Comparison Pages Win B2B AI Citations

We've got some great stories today on September 30th, episode 52, and, uh, the exclusive theme here for today is gonna be that the category is moving from can you track AI to can you measure the right market and influence the right sources? So let's get into the top story, a very important analysis here coming from Peak.

Peak challenges the more repeat runs equals better measurement assumption.

And by the way, that is definitely our assumption at GEOForge, which is that more runs leads to higher confidence and a smaller margin of error in AI visibility reporting.

So let's see what this is all about.

Peak analyzed roughly, uh, 120 million chats collected between February and August 2026 across major AI assistants.

In its controlled comparison, both scenarios used 1,500 monthly runs, 50 prompts run once daily, produced a plus or minus 6.1 percentage points of monthly portfolio uncertainty versus 13.2 points for 10 prompt runs five times daily.

So let's, let's take another look at that.

That's a big difference, more than two times difference in confidence here.

Peak concludes that broader prompt coverage improves estimates of portfolio-level vis- visibility more than repeatedly sampling a narrow prompt set.

It separately argues that repeated observations over time remain important for detecting actual change.

Um, this doesn't necessarily invalidate the repeated run testing at the individual prompt level argument, but it separates two different statistical questions, which is, one, estimating the behavior of one stochastic prompt versus estimating visibility across the broader market of possible buyer questions.

And these are two different things.

And what I'd like to do now is, uh, pull up the recommendations from Peak, uh, when they announced this.

There are five steps for putting this into practice.

Number one is grouping the prompts, branded versus non-branded, and, uh, we also have a tagging system in GEOForge within SignalForge that allows you to tag in, in any way that you want.

And sometimes it's, it's also helpful to tag your prompts by funnel stage, top of funnel, middle of funnel, bottom of funnel.

But definitely you wanna separate branded versus non-branded because your visibility for branded prompts will almost always be 100%, naturally.

Second recommendation is don't look at single prompts for visibility, and that is the real crux of this argument.

If you're only looking at one prompt or a small set of prompts, it's just a weaker sample of the whole universe of potential prompts.

Uh, it's just a smaller sample, so you have lower surface area.

And Peak is recommending that if you want to get more accurate visibility across your portfolio, then you should increase the size of the portfolio, increase the number of prompts that you're tracking, rather than doing more repeated runs on a relatively smaller set of prompts.

I generally do agree with that.

Uh, number three is to judge movement on the portfolio across time.

Because what we're doing here es- essentially is probabilistic sampling, it is important not to look at any snapshot of one particular day or one particular week of visibility, but keep your measurement, uh, keep your measurem- measurement approach consistent over time, and then look at the trend over time.

If you're trending up or trending down, uh, then directionally you can see where your AI visibility is trending, and you don't need to focus as much, or you shouldn't focus as much on the actual percentages that are reported because they always will imply some degree of error, some margin of error, and, um, this is not deterministic in, in kind of the way that SEO rank tracking used to be.