“AI tracking is just garbage dressed up as insight”: Pedro Dias tears down the false myths of modern visibility
Welcome back to SEO Confidential, the space where every week we hand the microphone to the most authoritative voices in international SEO to understand how being found online is changing, amid enthusiasm, fears and very concrete questions for entrepreneurs, publishers and digital professionals.
The guest of this episode is Pedro Dias, one of the most authoritative names when it comes to visibility on today’s web.
After his experience on Google’s Search Quality team and leadership roles at companies such as eBay, Reach plc and JoVE, Pedro now helps organizations and newsrooms build scalable systems that make content genuinely findable: in traditional search engines, in chatbot answers, and inside the ecosystems governed by artificial intelligence.
In this interview we tackle questions that cut across the entire industry: the increasingly thin line between automation and human expertise, the value of originality in the era of generative models, the unauthorized use of content by platforms, the fragility of the metrics that claim to quantify “AI visibility” and the risk, raised by many studies, of a web impoverished by synthetic data and models training on one another.
You will find sharp opinions, at times even divisive ones, but also practical pointers for navigating an ecosystem that is changing at breathtaking speed.
Enjoy the read.
“The so-called new visibility is just the old visibility wearing a fake mustache”, Pedro Dias told us
If a growing share of SEO work is going to be absorbed by AI, which skills become truly decisive for staying useful to clients and not being replaced by automation?
I believe artificial intelligence will absorb the operational aspects of SEO rather than the strategic ones. Automation will handle tasks such as tool usage, data analysis and pattern identification. As a result, I think SEOs need to prioritize critical thinking and strategy, proving their value through innovation and creative problem-solving. Ultimately, SEO professionals must evolve into growth-oriented decision-makers who oversee operational execution.
You say that publishing too much know-how ends up feeding models that then compete with the people who create content. Given that many systems ignore blocks and keep harvesting data, what strategy can a professional adopt today to protect their value without losing online visibility?
I believe that if you publish everything you know for free, you will probably end up with a weak competitive advantage, if any at all. Artificial intelligence was born to meet specific problem-solving needs, and most of us use it to kick-start tasks quickly. However, as AI’s capabilities grow, so does its expertise. The more training data it consumes, the more capable it becomes.
SEOs have historically been great champions of open knowledge sharing. But in an industry driven by strategy and the need to stay ahead of the curve, that openness could lead to a race to the bottom, as artificial intelligence absorbs your best insights.
It is now up to each professional to weigh how much they are willing to invest in publishing and how deep their content should go. The goal is to strike a balance: reveal enough to showcase your expertise, but hold back enough to protect your competitive edge.
Cloudflare gives websites much more explicit control over how AI uses their content. Could this be the start of a real redefinition of the pact between creators and platforms, or does it remain a symbolic gesture until Google, OpenAI, Anthropic and the other Big Tech players clarify how they intend to move?
We urgently need a control mechanism to safeguard our interests. Every content producer, creator, artist and researcher should have a say in how and by whom their content is used.
The open web should not be a free-for-all archive where anyone can simply take and reuse content without agreement or consent. We went through a similar shift in the early 2000s, when search engines emerged, and we should have learned from that experience.
While the implementation of the robots.txt exclusion protocol was a step in the right direction, in reality it is like an open glass door with a guest list written on a post-it note: anyone can walk in without asking permission or being on the list.
Regardless of goodwill gestures or the data controls offered by tech companies, we need a system that gives creators unilateral, unquestioned control. Otherwise, I fear the incentive to produce content will fade away, leaving us with a significantly poorer web.
Cloudflare attributed its global outage to a single faulty query that sent its bot protection system into a tailspin. Is it an isolated incident, or does it reveal a structural fragility in the infrastructure a huge portion of the web relies on?
I believe any single point of failure in such critical infrastructure should be cause for serious concern. We have become heavily dependent on web services for both our work and our daily lives. Almost every appliance, electronic device and smart gadget connects to a CDN or AWS instance somewhere.
As a result, it is unacceptable for the entire ecosystem to grind to a halt because of a single point of failure. Just as we have multiple routes and modes of travel to choose from, we need to build digital redundancies to ensure the web keeps working even if a single service provider goes down.
Many tools promise to measure visibility in AI answers, but there are enormous doubts about personalization, context and the technical limits of the models. If these factors are not taken into account, what kind of reality do those numbers actually represent? And why, in your view, does part of the SEO community seem so reluctant to genuinely question the quality and reliability of the data these tools provide?
They represent a statistical hallucination. When tools ignore the non-deterministic nature of LLMs, along with the lack of access to user context, the numbers they produce are essentially “garbage”.
We are dealing with stochastic black boxes, not deterministic indexes. Therefore, these metrics do not represent a ‘ranking’, but rather a “lottery ticket”. A tool might tell you that you are visible in an answer, but without understanding the temperature, the seed and the specific context of the user, that data point is just noise dressed up as insight.
As for the lack of pushback from SEOs, I believe it is because the alternative is admitting that we are currently flying blind. The industry is drowning in a sea of illusions; agencies and professionals are under enormous pressure to prove the ROI of “AI strategies”.
Questioning the data would mean admitting that the “directional” charts they show clients are largely based on guesswork. It is easier to accept the “mirage” of exclusive AI visibility than to admit that there is currently no scientific way to measure it, especially when even the engineers building these models cannot fully explain the inference path.
Many of GEO’s “novelties” are nothing more than SEO principles that should have always applied, such as brand, authority and product quality. How do you explain, then, that part of the industry keeps chasing superficial metrics and selling old practices as if they were revolutions? And what risks does anyone building their strategy on this narrative run?
It is a classic case of “magical thinking” fueled by the frantic need to stay relevant. By slapping acronyms like GEO or AEO onto fundamental principles such as ‘clarity’ and “structure”, the industry manufactures an artificial battle to sell new services.
It suggests that LLMs have developed a sophisticated, unique taste for quality that Google somehow overlooked, when in reality optimization is still just about how machines ingest and understand data.
The risk is that people are building their strategy on “temperature and noise” rather than on sustainable foundations. If you believe you have cracked a secret “AI-only” variable because a chatbot turned your brand into a sentence, you are mistaking a probabilistic dice roll for a strategy.
The danger is that when you optimize for stochastic token prediction rather than for the deterministic foundation (RAG), the results are random. You have not outsmarted an algorithm, you have just gotten lucky in the inference chain. That is not a business model, it is gambling.
When it comes to tracking visibility in AI answers, many compare prompt personalization to the personalization of traditional search results. How valid is that equivalence really, and what role do the now-evident limits of keyword tracking and query-cluster-based systems play in this context?
It is a fundamental misunderstanding of the difference between an index and a generator. Comparing the two is technically wrong. Traditional search is deterministic retrieval; AI answers are probabilistic predictions.
The only predictable part of an AI answer is the “grounding” (RAG), which relies on the exact same retrieval mechanisms (indexing, vector search) as traditional search. There is no magical “AI retrieval layer” separate from technical reality.
The limitations confirm that most “AI tracking” is useless. Since we cannot see or extract an LLM’s context-based personalization, and since we have no verifiable data on how users actually query these systems, any tool claiming to track this is “drawing on guesswork”
If you are not indexable and parsable according to standard search principles, you cannot be used for grounding. Therefore, the “new” visibility is just the old retrieval visibility wearing a fake mustache. Until we have data that is not just “garbage” noise, we will keep treating stochastic parrots as oracles.
The expression stochastic parrot is used to describe an AI that repeats what it has learned from data without truly understanding it. The model analyzes huge amounts of text, identifies the most probable patterns and combines them to create new sentences. The end effect resembles a parrot mimicking the human voice: the words sound sensible, but there is no understanding behind them, only statistics (Ed.).
The study by Kaiser and Schulze shows that traffic from ChatGPT is almost invisible and converts very poorly, while Google continues to dominate the decision-making stage. How do you read such a stark gap between “understanding” and “buying” in user behavior?
This gap is no surprise: it is the architecture of the technology. Search engines work like transit hubs: their value lies in redirecting you toward a destination. LLMs work like destinations: their value lies in synthesizing the answer so you never have to leave.
The study confirms exactly what technical reality dictates: when an AI successfully uses grounding (RAG) to answer a query, it satisfies the intent in situ. The “low traffic” statistic is actually proof that the models are doing their job: extracting information and eliminating the need to click.
As for the conversion gap: users are not stupid. When money is at stake, we want deterministic certainty. We trust a browsable list of verified websites/vendors (Google) over a probabilistic recommendation from a chatbot that might be unreliable. We use AI to summarize the manual, but we use search to buy the car.
The EBU–BBC studies show that AI assistants get it wrong almost half of the time and base many answers on absent or misleading sources. How do you assess the impact of this unreliability on the visibility and reputation of businesses that depend on online search?
This study confirms what technical realists have been warning about since day one: the race for “AI visibility” is actually a gamble.
I recently shared a study highlighting how hallucinations are a structural feature of the design, not a bug (here is the link to the paper).
These models are not designed for truth, but for plausibility. By definition, they prioritize fluency over accuracy. So when the BBC finds high error rates, we are not necessarily witnessing a malfunction, but the system working exactly as designed.
If these models hallucinate 50% of the time, fighting to be included in their answers means you are statistically just as likely to be part of a lie as part of a fact.
For a brand, this is not just an accuracy problem, it is a serious brand safety crisis waiting to happen. We need to stop pretending LLMs are knowledge bases; they are probabilistic sentence completers. When a model invents a source or attributes a false claim to a company, it is not “making a mistake” in the human sense of the word, it is simply predicting the next plausible token based on its training weights.
For businesses, this creates a dangerous paradox: the industry is pushing them to optimize for a platform that could confidently destroy their reputation in the very next inference.
Until grounding mechanisms (RAG) are rigorously enforced and the hallucination rate drops from “coin flip” to “near zero”, building a strategy on AI visibility is not marketing — it has a high probability of turning into liability management. That reinforces why deterministic, trustworthy search remains the only safe environment for transactional and factual queries.
Academic research talks about “brain rot” in models trained on low-quality content, triggering a degradation loop that is hard to reverse. What kind of risks do you see for the entire information ecosystem if models keep feeding on data generated by other AIs?
I see this dynamic turning into a self-destroying prophecy, or what researchers call Model Collapse. We are building a “digital ouroboros”, the snake devouring its own tail. AI models are probabilistic engines that operate by converging toward the mean; they smooth out the outliers to generate “average”, plausible content.
Innovation, however, is born in the outliers. If the ecosystem gets flooded with uniform synthetic content, and models start training on that material, the quality of information slides into a loop of errors and mediocrity.
This process moves even faster because of the economic factor mentioned earlier: incentive. If content creators are not rewarded for novelty because their work is extracted without consent or attribution, they will simply stop publishing.
AI cannot invent: it can only remix.
If you remove the human incentive to produce the “new”, you cut off the fuel that powers these models. The likely outcome will be a poorer AI and a culture standing still.
AI exposes brands to real risks, and solid content is the way to defend yourself
The interview with Pedro Dias shed light on a truth the debate often prefers to sidestep: the real value of SEO in the age of artificial intelligence does not lie in the mechanical production of content, but in the ability to maintain critical thinking in an environment dominated by probabilistic systems.
His remarks on the risk of a “digital ouroboros” show what can happen if models keep feeding on what they themselves generate: a gradual homogenization of information, with innovation thinning out until it almost disappears.
It is not an inevitable fate, in my opinion, but a warning bell we cannot afford to ignore.
To avoid it, however, we will need clear incentives for the creation of authentic, original and verifiable content, because no model can replace the human intuition that generates new ideas.
The concept of the stochastic parrot, which Pedro brought up several times, gives us a genuinely interesting insight: the models’ lack of real understanding makes the production of plausible but not necessarily accurate sequences inevitable.
Without careful oversight, these systems tend to replicate statistical patterns that may look correct on the surface, but translate into errors, distortions or unfounded attributions. AI hallucinations show just how fragile the link between these models and factual reality is, with potentially serious consequences for the reputation of companies that get cited improperly.
From my point of view, investing in useful, solid, verifiable content created by professionals is not just a quality choice, but a safeguard for the entire information system.
It means preserving a pool of reliable knowledge to draw from and preventing the web from being saturated with synthetic material incapable of standing up to complex queries.
An ecosystem built on verified data benefits everyone: businesses, users, and even the platforms that feed on that content to generate better answers.
We thank Pedro Dias for the depth of his analysis and the clarity with which he outlined both challenges and opportunities.
See you at the next episode of SEO Confidential with another unmissable guest.
#avantitutta
