Leo says: “Hey, David. My number one SEO tip in 2026 is a bit more GEO and AI visibility than SEO.

Something that we realised recently is that ChatGPT loves listicles. So, what we're doing right now, with the link-building budget that we have from clients, is that we are pushing, as much as possible, listicles that will rank in Google and will be taken by ChatGPT, to increase the visibility of our clients overall.

Why do we do that, and how do we do that? First of all, we build the listicle. To build a good listicle, you generally need an OL list – an HTML order list – at the beginning of your article, a table, and a detail component after that.

The second question that we have is how to find a good place to put your listicle. What we will be doing, generally, is that we will extract all the backlinks that are interesting. We'll search for the one with the highest number of top three keywords ranking with, if possible, the lowest price. Then, we'll try to push that.

Generally, we go for mid-media that are around like 300 euros. Like that, we have a good quality price ratio and a high chance of ranking the listicle. Then, the listicle is taken by ChatGPT, and you have a good link-building budget, which is invested as well to yield some performance in ChatGPT and GEO.”

I wonder if we can turn this podcast episode into a listicle: ‘X Steps to Producing the Optimum Listicle You Could Ever Dream Of.’ Maybe we'll turn it into that. We'll see.

In terms of defining what an optimum listicle looks like nowadays, compared with what it looked like a few years ago, has it changed at all?

Obviously, we're all familiar with the ‘Top Ten of This’ or ‘Five Simple Steps to Doing This’. Is there a particular format that works best in 2026?

“Yeah, so definitely there is a format. The format has improved year after year – and I've been doing affiliate marketing for a while, so the ‘Top 10 XYZ’ was my business for a while, so I improved my way of doing listicles with that.

What’s happened now (which is really important, in my opinion) is that the first path is generally two possibilities: put in a comparison table, with three, four, or five columns with different things that the people will use to compare the different solutions, and an ordered list.

Typically, I go for ‘Top 10’ because ChatGPT loves the Top 10, Top 5, Top 7, or whatever. With an ordered list, generally, you use an order because you want ChatGPT to feel that there is an order in this list exactly, and that this is the ‘Top 10’ – not a list of ‘10 good solutions,’ but one is better than the others. After that, what happens is that you have the detailed comparison.

This top stuff generally takes like 200-300 words, and your article will generally be around 3,000 words. The question is, what are the following words? For the following words, you go for a detailed comparison of the top 10, and you do an H3 for each solution. Under each H3, you detail why this solution is the best. Then, another H3 for a second solution, where you detail why this one is the best.

You add screenshots of every solution at every step, and like that, you can fill a big article, which is really detailed, with the pros and cons of each solution – why they are the best, why they are good, why they are better than the others – and, like that, you have a really solid listicle.

The last part that you will add to this listicle is, ‘We made this listicle’, ‘We compared the websites’, or ‘We decides to put one in front of the others’. This is something that is quite important.

After that, you can have a bit of semantics that you add at the end of that. That can be a conclusion, that can be, ‘What is the price of XYZ?’, that can be more details that are sometimes needed to rank on Google, and that will be asked by classic semantic tools and everything.”

Lovely. There are lots of different elements there that I'd like to go back to in greater depth.

You talked a bit about tables and the importance of ordering lists as well, but just in terms of word count, is there any length of listicle that is optimum?

“I think, personally, that the longer the better. It sounds a little bit like an old school answer, but the reality is that longer is better (if the content is properly written). You can see that because, in highly competitive niches, generally the content is really long – and really well written as well.

People will generally say, ‘You don't have to do something with 3K words. It’s better to have good intent and a clean world count and have better content and everything,’ but if you can do a really long and well-written piece of content, you have the best of everything.

Generally, go as long as possible – as long as you are not losing quality over quantity, and the most important thing will be to have a high rate of quality. I think, overall, the minimum word count will be around 1,500. That would be the minimum I’d go for. Generally, I don't go above 4,000 or 5,000. 4,000 is really quite a long article.

I don't go above 4,000, because I'm not in niches that need that. But if you are in super highly competitive niches… For example, I worked in gambling for a while, and having 10,000-word articles is quite normal in this type of niche.”

I love the specific answers like this as well.

Do you use AI to help you produce the content?

“Yeah, I do, because cooperators cost a lot of money on one side, and it saved me a lot of time as well, because when you have somebody else to do the article, and then you wait for them to come back, etc, etc. You have a lot of latency in the process. When you're trying to impact ChatGPT, you have to be fast because there’s already a lot of search and development and everything.

I use AI. I use Claude. I use the latest model (Opus 4.7, now). What I do as well is I feed it with some semantic tools. I use French ones (because I'm French), which are tools like SERPmantics, Guru, or Thot SEO. Some English ones like Surfer are doing really well as well and are doing a good job.

Just having some good semantics – What is the intent? What are the subheadings that you're supposed to insert? – and adding all of these semantics and headings, after the big, detailed comparison that you will do in the middle of the article. That’s generally how I produce my content.”

Why do you use Claude?

“I prefer Claude over ChatGPT. I feel that it falls less into the shortcuts of producing content.

ChatGPT will just do a lot of bullet lists, generally. If you try to produce something super long, you don't want so many bullet lists. You can see that Claude will naturally go less into ‘bullet list mode’.

I feel that the content that is written by Claude is, overall, a bit better. I prefer the type of content Claude produces. To be honest, for a while, I was on ChatGPT. After that, I switched to Claude, and I went back to ChatGPT. I changed a lot. I know some people who use Gemini as well. I think these three are the best ones to produce content.

Personally, my favourite is Claude, but those three all work well, generally.”

Sounds good.

Going back to your comment about tables, why do you incorporate tables in the listicle, and do you want AI to reproduce that table within the result?

“The table in listicles, first of all, is because I want a lot of different formats in my content in general. That's why I use lists, tables, and paragraphs as soon as possible, and I vary them. So, if Google wants one type of format, I provide it.

On the other hand, definitely, ChatGPT (or AI in general) likes to have different types of content, and different formats of content as well. If you have a table on one side, then a list after, and something else after that, and one of the formats is speaking to them a bit more, and they are getting it a bit more, and they like this format a bit more, they will use it, and they will reproduce it in their answers.

It used to be the case, around one year ago (right now, it's a bit less the case. I think they're more matching different sources, mixing them, and having the overall list that they will have built themselves, and they will not take your stuff for granted, but sometimes it's still the case), if you're one of the only sources that they can have and they feel is the most reliable, they will take it, and they will use it like that.”

You also talked about using an ordered list as well, and I've seen quite a few posts start off from the highest number and work down to number 1. So, you're starting off at number 10, then discovering number 1, perhaps even after going on to a new page.

Is there any benefit to doing it either way?

“For me, there is really a big benefit to starting from the best one. The first thing is that, if you use a classic HTML ordered list (from the basic, ‘last century’ type of HTML), the first one that you put will be marked as first. Basic HTML stuff is like, the first is the first, and the first is not number 10, and you do some reverse engineering stuff.

But sometimes, people will use the number 10 to have better retention, because people will scroll through it and will want to discover the last one and be like, ‘Okay, what are the best tips?’ and everything. Personally, I think that people don't have time anymore to stay on your page forever to try to discover the last tip.

The best thing, overall, is to give them as much information at the beginning, and the more information you give them, the more they will trust it, the more they will trust you, and the more they will stay on your article.

I'm more about giving as much as possible at the beginning, and giving a quick answer at the beginning, and after being like, ‘Oh, you want more detail? I will provide more detail later on.’ That way, the people who are interested will scroll, and they will dig into the article, and they will not just skim everything to go directly to the golden nugget at the end of the article.”

Lovely. Now you also say that ChatGPT uses Google to find solutions via fan-out. What do you mean by that?

“Typically, when you do a query to ChatGPT, there are two different possibilities. The first possibility is that ChatGPT knows the answer. For example, how long it takes to cook an egg, it will tell you, ‘Okay, it takes one minute to cook an egg,’ and it won’t do any Google search to know what is happening and how long it takes, because it's sure of the answer and it doesn't need to take precaution or anything.

The second possibility is, for example, if you ask, ‘What is the best podcast for SEO in 2026?’, because it’s not sure of the answer, it doesn't want to take the risk of providing fake info, so what it will do is search on Google, because Google is a truth and a good source for information.

Indexing everything for itself would be super hard, and crawling the web would be super hard, so what it does is that it will type different queries. One query might be, for example, ‘best SEO podcast 2026’. It will type another one just to cross different sources. Perhaps it will type ‘best marketing digital SEO reviews podcast YouTube 2026.’ It types a lot of different queries into Google that are super, super long tail, which are all the fan-outs.

That's the main queries, which are split into different sub-queries. and all the sub-queries end up as different queries that it has to Google. It mixes every source that it gets at the end and, according to all of that, it ends up giving you the answers.

The interesting thing is, if the main thing that you're trying to manipulate is super hard (for example, ‘best SEO agency in the world’ will be quite a hard query to manipulate in ChatGPT), it's quite hard to go directly for these queries – to try to rank on keywords like that, and to try to manipulate keywords like that.

Generally, what you will do is you will check the fan-out, you will check the sub-queries that ChatGPT asks on Google, and you will try to influence these ones specifically, instead of going directly to the main query. It's a bit like going from long tail to short tail in SEO; it's going from fan-out queries to direct main queries on ChatGPT.”

What's the best way to conduct keyword research nowadays, if you've got this extensive fan-out process with massively long, long-tail keyword phrases?

Do traditional keyword research tools still work, or are there other places you go to for this?

“The crazy thing is that, first of all, the volume overall on ChatGPT is way lower than on Google. So, already, keyword research is a bit truncated by the fact that you have fewer keywords overall and way more context, generally.

A lot of people will give a super long search, using voice/audio, to ChatGPT, and give a lot of different details and the stories and everything. Secondly, what I do is I take the best keywords that I have in SEO, and I try to translate them into a first-person type of query for ChatGPT. For example, if ‘best SEO agency’ is your keyword, I will translate it into something like ‘I searched for the best SEO agency.’ I keep the initial intent, and I just try to translate it into a more NLP/normal way of speaking and writing.

Then what happens is that, if you have some good tools (personally, I use a Belgian tool called MentionLab for that, but Peec, Profound, and all of these AI monitoring tools will give you the fan-out of your queries). You just enter your queries into your tool – your GEO/AI visibility tool – and they will give you the fan-out that you will need to influence.

Really quickly, you will end up with a lot of different things that will have super low volume. You just end up with a lot of different sources, a lot of different queries, and the volume is way lower than in SEO. So, the way to optimize for all of that is a bit weird sometimes.

Sometimes I just go for, for example, a massive strategy of sending a lot of different pages, a lot of different media presence – super cheap ones – because I know they are super long tail, then you can rank on a lot of different things. It ends up a bit spammy sometimes, when you try to go for all the fan-out.

What I try to do is check if there are some tendencies in the fan-out. If I see a tendency, for example, where everything is including reviews, I would try to optimize for review-type queries.”

Okay, and where do topic and competition come into this as well?

How important is potential keyword volume, versus competition and relevance to your target audience?

“There are some topics where the volume will be super high on ChatGPT, and some topics where the volume will be super low on ChatGPT. It really depends on your persona and the ICP that you're targeting.

For example, here in SEO – doing marketing for other SEO, tech people, B2B, and things like that – it's the type of people who use ChatGPT a lot, so obviously, we are in a sphere where we need a lot of optimization because a lot of the research will be done on ChatGPT.

I also help some people who are selling CBD in France for people who don't search on ChatGPT so much. If you're a CBD e-commerce, the volume that you will get on ChatGPT will be quite low. I also do, for example, some content for a company in Dubai, and for them, volume is super high because that content is B2B. In Dubai, there are a lot of people who are more tech-friendly, young, and so on. So, the volume tends to be quite high.

It really depends on the ICP that you're targeting and everything. According to that, sometimes the volume and the topic are really ChatGPT-friendly, and sometimes it's not really, and it's not yet a trend that you have to hop on.”

How do you monitor success?

“Oh man, it's one of the hardest parts. First of all, if you want to do that, you have to be able to do that and to jump on it without wanting a big impact that’s directly measurable and everything, because it's quite hard to measure success. But then you have tools to measure success.

Typically, Ahrefs has a ‘brand visibility in AI’ tool, and a lot of tools like Peec and Profound, for example (that I cited earlier), will have some things that can check how much you're cited when you ask the same queries to ChatGPT again, and again, and again.

Generally, what you would do is that you will rely on these tools to know how much you're cited on a certain query on a certain topic. Then, from a pure business perspective, there are two things that you can measure quite easily: the traffic that you're getting from AI on Google Analytics (or your analytics tool). For example, Analytics and the visits and sessions coming from ChatGPT, coming from Anthropic, and coming from all these AI tools. That would be a part of it.

But generally, with these sources, they are only here when you are one of the sources and not if you're just cited as a brand – as the best brand. For example, if you're searching for ‘the best tool for link building in SEO,’ they will cite Majestic, but they will perhaps not put a link to Majestic. What happens in this situation is that people will search for the brand, and you will normally get an increase in terms of brand search, and the conversion will be higher as well on brand search, because typically the conversion from ChatGPT is super, super high.

It will be that, and the best way to track it is just to ask people, ‘Where did you find us?’ and a lot of people will tell you it was AI, an LLM, or ChatGPT.”

Leo, what's the key takeaway from the tip you shared today?

“Share more listicles. Publish a lot more listicles about your brand on the web and try to build them as deep and as high-quality as you can, to get some listicles that will rank on Google.

If you're working on something which is super high competition, try to go for longer types of queries, which will be based on the fan-out of ChatGPT.”

Leo Poitevin is Co-Founder at LinkaVista and CEO at Amstrak. Find out more over at LinkaVista.com.