Is Relevance Engineering Just SEO With Better PR?

On June 30, Mike King of iPullRank published a piece arguing that the AI search era needs a new job function. He calls it Relevance Engineering, a cross-functional discipline that folds SEO, content strategy, digital PR, UX, and a bit of AI/ML engineering into one practice. It was probably the most argued-about strategy post of the month.

The skeptical reaction showed up fast, and it is worth taking seriously. Strip the framing away and you get a familiar list. Digital PR is link building with a nicer name. Content strategy, UX, technical hygiene, these are things any competent SEO team already owned. So the question underneath the debate is fair. Is this a real shift, or SEO with a rebrand and a consulting upsell attached?

Where the rename argument lands

Start with the strongest version of the skeptic's argument, because it lands more than SEO purists want to admit.

Every component King lists has been part of good SEO for years. Digital PR agencies were pitching journalists for links a decade ago. Content strategists were mapping topics to intent long before anyone said "passage." UX and page experience became ranking talk around the 2021 Core Web Vitals rollout. Put those on a slide, add an engineer, and you can call it a new discipline. You have not necessarily built one.

There is also a commercial read. New categories sell. An agency that names the category gets to position itself as the authority on it, and "Relevance Engineer" bills higher than "SEO." None of that makes the idea wrong. It does mean the burden of proof sits with the people making the claim, not the people doubting it.

So where does the rename argument actually break?

What changed underneath the label

The honest answer is that the retrieval system changed, and SEO's mental model was built for a different one.

Classic SEO optimizes for a machine that already read the web. Google crawls, indexes, and stores pages ahead of time, then matches a query against that index. Speed of your server at query time barely matters, because the copy Google scores was fetched days or weeks ago. AI search does not work that way in three specific places, and each one breaks a habit SEOs treat as settled.

Query fan-out comes first. An AI system takes one prompt, decides it is too broad, and splits it into a spread of sub-queries, often five to twenty, then retrieves against all of them at once. iPullRank traces the mechanism to a Google patent (US 2024/0289407 A1) describing a multi-step retrieval and synthesis process. One page tuned for one head term is the wrong unit of planning when the engine is asking twenty smaller questions you never saw.

Agentic RAG comes second. Some AI engines fetch pages live, at the moment of answering, rather than reading a pre-built index. That single change moves page speed from a comfort factor to a gate.

Semantic retrieval and Reciprocal Rank Fusion come third. The engine scores passages, not whole pages, then merges the ranked lists from all those sub-queries into one using RRF, which rewards passages that keep showing up near the top across different sub-queries. Your page does not get ranked. Your paragraphs get ranked, separately, against questions you were never targeting.

None of those three is "SEO but harder." They describe a different machine.

The page speed example nobody expects

The cleanest proof that the model changed is also the most counterintuitive tactical point in King's piece. Page speed matters more for AI visibility than it does for Google, and for the opposite reason people assume.

Google already has your page. A slow server costs you some crawl efficiency, but the copy being scored is sitting in the index. ChatGPT and Perplexity, when they fetch live, are on a timer. If your page does not respond fast enough, the agent gives up and answers from someone else. You are not ranked low. You are absent.

You can see it in log files. A 499 status is the server's record that the client closed the connection before a response came back. When an AI agent is the client, a cluster of 499 responses reads as agents timing out on your URL and walking away. The fix is unglamorous. Push content to the edge with a CDN so the first byte arrives before the agent quits.

Here is the trap. A green score on fast.com or PageSpeed Insights measures how a browser in a data center experiences your page. It says nothing about whether an AI crawler, hitting an uncached path from a different region, got a response in time. The tool most SEOs reach for to prove the site is fast is measuring the wrong client.

Content structure stops being page-shaped

The writing changes too, and this is where the discipline earns its own name rather than borrowing SEO's.

If engines extract passages, the passage becomes the asset. King's word for the principle is atomicity. One paragraph, one idea. A clear heading with the direct answer sitting right underneath it, not four sentences of throat-clearing first. Relationships written out as explicit subject, verb, object statements instead of implied across three paragraphs. "Search Agency runs AI visibility audits for Southeast Asian startups" is a fact a retriever can lift whole. The same information spread across a section is not.

The reason is mechanical. A tightly focused passage scores a higher similarity match against a narrow sub-query than a rambling one that touches five topics. Traditional on-page SEO optimized the page as the unit. This optimizes the paragraph, and a page full of loosely related prose is a page full of weak passages.

Reconciling the two

Both sides are half right, which is why the argument keeps going.

The skeptic is correct that most of the labor is recognizable. A Relevance Engineering team is still doing content, links, technical work, and structure. Nobody invented a new craft from nothing. The people who could do this well are mostly people who were already good at search.

Where the skeptic is wrong is the assumption that the same labor against a new machine is the same job. The retrieval model changed at the level of query expansion, live fetching, and passage scoring. Those are not PR problems dressed up. They need someone who reads information retrieval papers next to someone who writes, and that pairing is not what most SEO teams were staffed for.

The most useful way to read King's framing is not "SEO is dead" and not "SEO with better PR." It is SEO's scope expanding past the page, joined to genuinely new retrieval mechanics, run by a team where the SEO is one voice among several instead of the only one in the room. iPullRank says it generated 26 million dollars in additional value for a single client working this way, measured through AI referral traffic into revenue. Treat that as their reported figure rather than a law of nature. It still tells you the model is operational, not a keynote abstraction.

Arguments about the label will keep going, and some of that is just people defending a job title. The mechanics underneath are not up for debate. Query fan-out, live retrieval, and passage-level scoring are how the answer gets built now, and a practice organized around the page is going to keep missing them no matter what you call it.

If you want to know which of query fan-out, live retrieval, and passage scoring your own pages are already losing to, that is what my consultancy work is built to find.