Risk/avoid / recommendation spam

I wanna talk about the most effective strategy for being recommended by large language models like Chat, G P T. And Google's A I.

Mode. You've been watching my videos or following other people talking about G E O.

You might already have an idea of what I'm gonna say. It's listicles, right?

And that's half true. Listicles are incredibly effective right now.

But I'm seeing some data that makes me think it's not for the reason people think. And if I'm right, this could open up some new opportunities that are equally as impactful and less icky.

When someone goes to a large language model looking for a recommendation, the first thing the large language model is going to do is perform a search in a traditional search engine. And the reason Chat, GPT, or Google's A I.

Mode does this is because it's looking for recommendations across the internet, and that's why it loves listicles. These are lists of brands or products, and these pages typically have titles like Best X for Y or best X in Y...

ove listicles, but that's not quite true. The truth is that large language models love to search for things like best X and Y.

They search for recommendations. It just happens to be that the vast majority of pages that are optimized for these terms are listicles.

And this is because the vast majority of these listicles are automatically generated by directory websites. Once you have a directory and a way to rank different businesses or products, it's very easy to programmatically generate as many listicles as you want, subdividing by the specific product and service along with location, demographic, or use case.

And so, with hundreds of millions of listicles online, it's no wonder that we're commonly seeing large language models citing listicles. How over?

We've been tracking thousands of prompts across dozens of brands over the last six months. And after taking a closer look at that data, I'm starting to think that listicles aren't necessary at all.

As I was saying before, it can be kind of icky to put a listicle on your own website, ranking you number one...

going to have any impact on the large language model's recommendation. And you want that explicit recommendation as close to the top of the article as possible, preferably in the first sentence.

Now, that might still feel a little icky to you, but it sure beats putting a listicle on your website with you at the top. Now, to be completely honest, the brands that are having the most success are doing both the listicle and the how to choose the best article.

There's really no limit to the effectiveness of recommendation spam. Now, I should also note that if you're doing thousands of these kinds of posts every month, you are at risk of getting a penalty from Google.

It's called a scaled content abuse penalty, but it has to be pretty egregious before you're at risk. So my advice is to make a 2 by 2 grid in something like Google Sheets on the x axis.

You can list all of your products and services on The Y axis. You can list different demographics or use cases or service areas, whatever makes sense, and then start creating the articles prioritizing the most common combinations.

There are all these new tactics for being recommended by large language models. Now some of them feel a little silly or even spammy, and I hear a lot of people asking, sure, these work right now, but are they really gonna continue working?

And so I just wanna talk about why I think a lot of these tactics are going to continue working for the foreseeable future, and which ones you might want to avoid. The tactics that are especially effective right now, that are different from just traditional SEO, could collectively be summarized as recommendation spam.

You essentially want to spam recommendations for your brand all over the websites that these large language models trust. The most common form of this is probably spamming Reddit comments.

I think the most effective form is spamming listicles. This is where you create like top 10 best lists and put yourself at the top.

Incredibly effective even when it's on your own website. It can be done effectively using press releases or on any sites where users can post their own content, like YouTube, LinkedIn articles, or Medium...

this take is it's from other SEOs. SEOs that would prefer to just keep doing what they've always been doing and who really hate this idea that they have to change their methods.

It's a really good excuse to not have to learn something new. And honestly, I've fallen into this trap myself in the past.

I've been doing SEO for about 16 years, and maybe about 10 years ago I got it into my head that Google's getting a lot smarter and therefore these spammy tactics aren't gonna keep working. One example of a tactic I stopped using temporarily was service area pages.

If you're, say, a plumber in Houston, it doesn't necessarily make sense that you would have a page about your plumbing in every city around Houston and every neighborhood within Houston. You can just have one page that says you serve the greater Houston area.

However, if you want to rank in those neighborhoods and cities, this is incredibly effective. And 10 years ago I thought, you know what?

Google's gonna figure this out. Google doesn't want this to be effective.

It's super spammy...

really all they can do. Now, you might think well, eventually these models are gonna be smart enough to realize that the recommendation came from the brand's website or was probably placed by the brand themselves, but the truth is that would be way more expensive than just a simple search tool.

It would have to be causing a serious issue for the large search engines to be forced into doing that. It's much more likely they'll handle it the same way Google has.

They'll penalize the biggest offenders with manual actions. Instead of spending money having the models automatically vet every single web page on the internet, they can just manually identify the biggest troublemakers, brands that are spamming thousands of recommendations every week.

So my advice: don't try to automate this. Don't do this at scale.

Even if they never give out manual penalties for this, do you really want a bunch of spam about your brand out there on the Internet? However, I think it's equally as dangerous to ignore this tactic...