For years, the SEO industry has clung to a version of search that no longer reflects how search works.

We still talk about title tag length, keyword distribution, and page-level optimization as if search engines are simply counting words and ranking documents. But search hasn’t worked that way for a long time, and AI search makes the gap impossible to ignore.

In this AMA, Mike King explains how AI search works, how to structure content for retrieval, and why Relevance Engineering is the future.

1. What is the biggest gap between how SEOs perceive AI and how AI search really works?

The biggest gap is that people keep saying AI search is “just SEO” when it isn’t.

AI search expands what SEO has always done, but it works differently. AI systems use techniques such as query fan-out, agentic RAG, and semantic retrieval to break prompts down, evaluate passages, and decide which information to surface.

It creates a problem for SEOs because most tactics were built for an older version of search. 

For the last decade, search has moved through three stages: 

• Lexical search: Looks at the presence and distribution of words
• Semantic search: Uses machine learning to understand meaning
• Hybrid search: Combines both, then re-ranks results, typically through a process called Reciprocal rank fusion (RRF). 

Most SEO workflows are still built around the lexical model, even though modern search is already semantic and hybrid. 

AI search accelerates this evolution because it doesn’t look for a page to rank. It looks for the most relevant pieces of information to extract, compare, and use in an answer.

So when someone reduces AI search to advice like “make the title 60 characters,” they’re optimizing for a version of search that no longer exists.

2. Google’s new AI search guide says most AI optimization tactics are unnecessary. Does that validate your thinking, or do you disagree?

I disagree.

Google’s guidance is self-serving and naive because it treats AI search as if a single platform’s recommendations could explain the entire search ecosystem.

That’s the problem with the “it’s just SEO” argument. It gives people permission to keep doing the same thing, even when it’s producing diminishing returns.

Google’s guidance may be useful if you only care about Google. But it doesn’t explain what works in ChatGPT, Perplexity, Claude, or other AI search environments. Each system retrieves, evaluates, and surfaces information differently.

The only part I broadly agree with is the need to create non-commodity content. Beyond that, Google’s public guidance doesn’t line up with what we know about how these systems actually function.

Google also has a long history of saying one thing publicly while its systems behave differently in practice. So when someone points to Google’s guidance and says, “See, AI search is just SEO,” they’re looking for an excuse to stay comfortable.

3. AI search requires a fundamentally different strategy, but the consensus is that it layers on top of traditional SEO. What are your thoughts?

AI search is an expansion of SEO, but there’s enough of a step-change difference that it deserves to be treated separately.

The issue isn’t that every tactic is brand new because some tactics overlap. For example, Digital PR, link building, sales outreach, and fundraising all use similar motions. But no one argues they are the same discipline because they share tactics.

AI search works the same way. It may use some SEO tactics, but that doesn’t make it traditional SEO.

The problem with saying “it’s just SEO” is that it makes the work smaller at the exact moment teams are open to something new. It keeps AI search trapped inside the old perception of SEO, limiting what it can do. 

If we keep forcing it into the same box, we make it harder to get the budget, tools, team, and strategic ownership this work requires.

4. Given what we know about query fan-out, how should we write and structure content to attract more citations and visibility in LLMs?

First, stop thinking about this as a writing problem alone.

AI systems don’t only evaluate written content on your website. They also consider your broader content ecosystem, including video, social content, earned media, and shared properties. Your owned pages matter, but they are only one part of the surface area AI systems can pull from.

When we’re talking specifically about copy on a page, the goal is atomicity. Each paragraph should focus on one clear idea rather than mixing several topics.

It’s important because query fan-out breaks a prompt into multiple related searches and pulls the most relevant passages from the results. Those passages are scored for relevance using distance measures like cosine similarity or Euclidean distance.

For context, cosine similarity measures how closely two pieces of text align in meaning by comparing the angle between their vector representations, which is why tightly-focused passages tend to score higher than those covering several ideas at once.

If one paragraph covers ten different subjects, it can score lower for individual topics because it dilutes meaning, as seen in the example below.

The better approach is to make each passage easy to isolate, understand, and lift into an answer.

That means:

• Use clear headings
• Put the answer close to the heading
• Keep each paragraph focused on one idea
• Include useful data points
• Use semantic triples that make relationships clear

The more focused the passage, the easier it is for an AI system to understand why that passage answers the prompt.

Further reading:

John Iwuozor also covers this in his article on optimizing content for Google’s AI mode.

5. For enterprise sites with massive content libraries, what does a content maintenance program look like?

Lean into AI, with checks and balances built in from the start.

The scalable approach is a RAG pipeline with two functions: one that generates content, and one that critiques it. If the critic flags a problem, the draft goes back to the generator, and the loop continues until the content clears the bar. 

Build the pipeline on an index of your own data, including:

• Legal documentation
• Internal content
• Product data
• Customer data
• Brand guidelines
• Subject matter expertise

This gives the system the context it needs to maintain content at scale without drifting away from your source of truth.

If you have a large content marketing team capable of doing this review manually, that works too.

AI can help identify what needs updating, draft changes, and review content against internal standards. Humans should handle judgment, strategy, and final quality control.

What doesn't work is publishing AI-generated content at scale without a review process.  

Traffic might spike but eventually collapse, as Lily Ray noted in the above post. 

The reality is that Google can't reliably detect generative content and demote it outright. What happens instead is that new content receives a provisional site quality score, which boosts it for a short period. 

After a while, Google looks at user signals to determine if the content should continue to rank. If those signals say the content isn't good, the score falls and so does the traffic. In that sense, it’s more of a UX problem than a generative AI problem.

6. How should brands handle JavaScript and its impact on AI visibility?

JavaScript is not a major issue in Google or Bing environments, unlike in AI systems like ChatGPT and Perplexity, which don’t render JavaScript. Hence, anything served client-side through JavaScript is invisible to them.

Vercel’s analysis of major AI crawlers found the same gap: many AI crawlers fetch raw HTML but do not render JavaScript.

So brands need to think carefully about what they serve in HTML and what they leave behind in JavaScript. 

iPullRank's Context Parity Explorer is a free tool that solves this problem. Enter a URL, and it crawls the page as Googlebot, AI bot, and a regular browser and shows you what each one actually sees. 

For content you want AI systems to see, make sure it is available in the raw HTML. That includes key product details, descriptions, answers, data points, and anything that helps explain your brand or offering.

Conversely, JavaScript can act as a barrier against competitors or AI scrapers. Rendering assets with JavaScript makes them harder to extract in environments that do not execute it, effectively reducing visibility for content you wish to protect.

7. iPullRank generated $26 million in additional value for one client using Relevance Engineering. How do you measure AI search impact?

That number comes from referral traffic, but AI search measurement can’t stop there.

We break AI search impact into three buckets:

Input metrics: These are the levers we can directly influence. We track rankings for query fan-out queries, passage relevance scores, and bot activity to understand whether the work is improving the signals that feed AI visibility.

Channel metrics: How visible the brand is inside AI search environments. It includes share of voice, citation rate, citation sentiment, and citation accuracy.

Performance metrics: How AI referral traffic performs after it reaches the site. It includes conversions, revenue, lead quality, and any business outcome tied to that traffic.

Each layer feeds the next. The input metrics influence the channel metrics, which in turn influence downstream performance.

So when we talk about $26 million in additional value, we’re referring to the traffic we drove from AI search and its performance. But the broader measurement model has to connect what we can control, how AI systems respond, and whether that visibility turns into business value.

8. You coined the term Relevance Engineering. Why do you think this framework is important for AI search?

I coined Relevance Engineering because there’s a clear gap between what SEO has been and what AI search now requires.

Relevance Engineering sits at the confluence of information retrieval, content strategy, AI, digital PR, and UX. You’ll notice I didn’t include SEO on that list, and for good reason. 

You can be a good SEO without understanding how any of this actually works. Knowing regex or how to set up a 301 redirect is a checklist skill. It doesn't tell you how the web functions underneath it. 

Everything is being built in front of us. There are open-source equivalents for many of the systems behind ChatGPT, AI Overviews, and other AI search experiences. 

It's an opportunity to understand the minutiae and build a strategy that reflects how these systems work.

It’s important to separate AI Search from SEO because when something is labeled as an SEO task, it’s treated as a line item on a checklist: meta tags, free traffic, and one person in the corner cleaning up everyone else’s mistakes.

Calling this something else is a blank slate. It allows us to build cross-functional teams that are properly funded from the start.

Further reading:

Learn more about Relevance Engineering from iPullRank

9. What does the ideal Relevance Engineering team look like?

At a minimum, a Relevance Engineering team needs people who can own each discipline, including:

• An engineer who understands AI
• A content strategist
• A UX specialist
• A digital PR specialist
• Someone with SEO experience

I introduced this framework at SEO Week, where I positioned Relevance Engineering as a response to the limits of traditional SEO in AI search.

SEO experience still matters, but it doesn't need its own dedicated seat. Increasingly, the engineer on the team is the person who came up through SEO and built the technical skill set to go with it.

SEOs are still the best-positioned people to lead AI search work. Most folks just need to upskill into this environment, and paradigms like vibe coding have removed any real excuse not to.

10. What is the value of digital PR to AI search, and how does it work technically?

When I talk about digital PR for AI search, I’m not talking about link building. The value is having your brand, message, and expertise covered on sites that AI systems already consider authoritative.

AI systems look for consensus and authority when deciding what to trust. Having your message appear across your own site, news outlets, and partner content creates multiple instances within the query fan-out that validate the same answer. 

Digital PR focuses on placements and mentions, rather than links, to influence the narrative AI systems use when describing your brand.

11. AI Mode in five years — does Google still exist as we know it?

Google still exists, but search becomes less of a place we go and more of something built into everything we use.

I think Google is best positioned to win the generative search race. But the experience of searching will look very different. 

We’re moving toward a world that feels more ambient (like the movie Her), where information comes to you proactively instead of waiting for you to type a query into a search box.

As a search marketer, it’s unsettling, but as a user, it’s exciting because this is the world I want to live in.

12. What is the most underrated technical move for AI visibility right now?

Page speed.

Not necessarily for Google, because Google already has an index. But for systems like ChatGPT and Perplexity, that request pages in real time. If your page is too slow, it gets skipped and becomes ineligible.

You’ll see this as a 499 response in your log files. It’s not a status code most SEOs are used to seeing, because it’s not part of the classic HTTP spec, but it indicates that the client closed the request before the server finished responding.

The fastest fix is often edge caching through your CDN. It makes your content easier to retrieve, giving AI systems a better chance to access and use it in real time.

SEO has been outgrowing its old definition for years, and AI search makes the gap impossible to ignore. 

Relevance Engineering closes that gap by giving the work a name, a team, and a budget that match its requirements. 

However, SEOs are best positioned to lead this next phase of search, and the way to execute is through a cross-functional relevance-engineering team.