A conversation about the signals that make a website readable by machines, the limits of controlling hallucinations, and the need to get back to the fundamentals to keep AI from distorting your brand’s narrative
Organic traffic is collapsing and the rules that govern search are no longer the same. Today your potential customers no longer just scroll through Google’s results page: they query AI agents, LLMs and answer engines.
But what happens when these new artificial intelligences ignore your business or, even worse, “hallucinate”, making up information out of thin air that destroys your brand’s reputation and stalls your sales?
To tackle these topics with pragmatism and no false promises, in this episode of SEO Confidential my guest is Kristine Schachinger.
A world-class consultant specializing in Technical SEO and LLM Visibility Strategies, Kristine is also the voice of the authoritative podcast Webcology, where, together with Jim Hedger, she walks businesses and professionals through the real-world dynamics that make a web marketing project sell or fail.
In this interview we get straight to the point. Kristine reveals which structural mistakes push Google to “dump” your website, how to make your brand recommendable by AI agents (Agent-Ready), and why many of today’s “AI visibility” trackers are selling nothing but smoke.
Above all, we take a concrete look at how to defend your company’s reputation and leads from language model hallucinations.
Essential reading if you want to stop chasing the machines and start steering them.
“Search engines remain the primary source: SEO is still the foundation of everything”
If core updates really work like “street sweepers” that clear out websites that are expensive to crawl, which technical signals would you look at to understand whether a domain is becoming a “burden” for Google and risks being dumped by crawling agents?
I believe that’s the case, because when a site is hit hard by a core update, in most cases the problem is big, obvious and has been there for a long time, even if Google previously didn’t seem to give it much weight.
If these updates really work like “street sweepers” that remove sites that are too expensive to crawl, the technical signals to watch are the ones that point to persistent structural inefficiencies: widespread technical issues, messy architecture, crawling or rendering difficulties, elements that make the site complex for crawling agents to process.
The impact is significant because core updates affect all the major ranking signals. That’s why, in my approach, I always tackle the technical side first. Experience shows that by properly fixing the technical layer, you often get a recovery very close to 100% even before moving on to the next level of work.
Now that assistants are becoming “summarization engines” and pick the sources they synthesize, what makes a site more “ingestible” by agents than its competitors: technical quality, semantic structure, or external reputation, and how do you verify it without chasing visibility metrics that remain unreliable?
The process we call AI Search stems from RAG, a technique designed to “ground” the predictive text engine and reduce hallucinations. That’s where so-called AI Search developed from. There is no method that guarantees a presence in citations, unless Google integrates the Core Ranking Signals into AIO and AI Mode. The same indexes are already being used, but the list of citations we see is incomplete and random.
Some elements that can help are:
- a clear page structure (Hns);
- detailed structured data (schema);
- asking questions and providing answers (a function tied to natural language processing);
- correct use of entities;
- excellent technical SEO and little or no JavaScript in the content creation layer;
- an opening paragraph built on the inverted pyramid, because the context window of LLMs is limited.
Agents don’t “rank” — they select documents based on quality and readability. To understand which technical fixes have the fastest impact on your chances of being chosen as a source, I recommend reading this fascinating study by SparkToro. I think it can prove very useful!
You said that “LLM visibility” trackers risk telling a story, because citations and sources change even with the same prompt: how would you set up a serious “agent readiness” audit that a company can repeat every month with verifiable, non-narrative indicators?
Agent readiness rests on the factors I’ve already mentioned. There can be no guarantee of citation until the ranking signals are integrated into Google’s LLMs.
Keep in mind that these sources generate very little traffic. Focusing on this at the expense of traditional SEO is counterproductive, because search engines remain the primary source of citations and Google still sends almost all organic traffic. SEO is still the foundation of everything, with small adjustments in priorities.
From your point of view, what is the most underrated technical signal that gets a site excluded from agents’ summarization flows today, even when the content is correct: incomplete rendering, crawling costs, or a confusing internal architecture?
To be interpreted correctly by machine learning parsers (software that reads the code of a web page and interprets its structure to make it understandable to search engines and AI systems, Ed.), pages must be technically clean, with correct, well-structured HTML markup.
It is essential to keep the use of JavaScript to a minimum, especially in the core content elements, because everything that is essential should be immediately readable and processable by machines without depending on complex rendering.
If MCP becomes the standard and AI agents stop “reading” websites the way a human does, what are the first technical signals showing that a brand is truly agent-ready, and not just “in good shape” in traditional SEO terms?
AI agents already don’t read websites the way a person would: they analyze structure, signals and data in an automated way. Rather than establishing a rigid hierarchy of technical signals, what matters is that the site is fully understandable and processable by machines, with a clear structure and clean code.
Digital PR remains decisive in any case, because external presence and authority significantly affect the likelihood of being cited.
In the era of “Personal Intelligence“, where the answer changes based on history, email and context, what is the most serious way to measure a brand’s real visibility without selling “share of voice” as smoke: which signals should we watch, how often, and how do we tie them to leads and sales?
You cannot accurately track elements that don’t appear in a complete, stable list of citations, not least because many of them aren’t visible. There are black hat techniques that can increase mentions inside an LLM, but to earn a genuine citation it is essential that the page content be tightly coherent and relevant to the query.
In organic search this mechanism is known as neural matching and it is one of the factors that determine the final ordering of results.
From an operational standpoint, referral tracking becomes central. Google Search Console can offer some useful clues to reconstruct part of the traffic flows, while log file analysis makes it possible to understand more accurately who is accessing the site and how often.
LLMs and answer engines risk creating a kind of parallel brand, confusing the public and undermining brand reputation. What can we do to defend ourselves and prevent AI from describing us in an incorrect, “hallucinated” way?
No one can eliminate hallucinations entirely, because they stem from the mathematical structure language models are built on. The scientific literature explains it clearly: these systems generate text predictively, they don’t understand what they write, but they detect patterns in the training data and return the statistically most probable word sequences.
You can only partially influence this behavior, for example by adjusting parameters such as temperature, but only if you have direct control over the model. In all other cases, the room for intervention remains very limited.
Which signals increase the odds that a correct version prevails over misinformation: markup and structured data, consistency across pages, frequent updates, off-site authority, or presence on the platforms where narratives are born (e.g. forums and social media)? In practice, how do you design an anti-hallucination defense that is measurable and repeatable?
As we know, there is no method that can completely eliminate hallucinations. You can, however, reduce their incidence through interventions such as targeted training, the integration of RAG systems and the introduction of guardrails, but only if you have direct control over the model.
Otherwise, the room for intervention is very limited, because the system generates answers based on the data it was trained on and the statistical probabilities that derive from it.
You can try to influence the outcome indirectly by publishing content on public sources known to be used in training processes, or by creating evergreen content picked up by LLMs not connected to search engines. This strategy, however, produces no concrete effects on ecosystems like Google or Bing.
The illusion of control and the harsh reality: how to bulletproof your brand in the era of AI agents
Wrapping up this conversation with Kristine Schachinger, a crystal-clear picture emerges: the race to shelter from Artificial Intelligence is not won by chasing new vanity metrics, but by going back to the technical fundamentals with almost ruthless rigor.
Clean code, a drastic reduction in JavaScript and a machine-proof information architecture.
The plain, hard truth is that today chasing “LLM visibility” hoping for some secret hack is a strategic mistake that only burns budget. Machine hallucinations are a frightening, concrete risk for any brand’s reputation and sales, but our defense does not run through tracking tools that often sell nothing but illusions.
True Agent Readiness is engineering: answer engines don’t grasp nuance, they process structures. If you want AI to become an ally that recommends your business rather than a misinformation generator, you have to stop feeding the hype and give your digital ecosystem technically unassailable foundations and unquestionable external authority.
My heartfelt thanks to Kristine for sharing her expertise and sweeping away much of the hot air circulating in the industry today.
See you in the next episode of SEO Confidential.
Keep following us: our journey into the gears of search continues.
