From user experience to LLM experience. Here’s why information gain will truly make the difference
If you work in SEO, you already know: something has changed. This isn’t just another algorithm update, not one of those turbulences that blow over in a few weeks. This time the ground has genuinely shifted.
Users are still searching, but the answers increasingly arrive without ever passing through your site. No clicks, no traffic, no conversions. Just a generated, synthesized answer — and your brand might not even be in it.
There is good news, though. And it’s worth hearing: those who have always done serious work, building real authority and a consistent presence across the web, now find themselves at an advantage. Not by luck, but because the fundamentals remain fundamental (pardon the wordplay!), even when everything around them changes.
Welcome back to SEO Confidential, the series where we meet the most authoritative and internationally recognized SEO professionals to reflect together on the great upheavals in search, in the era of language models and answer engines.
Today we have the pleasure of talking with Peter Rota, an SEO strategist with over 15 years of hands-on experience. Peter has built his career across affiliate SEO, international consulting, and direct testing on hundreds of real projects. Today he is one of the most-followed voices in the global community on AI Search, GEO, and advanced technical optimization.
Do you work in SEO, or do you have a brand to grow online? Then this interview is for you.
Peter pulls no punches and doesn’t dance around things: you’ll find concrete, practice-oriented answers, sharp points of view, and a few thought-provoking challenges. Enjoy the read.
“When an AI agent lands on your site, can it get the job done easily?”
Over the years, how has your day-to-day SEO work concretely changed compared to when you started? Which activities that were central a few years ago matter much less today, and what do you really focus on now to get results?
When I started, my SEO work was much more focused on backlinks. I created social bookmarks, built local citations by hand, used software like GSA, and bet everything on the keywords with the highest search volumes. A few years ago, instead, the core focus had shifted heavily toward affiliate work and content creation.
Today the work has changed again: I do far more reporting — meaning data analysis and interpretation — and I focus on clearly communicating the most relevant KPIs such as conversions, qualified traffic, and revenue. There is definitely a very strong emphasis today on using AI to speed up processes, with much less attention on short-term tactics that, in the long run, only end up “burning” you.
AI-based models don’t work like search engines but as probabilistic, pattern-based systems: in your view, what mistakes do brands make most often when they apply traditional SEO logic to this new context?
The biggest mistake I see right now is that people treat LLMs as if they were search engines, but in this environment “rankings” are completely unpredictable and unstable. I’ve entered the same prompt just a few minutes apart and watched a citation move 10 positions.
Another common mistake is trying to recover, through LLMs, the traffic sites have lost from classic search engines. The truth, in my opinion, is that LLMs weren’t born to be search engines but to answer questions. If they don’t have a piece of information in their training data, they go look for it on the web — but even then there’s no guarantee they’ll cite you or send you traffic.
Finally, another big mistake is treating prompt tracking as if it were keyword tracking. LLMs don’t provide real data on what people are actually searching for. Many tools today use synthetic prompts or connect via API, but users’ real search data simply isn’t accessible.
The rise of AI-generated content risks creating a self-referential, unreliable ecosystem: how real is the risk today that models will “feed on themselves,” degrading the quality of information online?
It’s something I think about often: machines feeding machines. I believe talent and overall quality will degrade. I think the risk is extremely high, and the real problem is that many LLMs are getting so good that sometimes you can’t tell whether a text was written by a human or by artificial intelligence.
I don’t believe this process will slow down, because today anyone can create content quickly, and we’re seeing that even Google is having serious trouble handling this flood of information. From my point of view, it’s a very concrete risk.
LLM-related crawlers can ignore rules like robots.txt and access content never meant to be public: in your experience, how critical has technical control over data access become today for protecting a brand?
Yes, LLMs can happily ignore the robots.txt file. At this point, if you have sensitive information you don’t want ending up on the internet, my advice is to put it behind a paywall or a login. Technical control today is absolutely necessary. I always recommend analyzing your server logs to understand exactly which content is being crawled.
You should also be fully aware of the kind of technological and business information you make public on your site: maybe in the past search engines didn’t care at all about technical documentation or internal FAQs, but today, for AI systems, that data is relevant.
A lot of essential content isn’t visible to LLMs because it’s generated via JavaScript or structured in ways that make it hard to access: how much does this problem affect the AI’s ability to correctly interpret a site, and which technical fixes are most effective?
It matters a lot. LLMs can crawl the web, but if they can’t validate and interpret the content on your site, they’ll lose trust in your brand, and that will obviously hurt you. That said, if something gets indexed in Google via JavaScript, the AI can often still scrape the title tags and some meta tags, forming a rough — if partial — idea of the content.
To fix the problem at the root, the best and simplest technical solutions are two: pre-rendering all your content into HTML (using services like Prerender) or switching to Server-Side Rendering (SSR).
LLMs are highly volatile and non-deterministic, making it pointless to chase a stable ranking: which KPIs truly become priorities for evaluating an AI SEO strategy?
As I said before, talking about “rankings” here is almost an oxymoron: if a position isn’t stable, what’s it worth? As a result, I’m focusing in particular on business KPIs such as direct conversions coming from AI, assisted conversions, or better yet revenue and the income generated.
I also monitor brand appearance for specific prompts that intercept different stages of the conversion funnel. Another important indicator, even though it’s hard to translate into an exact number, is the brand’s overall “presence” and consistency across the entire web. Maybe in the future tools will manage to calculate an “average ranking” across prompts, but for now the priority is tracking real impact using the tools we have available.
Visibility in LLMs can change even within minutes: how do you build a solid presence that doesn’t depend on such unpredictable fluctuations?
This brings us back to the basic principles of SEO: you can’t build a strategy by chasing algorithms. You have to build a strategy based on the fundamentals of good marketing. You need to be present wherever your audience is, to understand how the brand appears online.
That means fixing technical issues, building solid Schema markup on the site, and creating a good internal linking architecture. Every page should be about a single topic and answer that topic’s intent immediately. In short, it’s about applying the fundamentals and fully understanding how search engines and AIs track your entity online at a global level.
Many GEO strategies are built on correlations mistaken for causation: how do you distinguish today between factors that genuinely influence AI citations and mere characteristics of content that is already authoritative?
It all comes down to going back and understanding how LLMs really work. We know they’re predictive models, but fundamentally they try to corroborate (confirm and cross-check) the information present on the internet. As with most of SEO, to separate cause from mere correlation you have to run tests.
You need to be able to say: “We applied this change — what was the actual result?” It’s not easy, but you do it by keeping obsessive track of the situation before and after the changes you introduce, to assess the real impact.
Research shows that most AI citations come from third-party sources rather than the brand’s own content: how central are distribution and presence on other sites today, compared to publishing on your own domain?
It becomes essential precisely because of that information “corroboration” mechanism I was talking about. You get proof of it when you start testing outside your own site: maybe you launch a press release (Digital PR) and notice your mentions in prompts increasing, or you actively go and correct wrong information about your brand spread across third-party sites, and shortly afterwards you see the AI update your presence and its answers. It’s essential to keep a historical record and track how the brand appears on these third-party sites, because they directly influence the answers AI systems give.
Many people start out studying without ever testing in the field: what was the moment you realized that practice is worth more than theory in your SEO journey? Any anecdotes to share?
Oh, wow. Practice versus theory… I actually wrote a LinkedIn post about exactly this — it really resonates with me! When I started out in SEO, I was hyper-focused on reading every single guide and article. That gave me a great foundation, but I truly learned an enormous amount (and much more) when I started applying that knowledge directly on my own sites.
Back then I was doing Affiliate SEO, so I could test in the field and see with my own eyes what was ranking. I’d tell myself: “Okay, I made this change, and this is exactly what’s impacting the ranking.” Through testing you realize that a lot of what Google says is — pardon the expression — nonsense. They might say “this shouldn’t happen,” and yet you see it working. Official information should be read, sure, but you always have to validate it in the field. For me, my affiliate sites and my freelance work were the real turning point.
With the arrival of AI agents that choose, compare, and purchase on behalf of users, what do you think will be the main priority for an SEO in the coming years?
I think the main priority will be a mix of all-around marketing and technical work. It will become a kind of ORM (Online Reputation Management): what does the internet say about you? Is what it says correct? Can we influence it? So SEO will look a lot more like classic digital PR.
The second cornerstone will be purely technical: analyzing server logs, making sure there are no crawling issues, and optimizing the infrastructure. Then there will be the user experience side… or rather, the LLM experience! When an AI agent lands on your site to complete an action (like making a purchase on behalf of a user), can it do so easily? Can it navigate and complete the operation?
Of course, content will remain an essential component, especially original content capable of delivering “Information Gain” (genuine added value and new information); that will do just fine. Search engines will keep existing in some form, but AI agents will be a fully integrated part of them.
Being present — but above all consistent — across the entire web
Every time I read or listen to someone who really works in the field — with logs open, test sites running, and data at hand — I get the same feeling: SEO never stands still, it keeps transforming.
And those who have always worked on the fundamentals find themselves, almost paradoxically, in the best position to face the most radical changes.
The point that struck me most? It’s not about learning new tricks, but about truly understanding how the medium works.
They shouldn’t be treated like Google.
And those who understand this now, instead of waiting for it to become obvious to everyone, already have a huge advantage. This is exactly the kind of thinking I’d love to see more often in the industry.
A heartfelt thank you to Peter Rota for his invaluable, always practice-oriented insights. This was one of those interviews you read in one sitting — the kind that makes you want to go and test some of the advice right away, am I wrong?
Clear, direct, no frills — exactly the way we like it. It was a real pleasure to host Peter, and we’re sure you found his insights just as fascinating.
See you at the next episode of SEO Confidential, with another incredible figure from the world of SEO (and GEO!). Don’t miss it.
