for, the problem you lead with, the point of view only you hold, and the alternatives you're deliberately not. Then re-run test one in ninety days and see whether you've reached three of five.

That written decision is what we build as a Magnetic Messaging Framework, and it matters here for a plain mechanical reason. An engine can only repeat a decision you've already made. Hand it nine half-decisions and it will cite you politely, then recommend the company that made one.

“An engine can only recommend a decision you've already made, and publishing more is how companies avoid making one.”

... Greg Rosner, founder of PitchKitchen