Publish Less, Win More

Why the most comfortable lever in marketing is also one of the weakest.

Chances are you’ve sat in this meeting. Visibility in AI search starts slipping. Somebody shares their screen. The numbers are not good. And the room lands, every single time, on the same decision: we need to publish more.

I understand the instinct, because I had it too. When I first started taking AI search seriously, my first move was to point the content machine at the problem and pull the lever harder. More pages. More answers. More coverage. It felt like progress, mostly because I could measure the output, and output is the one part of this whole thing you completely control.

It mostly did not work. The wins, when they showed up, came from somewhere else entirely.

Here is the number that I can’t stop referencing. Ahrefs ran a study this year on what correlates with getting cited by AI engines. The relationship between how many pages a site has and how often it gets cited landed at 0.194. For anyone who has not stared at a correlation coefficient lately, that is a polite way of saying there is almost no relationship at all. You can triple your page count and the models will shrug.

So why does everyone keep reaching for it anyway?

Because publishing is the one move you can make alone, today, with nobody’s sign-off. You can ship a post by Friday. You cannot ship a Reddit community deciding you are worth trusting by Friday. One of those feels like work. The other one gets you cited. We tend to pick the one that feels like work and then call it a strategy.

Most of what gets sold as AEO right now is a content marketing program with an AI label quietly stapled to it. Same calendar, same briefs, new dashboard. The teams pulling ahead are doing something less glamorous and a lot more useful. They have stopped trying to out-publish the problem and started building systems that watch where answers are forming and put themselves in front of them.

Three of those systems are doing most of the work for my clients right now. Steal all three.

(Source: Ahrefs)

1. Hunt for the gaps, not the wins

The first one is almost embarrassingly simple. I take a fixed list of the questions our buyers ask, run them across ChatGPT, Perplexity, Google’s AI Mode, and AI Overviews on a schedule, and log who gets cited. Brand, URL, type of source.

Everyone wants to look at who showed up. The part that matters is who did not. Specifically, us, on questions where we already rank perfectly fine in regular Google.

When that happens, and it happens constantly, it is almost never because we did not publish enough. Something else is off. The model trusts a different source. A competitor answered the question more cleanly. The citation is coming from a review site, or a Reddit thread, or a comparison page we have no presence on whatsoever. That gap is the to-do list. Not “go write more,” but “go figure out why a G2 page is beating our own product page inside Perplexity.” That second question is where the work gets interesting.

2. Catch the decay before the traffic does

The second system exists because of a timing problem.

Most teams find out they have lost AI visibility the way you find out about a roof leak: once the ceiling is already stained. By then you are not preventing anything, you are explaining it in a QBR.

So we stop waiting. We watch the pages and topics that matter for early signs of slippage. Fewer citations than last month. Showing up lower in the answer. The evidence on the page going stale. A competitor quietly taking a spot we used to own. The second one of those trips, we do not open a blank doc. We go back to the winning page and make it better. Answer the question in the first two sentences. Add a data point nobody else has. Tighten the parts of the page the model uses to figure out what it is even about.

The question moves from “what should we publish next” to “which of our winners is starting to slide.” That is a much shorter list, and it pays back far more than anything net-new.

3. Get to the question before anyone answers it well

If you made me delete the other two and keep one, this is the one I keep.

We are constantly pulling fresh buyer questions out of the places people actually ask them. Reddit. Quora. YouTube comments. Review sites. The private communities. And the most underrated source on the list by a mile, sales calls, where your prospects describe exactly what confuses them, for free, in real time.

Then we rank them. Not by search volume, which is the old reflex. By how likely the question is to get cited. Is it coming up over and over? Are the AI engines answering it badly today? Are the sources they lean on weak enough that we could realistically take the spot? When the answer is yes, we do not default to a blog post and a prayer. We go put a real answer in the places the models are already reading: the community threads, the videos, the third-party write-ups, the comparison pages, and our own site too.

Reddit does not care about your content calendar. It cares whether you said something true and useful when someone asked.

The actual job

None of this is more work than running a content calendar. It is different work. Less writing, more listening. Less guessing about what to make, more watching where the answers are forming and showing up there before your competitor does.

That is the whole shift. The brands winning AI search this year did not get there by making the most content. A few of them are making noticeably less of it than they used to. They got there by being the answer in the rooms where the question gets asked.

So publish less. Pay attention more.

Then go steal all three.