Between LLMs, organic traffic loss, Discover and AI training data, new tensions are emerging between platforms and publishers (and new dynamics for online visibility)
SEO is going through one of the most turbulent phases in its history. Between AI Overviews, declining organic traffic, changes in Google Discover and new balances between platforms and publishers, many of the certainties on which editorial strategy has been built over the past fifteen years are rapidly wavering.
Anyone working in publishing, marketing or SEO feels it every day: the relationship between content, visibility and distribution is changing profoundly.
To better understand what is happening, in this new episode of SEO Confidential we interviewed Harry Clarkson-Bennett, SEO Director at The Telegraph and author of the international newsletter Leadership in SEO, which I strongly recommend you follow.
Harry has almost a decade of experience across agencies, editorial projects, affiliate sites and organic growth strategies, and he observes search from a privileged point of view: that of someone who has to turn theory and algorithmic models into concrete results for one of Europe’s most important publishers.
In our interview we will discuss some of the hottest topics of the moment: the role of content clarity in the age of LLMs, the problem of source attribution in AI Overviews, the impact of new technologies on publisher visibility, the future of Google Discover, all the way to the most controversial dynamics of contemporary SEO, including drop domains, canonical abuse and AI-generated spam.
I recommend reading his answers carefully — so direct and full of insights — to understand how search is really changing. Here is our conversation with Harry!
“At some point AI and LLMs will have to pay for quality data, but for that to happen publishers must stay united”
In your article “Information Retrieval Part 1: Disambiguation” you argue that search engines and LLMs reward content that is “easy to understand”, rather than the content that is genuinely better. Isn’t there a risk that this logic ends up penalising complexity, critical thinking and truly original content, favouring a standardisation of information that Google claims it wants to fight?
Answering the user’s question directly has always been a central principle in search. Giving people what they are looking for in the shortest possible time encourages a positive interaction with the content, not least because attention spans are limited. This approach clashes with an old SEO habit of producing very long texts, often around 2,000 words, even when it was not really necessary.
Google has introduced evaluation systems based on broad consensus among sources, designed to limit the spread of misinformation. This does not mean that more complex or nuanced queries must have a single answer. In many cases it is legitimate to present multiple interpretations, opinions or different perspectives.
It becomes essential, however, to maintain clarity in the structure and presentation of information. Well-organised headings, explicit answers and a logical sequence of content help both users and search systems understand the meaning of the page.
Clarity also concerns technical aspects: structured data, author attribution, relationships between entities, presence in the knowledge graph, social profiles and other signals that help search engines precisely identify people, organisations and content.
In a landscape increasingly crowded with AI-generated content, the ability to be clear, verifiable and free of ambiguity becomes even more important. Information disambiguation is one of the key factors in maintaining credibility and visibility in search systems.
Getting into AI training data is becoming increasingly decisive in order to matter in the AI-based search ecosystem, but access to these datasets is now mediated by commercial agreements, paywalls and big platforms. Aren’t we moving from an open web to a visibility system reserved for those who can afford to be acquired or integrated?
All publishers need to make an impact. It is not enough to produce content that repeats answers already available, or to publish news without real reasoning or without a point of view that differs from the dominant one. This is even more true for smaller publishers, who need to cut through the informational noise, distribute their content and get it to the right people. In other words, they must claim their space with determination.
A publisher can only control what happens within its own pages. It can choose to be clear, rigorous and original. These are fundamental elements. At the same time, it is not necessarily essential to block bot access. More than 60% of large language models have been trained using datasets from Common Crawl.
Since many large publishers are preventing their content from being used for training, this situation could even represent an opportunity for smaller publishers, at least for those who believe LLMs can offer visibility or value.
Many publishers are blocking bots anyway, for two main reasons. The first is that content has often been used by third parties for commercial purposes without authorisation. The second concerns the desire to negotiate commercial agreements with the companies developing these models.
In the short term this involves a trade-off in terms of visibility in AI-based systems. For most publishers, however, the immediate value remains limited. This is a structural change that is redefining the relationship between platforms and content. In several cases publishers are starting to coordinate in order to put pressure on the AI companies that need reliable, high-quality data to continuously improve their models.
You speak openly about a risk of model collapse due to the scarcity of quality data and the growing use of synthetic data. Isn’t there a danger that AI ends up reinforcing an increasingly self-referential vision of the web, where a few brands and a few points of view keep bouncing around endlessly?
Yes, that is exactly the risk! I believe that at some point these companies will have to pay for quality “data”, but for that to happen publishers must stay united.
Google Discover seems to reward above all the initial performance of content, such as CTR and engagement in the first hours after publication. To what extent does it make sense to push on “clickability” without risking long-term damage to editorial credibility and the relationship with readers?
I believe it is no longer worth focusing exclusively on clickability, especially on Discover. I think the Discover algorithm update was designed precisely to fight this type of content. The headline and the image must be highly clickable, but they still need to accurately describe the content of the page.
But Google Discover does not only reward initial engagement. If a piece of content proves interesting to a specific group of users and that group interacts positively, the algorithm tends to distribute it more and more within the same audience and towards similar segments. The first clicks therefore remain very important, but what really matters is the quality of the interaction. A piece of content must manage to exceed the level of engagement expected in the early stages of distribution.
In this context, the evolution of journalism rewards behaviours that foster a more direct relationship with the audience. One example concerns the role of journalists as recognisable figures: encouraging them to build their own profile and share their content helps amplify its reach and strengthen the relationship with readers.
In recent months fairly blatant spam content, recycled expired domains and sites built solely for monetisation seem to be reappearing even in the most competitive SERPs. Is this a temporary problem linked to the AI wave, or a structural sign that Google is losing control?
Over the past ten years many publishers have made excessive use of technology. Search should be a tool at the service of journalism, not the factor that drives its editorial choices. The ability to produce content on a scale never seen before has made it more complex to maintain high quality standards. At the same time, on the spam front, Google seems to have lost part of its control over the situation.
It may be a temporary phase, but it is likely that some problematic practices, such as the drop domain phenomenon or canonical abuse, will continue to circulate for some time.
(A drop domain is an internet domain that has not been renewed by its owner and therefore expires and becomes available for anyone to register. In SEO it is often purchased because it still retains backlinks, authority or history, which some try to exploit to gain ranking advantages, Ed.).
(Canonical abuse occurs when the
rel="canonical"tag is used incorrectly to make Google believe that different pages are actually the same page. In practice, the canonical is set towards another URL to transfer value or manipulate indexing, even when the content is not genuinely duplicated, Ed.).
The data show a significant collapse in traffic from Google Search and Discover to publisher sites, especially after the introduction of AI Overviews. Is this a phase of technological adjustment, or a structural change that makes the role of the media as a direct source of traffic less and less central?
A structural change, no doubt about it for me. AI Overviews present several problems. One of the most evident concerns source attribution: Google often does not clearly indicate where the information used in AI-generated answers comes from. It does not look like a simple technical shortcoming, but a choice tied to the platform’s business model.
This dynamic is producing a structural change in the publishing industry. Many publishers are starting to diversify their content, investing more in individual journalists, in video and in a stronger presence on social media.
The reason is simple: the audience’s attention has shifted. More and more people consume information through formats and platforms different from those on which traditional publishing built its presence. This is why publishers are forced to rethink their strategies and adapt to an information ecosystem that is changing rapidly.
Beyond rankings: source credibility and brand strength are becoming the new competitive battleground
After reading this interview, the question is almost inevitable: where is online search really heading?
The conversation with Harry C. Bennett highlighted several signals that cannot be ignored. What emerged strongly is the importance of content clarity, of information disambiguation and of the ability to make articles, authors and entities easily understandable for search systems and LLMs.
At the same time, we saw how AI Overviews and the new AI-based systems have already produced a structural change in traffic distribution.
Other decisive points also emerged during the interview: the real role of engagement on Discover, the limits of strategies based solely on clickability, the increasingly delicate issue of the data used to train artificial intelligence, all the way to the problems linked to spam, drop domains and certain SEO practices that are making a comeback.
From a business perspective, artificial intelligence represents a profound change in the way information is distributed and consumed. When answers are generated directly within the platforms, competition shifts from mere rankings to source credibility and brand strength.
The most forward-looking companies (yes, your competitors too!) are investing in authority, in the recognisability of their authors and in a solid presence across multiple channels. For these reasons, turning expertise and content into real trust becomes central.
Trust means attention, relationships with your audience and, in the end, concrete results: clients, conversions and growth.
We thank Harry for this exchange of views, and we thank you for reading this far.
We will be back next week with a new guest and a new interview, right here on SEO Confidential. Because understanding where search is heading today means making better decisions for tomorrow’s business.
