For years SEO focused on rankings and clicks. AI Search introduces a different question: why does an AI choose one source over another?

For most of the history of SEO, the objective has been surprisingly simple. Build a better page, optimize it more effectively than your competitors and try to appear as high as possible in the search results. Everything else, from technical audits to keyword research and link building, was ultimately designed to support that single goal.

That way of thinking still has value. Rankings haven't disappeared, organic traffic still matters and search engines continue to send billions of visits to websites every day. But after spending the last few months studying AI-powered search, I've started wondering whether we're focusing on the wrong question.

Instead of asking "How do I rank higher?", perhaps we should also be asking something else.

Why does an AI decide to use one page as a source instead of another?

It sounds like a subtle difference, but I don't think it is. Traditional search engines primarily rank documents. AI-powered search systems generate answers by retrieving information from multiple sources, interpreting it and combining it into something new. The final answer isn't simply copied from one page. It's reconstructed from pieces of information that the system considers reliable, relevant and easy to reuse.

That changes the way I look at content.

For years the common advice was to create the most complete article possible. If a competitor published a guide with 2,000 words, you wrote 3,000. If they covered ten subtopics, you covered twelve. It wasn't necessarily bad advice, because comprehensive content often deserves to rank well.

However, some of the examples we've been collecting recently tell a more complicated story.

In one case we analysed, Google AI Overview cited a commercial product page before Wikipedia and before another article that explained the same subject in much greater detail. It doesn't prove that shorter pages are better, and it certainly doesn't reveal Google's algorithms. What it does show is that the relationship between completeness and citability may not be as straightforward as many of us assumed.

The more examples I collect, the more I find myself paying attention to different characteristics. Is the page focused on a single topic? Are the key concepts explained clearly? Does every paragraph introduce useful information, or is the important part buried inside long narrative sections? Could an AI retrieve a small section of the page and understand it without needing the surrounding context?

Those questions barely crossed my mind a few years ago. Today they seem increasingly relevant.

One idea I've been thinking about is something I informally call reconstructability. It isn't an official SEO metric, and I don't claim that Google uses it. It's simply a way of describing content that can be understood, extracted and reused with very little ambiguity. When an AI retrieves a paragraph, it has to decide whether that information is clear enough to become part of a generated answer. Pages that communicate concepts cleanly may have an advantage, even when they aren't the longest resources available.

That doesn't mean technical SEO suddenly becomes irrelevant. Fast websites, structured data, internal linking, crawlability and a solid site architecture still matter because they help search engines discover and understand your content. Brand authority still matters. Expertise still matters. None of those fundamentals disappear simply because AI-powered search is becoming more common.

What changes is the destination.

For years we optimized almost exclusively for rankings. Now we're beginning to think about AI visibility as well. Those are related goals, but they aren't always the same. A page can rank well without becoming a frequently cited source. Likewise, a page might influence AI-generated answers even if it doesn't dominate every search result.

That's why I think the SEO conversation is slowly evolving.

Instead of asking only how to create content that ranks, we should also be asking how to create content that an AI can confidently understand, verify and reuse. Those are different problems, and they may require slightly different approaches.

The truth is that nobody outside Google knows exactly how AI Overview selects every citation. Anyone claiming to have a complete formula is almost certainly oversimplifying a very complex system. What we can do, however, is observe real examples, compare them, look for recurring patterns and challenge our own assumptions whenever the evidence points in a different direction.

That's the approach we want to take at NetContentSEO.

Rather than publishing absolute answers, we'd rather document experiments, analyse real-world case studies and learn from the behaviour of AI search systems as they continue to evolve. Some hypotheses will turn out to be wrong, others may prove surprisingly accurate, but the process itself is where the value lies.

Perhaps the biggest change isn't that SEO is disappearing. It's that we're being asked to think beyond rankings for the first time in a long while.

Maybe the next generation of websites won't be remembered simply because they ranked first.

Maybe they'll be remembered because they became the sources that AI systems trusted enough to cite.