Search engine optimization went from keywords to entities to answers. The next step puts the publisher back into the search.

User intent is person-based. It is the whole field of thoughts, questions and wishes someone carries at the kitchen table, in the car on the way to work, in a waiting room or awake at night. Search intent is query-based. It is the small part of that field which reaches a search box, read by a search engine as a class such as informational, navigational or transactional.

The two form no yes-or-no pair. Search intent sits inside user intent, and user intent reaches far beyond it. A person wants to get rid of a headache and types headache tablet. The search intent is the tablet. The user intent is the end of the headache. She does not like pills and would prefer acupressure, and she does not yet know that acupressure exists.

Search engines read search intent from the SERP consensus, from what already ranks and from what people clicked on it. Clicks follow position, so each reading tends to confirm the result page it came from. Over decades, search intent has built a confirmation bias into the result page. Andrei Broder, who introduced the navigational, informational and transactional classes in 2002, called inferring the user intent from the query “at best an inexact science, but usually a wild guess”.

A page that answers the user intent also carries the search intent inside it. Whoever sells a book or a course on acupressure has to treat headache tablets too, with their advantages and disadvantages, and explain how headaches occur and why faster alternatives to the tablet often exist.

If the query revealed what a person wants, asking the person and reading the query would give the same shares. Reading the query put informational searches at 48 %. Asking people gave an estimated 39 %.

Classic SEO worked on a machine that counted words and links. Google itself started from links: in April 1998, Sergey Brin and Lawrence Page presented a search engine built on the link structure of the web, with PageRank as a measure of a page's citation importance.

The practices of that era followed from it. Keywords went into the title, the meta description, the headings and the first paragraph, often at a fixed density. Links were collected from directories, article sites and link exchanges, with anchor texts that repeated the keyword. Exact-match domains, doorway pages, hidden text and paid link networks were the black-hat end of the same logic.

The technical craft came with it: crawlable HTML, robots.txt, clean redirects and XML sitemaps. Google launched Google Sitemaps in June 2005, and Yahoo! and Microsoft joined the shared protocol in November 2006.

Step by step, and in public. In January 2005, Google introduced rel="nofollow" so that links carrying it receive no ranking credit. In September 2009 it stated that its ranking does not use the keywords meta tag. The update against low-quality sites in February 2011 and the update against webspam in April 2012 took the ground from content farms and bought links.

What remains of Classic SEO is the technical floor. A page has to be crawlable, indexable, fast, usable on a phone and served over HTTPS. Google renamed its Webmaster Guidelines to Search Essentials in October 2022 and completed the move to mobile-first indexing in October 2023. The floor is required, and it wins nothing on its own.

If ranking had kept rewarding the classic playbook, Google's changes would have touched a few queries at the edge. One update alone noticeably changed 11.8 % of queries in the United States.

Semantic SEO works on meaning, and meaning is carried by words. Search systems added models of things to the matching of words. In May 2012, Google introduced the Knowledge Graph with more than 500 million objects and more than 3.5 billion facts about them. Hummingbird followed in August 2013, RankBrain in 2015 as the first deep learning system in Search, neural matching in 2018, BERT in October 2019, passage understanding in October 2020 and MUM in May 2021.

Structured data gave pages a shared vocabulary for the same things. Google introduced Rich Snippets in May 2009, and Google, Bing and Yahoo! launched schema.org in June 2011.

No. The lexical level is where the knowledge of a field lives. Studies and books carry it as words: the terms a discipline has agreed on, its definitions, its names and the exact phrasing of its findings. A search system can only model the meaning of a subject from the words in which that subject has been written down.

Retrieval research still measures meaning-based search against keyword matching. In the study that introduced dense passage retrieval, keyword matching with BM25 already placed a passage containing the answer among the first 20 results for 59.1 % of Natural Questions and 66.9 % of TriviaQA questions, and the meaning-based model raised these shares to 78.4 % and 79.4 % (Vladimir Karpukhin and colleagues, EMNLP 2020).

Semantic SEO therefore works on both levels at once: the exact vocabulary of a field, taken from its studies and books, and the meaning that vocabulary builds.

The entity and the topic joined the word as units of work. A page is expected to name the things it is about, relate them correctly and answer the questions that belong to them.

Quality became a written standard. In December 2022, Google added Experience to E-A-T in its search quality rater guidelines, which made it E-E-A-T. The helpful content system from August 2022 became part of the core ranking systems in March 2024, and Google reported 45 % less low-quality, unoriginal content in its results after that month's core update.

The answer moved above the list. OpenAI released ChatGPT on 30 November 2022. Microsoft launched an AI-powered Bing in February 2023, and Google opened its Search Generative Experience in May 2023. Google rolled out AI Overviews to everyone in the United States in May 2024 and reported more than 2 billion monthly users in July 2025. OpenAI introduced ChatGPT search in October 2024, Anthropic added web search to Claude in March 2025, and Google brought AI Mode to everyone in the United States in May 2025.

Two disciplines grew out of this. Answer engine optimization (AEO) works on being the source a direct answer is taken from: featured snippets, voice answers and answer boxes. Generative engine optimization (GEO) works on being named in an answer that a language model writes: citable passages, verifiable figures and entities a model can ground.

Both kinds of system learned from human judgement. Ranking models have learned from relevance labels since RankNet in 2005. InstructGPT and ChatGPT were trained on human rankings of model outputs. Google ran 719,326 search quality tests with external raters in 2023 and states that these ratings evaluate its systems without directly affecting the ranking of any page. The answer a person reads was shaped by judgements other people made before.

W. Chan Kim and Renée Mauborgne describe red oceans as “all the industries in existence today – the known market space”, where “cut-throat competition in existing industries turns the ocean bloody red”. Their study of business launches in 108 companies shows what that competition returns.

In search, the red ocean is the SERP consensus: the settled core of a subject that every ranking page repeats. Search intent is read from this consensus and confirms it. Everyone can cover it, everyone does, and each further restatement is worth less. Answer systems compress exactly this part into one paragraph and hand it over without a visit.

If extending what a market already offers paid best, the 86 % of launches built that way would earn most of the profit. They earn 39 %.

Three findings show where the spend on consensus pages goes. Almost all pages receive no search traffic at all. Most searches end without a click, and fewer end in one when an AI summary appears. Paying for a position on demand that would have arrived anyway returned a loss in the largest controlled experiment on paid search.

If publishing a page were enough to earn search visitors, most pages would receive some. 96.55 % receive none.

If a place on the first result page still delivered the audience, most searches would end in a click. In 2024, most of them ended without one.

If buying ads for demand that already exists paid off, the controlled experiment would show a positive return. Its best estimate is minus 63 %.

W. Chan Kim and Renée Mauborgne coined the terms red ocean and blue ocean. W. Chan Kim is Distinguished Professor of Strategy and International Management, Emeritus, at INSEAD. Renée Mauborgne is the INSEAD Distinguished Fellow and Affiliate Professor of Strategy. Together they direct the INSEAD Blue Ocean Strategy Institute in Fontainebleau, France. They introduced blue ocean strategy in Harvard Business Review in October 2004 and in the book Blue Ocean Strategy, published by Harvard Business School Press in 2005, in an expanded edition in 2015 and in an enhanced 20th anniversary edition in August 2026. The book is based on a study of 150 strategic moves spanning more than 100 years across 30 industries and has sold over 4 million copies in 49 languages. Blue Ocean Shift followed in 2017.

Their tools are theirs by name: value innovation, the simultaneous pursuit of differentiation and low cost, the strategy canvas, the Four Actions Framework and the Eliminate-Reduce-Raise-Create Grid. Thinkers50 named them the most influential management thinkers in the world in 2019, and Renée Mauborgne was the first woman to reach that position. In 2023, Harvard Business Review recognised them as two of the four leading thinkers in its 100-year history.

The transfer of their concept to search results is the work of this site. The concept, its terms and its tools belong to Kim and Mauborgne.

In the words of Kim and Mauborgne, blue oceans are “all the industries not in existence today – the unknown market space”. Beyond the SERP consensus, the blue ocean is made of the part of user intent that reaches beyond search intent: the questions and wishes that belong to a market, that people demonstrably carry, and that no first result page answers. Nobody competes there yet. There is no keyword price, no ranking to take from anyone and no summary an answer system could reuse.

If the first result pages already covered what people ask, new searches would be rare. Google reports that 15 % of the searches it sees every day are new.

If the first result page answered what people want to know, open questions would be the exception. On 90 % of result pages, every top-3 page leaves at least one open.

Johannes Faupel finds it with the IgnoranceGraph Engine. The engine is available only in collaboration with him and a small number of partners. It is not sold as software, and there is no SaaS version and no public access.

A collaboration delivers the result: the subjects and questions a market is missing, named and ordered, ready to be written, and ready to be built into Supervised Search.

The next step is: Contact