It was a landmark year in SEO, largely driven by the uncertainty introduced by AI Search. The year began with the digital marketing community questioning its relevance and ended with a strong affirmation of its central position as it gradually adjusted to new realities. WordPress entered the year with uncertainty about whether the core would see meaningful updates and closed out the year with version 6.9, an update that strongly positions it for AI-led innovations.
GEO Is Recognized But Remains An Inchoate Concept
SEOs Turn To Geo
2025 is the year that GEO went mainstream, energized by client demand for solutions that are specific to AI Search. This resulted in the somewhat awkward situation of some SEOs pivoting to providing GEO-specific services while simultaneously affirming SEO best practices for ranking in AI search. Attempts to define GEO as a process distinct from SEO generally fell short.
WordPress SEO Plugins Go GEO
WordPress SEO plugins faced a similar issue with clients demanding GEO-specific solutions, leading to the introduction of LLMs.txt generation features. LLMs.txt is a proposed standard for providing content to AI; however, it’s a solution in search of existential justification because no AI companies use or have plans to adopt the standard.
While other WordPress SEO plugins justified LLMs.txt support as a future-proofing feature, the Squirrly SEO WordPress plugin was refreshingly candid about its reasons for introducing it:
“I know that many of you love using Squirrly SEO and want to keep using it. Which is why you’ve asked us to bring this feature.
So we brought it.
But, because I care about you: know that LLMs txt will not help you magically appear in AI search. There is currently zero proof that it helps with being promoted by AI search engines.”
Google Accidentally GEOs Itself
Google’s John Mueller has strongly and unambiguously insisted there are many reasons why the LLMs.txt proposal is not viable. Thus, many were startled and amused when Lidia Infante discovered that Google itself was using LLMs.txt. The LLMs.txt file was quickly removed, but that didn’t stop some GEO “experts” from crowing that Google’s use of the file validates LLMs.txt, apparently unaware that Google had already removed it.
Google’s Advice For Better Rankings: Become A Brand
In remarks at the New York City Search Central Live event (which I attended), Google’s Danny Sullivan encouraged SEOs and businesses to think about how they can differentiate themselves as brands in order to improve their search visibility.
Sullivan explained:
“And I’ve seen where people do research and say, ‘I’ve figured out that if you have a lot of branded searches…’ That’s kind of valid in some sense.
…What it’s saying is that people have recognized you as a brand, which is a good thing. We like brands. Some brands we don’t like, but at least we recognize them, right?
So if you’re trying to be found in the sea of content and you have the 150,000th fried chicken recipe, it’s very difficult to understand which ones of those are necessarily better than anybody else’s out there.
But if you are recognized as a brand in your field, big, small, whatever, just a brand, then that’s important.
That correlates with a lot of signals of perhaps success with search. Not that you’re a brand but that people are recognizing you. People may be coming to you directly, people, may be referring to you in lots of different ways… You’re not just sort of this anonymous type of thing.”
Sullivan’s reference to “branded searches” may have been a reference to an article I wrote about Google’s branded search patent that describes the use of branded search queries as ranking factors.
People think of “brand” in terms of something that big sites have and little sites do not. But the reality is that brand is just what people think about a company, and the challenge for any business is to distinguish itself from its competitors in such a way that its customers and site visitors remember it, ask for it by name on Google search, and recommend the site to their friends. That, in a nutshell, is how I interpret what Danny Sullivan was communicating.
User behavior is a trusted source of signals that can indicate qualities like expertise, experience, authoritativeness, and trustworthiness (E-E-A-T). E-E-A-T is not something that an SEO adds to a website. While Google has cryptically referred to signals that it uses to determine qualities related to E-E-A-T, in my opinion, those signals are likely related to how users react to a website, user behavior signals.
Read what Danny Sullivan said: Google’s SEO Tips For Better Rankings – Search Central Live NYC
Advances In AI And Search
This year saw the publication of a number of research papers and patents that point to improvements in AI and algorithms that may play a role in how webpages are ranked.
Google’s Thematic Search Patent
Google filed a patent that describes how an LLM can organize related search results by themes and then provide a short summary. It describes a deep research method that closely parallels what we see happening in AI Mode.
The patent describes the invention:
“In some examples, in response to the search query being generated, the thematic search engine may generate thematic data from at least a portion of the search results. For example, the thematic search engine may obtain the search results and may generate narrower themes (e.g., sub-themes) (e.g., “neighborhood A”, “neighborhood B”, “neighborhood C”) from the responsive documents of the search results. The search results page may display the sub-themes of theme and/or the thematic search results for the search query. The process may continue, where selection of a sub-theme of theme may cause the thematic search engine to obtain another set of search results from the search engine and may generate narrower themes (e.g., sub-sub-themes of theme) from the search results and so forth.”
The takeaway from the above passage is that an AI system that incorporates what’s in the patent is still relying on a search engine for retrieving the documents. What those who are interested in GEO need to wrap their heads around is that what’s being ranked for a given search query is vastly different from classic search because it’s generating “sub-themes” of the initial query and then ranking those webpages in addition to the initial query.
Insight About GEO: While the underlying infrastructure is still classic search, what’s getting ranked is not classic search relative to the initial query. This is the nuance that genuinely distinguishes GEO from SEO.
The patent also describes a summary generator that groups answers by themes using data from passages from documents, but may also use data from titles, metadata, and surrounding passages.
Read more: Google’s Thematic Search Patent
Google’s Patent On Personalization In AI Answers
Google filed a patent about using five real-world contextual signals to influence the answers that an AI answer engine provides.
The five factors that this system describes as influencing LLM answers are:
- Time, Location, And Environmental Context.
- User-Specific Context.
- Dialog Intent And Prior Interactions.
- Inputs (text, touch, and speech).
- System And Device Context.
The first four factors influence the answers provided by the LLM. The last one influences whether to turn off LLM-assisted answers and revert to standard AI answers.
An interesting part of this patent is about the concept of “related intents.”
The patent describes how this works:
“For example, …one or more of the LLMs can determine an intent associated with the given assistant query… Further, the one or more of the LLMs can identify, based on the intent associated with the given assistant query, at least one related intent that is related to the intent associated with the given assistant query… Moreover, the one or more of the LLMs can generate the additional assistant query based on the at least one related intent.”
This patent is useful for understanding how AI Search differs from Classic Search. It describes a way that AI systems can personalize answers with context-aware responses.
Read more: Google Patent On Using Contextual Signals Beyond Query Semantics
Google’s Patent On Personal History-Based Search
This patent is about solving a user’s problem of identifying where they read about a certain topic, whether the topic was in an email or a webpage. The name of the patent is Generating Query Answers From A User’s History.
Traditional email search did not enable natural language querying; it still relied on basic keyword-matching algorithms. This patent solves that problem, partially through the ability to understand fuzzy queries.
The patent describes this process:
“For example, the browser history collection… may include a list of web pages that were accessed by the user. The search engine… may obtain documents from the index… based on the filters from the formatted query.
For example, if the formatted query… includes a date filter (e.g., “last week”) and a topic filter (e.g., “chess story”), the search engine… may retrieve only documents from the collection… that satisfy these filters, i.e., documents that the user accessed in the previous week that relate to a “chess story.””
Read more: Google Files New Patent On Personal History-Based Search
Google’s Sufficient Context Signal
Google published a research paper introducing a new method for determining whether retrieved content provides enough information to answer a query. The breakthrough makes it possible to identify when retrieved context is incomplete or insufficient, which is a major source of hallucinations in RAG systems.
The paper’s contributions and insights are:
- Defines “sufficient context” as a content passage that contains enough information to answer the question.
- Builds an autorater that classifies whether a retrieved passage has sufficient context.
- Provides the insight that hallucinations can still happen when context is sufficient, meaning that hallucinations are not only a retrieval problem.
- Provides the insight that models can provide correct answers with insufficient context, sometimes because of “parametric memory,” which is the knowledge from their model training.
- Proposes a selective generation framework that uses the sufficient-context signal plus a confidence signal to reduce hallucinations by 2-10%.
SEO Takeaway: The research paper underscores the importance of ensuring that published content contains the necessary context to fully support the topics it covers.
Read more: Google Researchers Improve RAG With “Sufficient Context” Signal
MUVERA
Google’s MUVERA enables multi-vector models to retrieve at speeds comparable to single-vector systems while preserving their ability to perform token-level matching. Token-level matching means the model compares each individual word in the query to individual words in the content it evaluates. MUVERA keeps the accuracy advantages of multi-vector models while removing the heavy computation in the retrieval step by learning efficient virtual document vectors that approximate multi-vector scoring.
Read about Google MUVERA
WordPress And AI
WordPress generated buzz in the developer community with the announcement of the WordPress Abilities API, a way to safely integrate external plugin functionalities into WordPress in a more unified, less fragmented way. This also lays the foundation for a dramatic expansion of capabilities with AI.
According to WordPress:
“This API creates a centralized registry where all functionalities can be formally registered with well-defined schemas, comprehensive descriptions, and explicit permissions. By adopting this common language, plugins and themes will empower AI-driven solutions to seamlessly discover, interpret, utilize, and coordinate capabilities throughout the entire WordPress ecosystem.”
The December State of the Word event in San Francisco provided a sneak peek at the improvements AI will play in online publishing. WordPress co-creator Matt Mullenweg said that he envisions hundreds, if not thousands, of specialized AI models integrated into different levels of the WordPress workflow.
Mullenweg explained:
“So I imagine that in the future, we’ll actually have hundreds, if not thousands, of different specialized models that might be tuned for different things. In fact, in some of our work at Automattic around like a site builder, we’re finding that models that are tuned specifically for like logo creation can be essentially fine-tuned or smaller, cheaper to run, sort of less memory, etcetera, can do more specialized tasks.”
Mullenweg views a future in which narrowly focused models contribute to different parts of the publishing process, showing how WordPress expects AI to take on routine creative tasks so that users can focus on the work that matters, further democratizing the act of publishing online.
2025 “Low-Lights”
Google Blocks Rank Trackers
Google blocked rank trackers from scraping the top 100 search results. An unexpected consequence of blocking rank trackers from scraping the top 100 search results is that Google Search Console began reporting fewer keyword impressions, sending SEOs and businesses into a panic. It turned out that rank trackers had been inflating the Search Console impression data.
This, in turn, caused some SEOs to revise the idea of zero-click searches, an idea dating from at least 2019, that blamed a low click-to-impression ratio on things like Featured Snippets. In hindsight, that low ratio of clicks to impressions was likely due to inflated impression data.
Declining Clicks Is A Reality
The irony of the zero-click idea being revisited is that businesses in 2025 are reporting declines in traffic that are blamed on Google’s AI Overviews and AI Mode. The biggest story of the year related to SEO is arguably the decline of search clicks.
While Google’s CEO Sundar Pichai insisted that Google’s AI Overviews is sending more clicks than ever, SEOs and their clients strongly disagreed with that point of view.
WordPress Versus WP Engine
The news dominating the WordPress world in 2025 was Automattic and WordPress co-creator Matt Mullenweg’s self-described “nuclear” attacks against WP Engine, which included publishing a website with the goal of encouraging WP Engine’s customers to migrate away, locking WP Engine out of the WordPress ecosystem, and creating a copy of WP Engine’s premium version of their ACF plugin.
The basis for the conflict is what Mullenweg describes as WP Engine’s lack of support for the open-source WordPress project. WP Engine responded with a federal lawsuit against Mullenweg and Automattic, seeking to hold them responsible for actions that WP Engine argued hindered its business.
Many months later, Automattic responded with a counterclaim against WP Engine, using creative statistics about WP Engine’s use of SEO that, in my expert opinion, don’t hold up on closer scrutiny (Read: Automattic’s Legal Claims About SEO… Is This Real?).
Automattic and Matt Mullenweg are on solid ground to encourage big corporations to give back to the WordPress community because it supports the long-term viability of the WordPress open source project. It’s quite likely that many in the WordPress community would have rallied behind Mullenweg against WP Engine if he had pursued a less extreme approach toward WP Engine.
Negative Sentiment Against WordPress Co-Creator
What happened between Mullenweg and WP Engine arguably backfired on Mullenweg, generating substantial negative sentiment against him that persists to this very day. The effect is that many in the community are siding against Mullenweg while simultaneously not necessarily siding with WP Engine.
An example of how the negativity persists, Kevin Geary, the creator of the Etch WordPress page builder, recently tweeted:
“As usual, the adults do sensible things and serve the community, and all Matt can do is p— on us and wreak havoc.
WP is an unserious org led by an unserious person. Embarrassing.”
Another example: It didn’t take long for negative sentiment against Mullenweg to arise in a recent popular Reddit discussion about Automattic’s SCF plugin, a fork of WP Engine’s premium ACF plugin.
A Redditor asked:
“ACF vs SCF this far along – have they diverged?
Politics and such aside – , what is the difference now between Advanced Custom Fields and Secure Custom Fields after some time developing?”
A typical comment:
“When you say “politics and such aside,” it’s pretty hard to put GPL theft of the most extreme WordPress has ever seen aside.
Just don’t use SCF. Plain and simple.”
Another Redditor responded:
“Man u must have missed it when the wordpress owner had a feud with wpengine over their branding and spiraled and then stole the ACF plugin and renamed it and started just burning bridges and flexing ownership ability
He even put some petty checkbox on the wp login screen like check this box that you’re in no way working with WPEngine or you can’t log in
It was crazy / petty / weird and then in the end scary for all plugin devs that what you thought was open source could be manhandled and banned and stolen or replaced by one guy at the top of wordpress
Sad to see”
Many people are grateful to Matt Mullenweg for what he’s accomplished with WordPress. But, as the Redditor commented, the conflict was “sad to see.” One doesn’t have to click around the web for long to discover evidence of the extremely negative sentiment that follows Mullenweg around across the internet.
2025 Online Marketing Wrapped
2025 was largely a year of transition. Everything, from SEO to WordPress to the tools that online businesses use, was in the process of preparing for what comes next. In terms of internet marketing, 2025 was the gateway to 2026.
More Resources:
Featured Image: Emre Akkoyun/Shutterstock
