Visibility, citations, sentiment and new criteria of authority: Jules de Bruin tells us how search is changing in the AI era and which strategies can make the difference
Welcome back to SEO Confidential. Today we are publishing an episode that arrives at just the right moment, on a topic the industry is still trying to bring into sharp focus.
I spoke with Jules de Bruin, Chief of Staff at Rankscale AI, one of Europe’s most advanced platforms for measuring brand visibility in AI-powered answer engines.
Jules works every day in close contact with companies that truly want to understand how they are represented by ChatGPT, Gemini, Claude, Google AI Overviews and the new AI Mode. What he shared in this interview comes straight from the field: real data, concrete cases, mistakes that keep repeating themselves and a few truths the SEO (and GEO) community still struggles to accept.
The common thread is a very simple question: does your brand actually exist in AI Search, or is it invisible without you even knowing it?
In the lines that follow you will find precise answers on how the logic of visibility changes when the one deciding who gets cited is an artificial intelligence rather than a traditional ranking algorithm.
Jules told us why monitoring only ChatGPT is already an outdated strategy, what sentiment in AI answers means and why it can do more damage than a negative review, how to build content that language models choose as a source, and which metric almost everyone ignores even though it tells you more than any Google ranking.
There is one sentence that has been going around in my head ever since I read his answers: the AI answer is your new homepage. Okay, put like that it sounds a bit like a provocation, but if you read the interview you will understand that this statement of his is not so crazy after all.
If you already know what answer engines are, this interview will give you the tools to act. If you have not thought about it yet, now is the right time to do so.
GEO is exposing your brand’s weak spots
We are witnessing the rise of a new form of competition in which winning the first position on Google no longer matters; what matters is becoming the product recommended by artificial intelligences. What consequences will this shift have for brands that keep investing exclusively in traditional SEO?
Brands that invest exclusively in traditional SEO will lose visibility in the AI space, but not because SEO has stopped being important. AI search rests on three tightly connected factors: technical SEO, digital PR (positive mentions) and content. Agencies that specialize exclusively in SEO usually optimize just one element and neglect the other two.
The underlying mechanism has changed. Authoritative link building is turning into mention building: your brand needs to be cited on the specific third-party pages that every AI search engine already trusts and draws information from. Technical SEO remains fundamental, and the standards are rising. Teams that have already done SEO properly (a solid technical foundation, high-quality content, authentic PR) are in a privileged position. The ones at risk are those who defined link volume as “SEO” and left out the rest. GEO demands that all three factors work in unison.
There is an analogy I like in this context. For years we all swam in the same pool and everything seemed fine. Authority was the water. As AI search drains part of that water away, you start to see who has been swimming naked the whole time, that is, who was propped up by authority alone and never truly invested in the fundamentals.
Until recently, much of the industry focused on monitoring visibility in ChatGPT. At Rankscale you argue that this approach is now insufficient, and you have extended tracking to Google AI Mode and Microsoft Copilot Shopping. Does this mean that many companies are still measuring the phenomenon with tools that are already outdated? What are the most common mistakes you are observing?
Focusing exclusively on ChatGPT is a mistake today, even though two years ago it was the right choice, back when ChatGPT was the dominant platform for AI search and concentrating resources on it made sense.
ChatGPT remains the largest AI search platform overall, but it no longer holds a near-monopoly: Google AI Overviews and AI Mode, Perplexity, Copilot, Gemini and Claude have fragmented usage, and the trend is moving away from ChatGPT, not toward it. Different audiences now use different engines, and each engine has its own retrieval preferences.
So the real question is not “am I monitoring everything?” but “which engines does my audience actually use?”
A practical guide to get started:
Audience Track these first Enterprise ChatGPT, Claude, Copilot SMBs ChatGPT, Gemini, Google AI Mode Startups / developer tools ChatGPT, Claude, Gemini (+ Grok for the US) Consumers ChatGPT, Google AI Overviews and AI Mode Marketing professionals Add Perplexity Europe (especially France) Add Mistral Asia Prioritize DeepSeek Beyond that, three recurring mistakes:
- Monitoring everything, or only ChatGPT. Both strategies are wrong. Match the engines to your audience.
- Branded prompts. Including your own name inflates the score. Of course you rank for yourself, but that has always been true on Google as well. The real challenge lies in unbranded, solution-seeking prompts, where nobody names a brand. That is where buyers actually discover you.
- Black-box dependency. Many tools hand you a score without revealing how it was calculated, which merely replaces one dependency with another. They also tend to give generic recommendations, which makes little sense when they know nothing about your priorities. At Rankscale we show the data behind the number, so teams can build AI search capabilities in-house instead of outsourcing the understanding, and we provide recommendations for every prompt and every AI engine, aligned with what actually matters for your business.
Many business owners still keep checking where their site ranks on Google. With AI Mode, though, that may no longer be enough. Which new metrics should they start monitoring to understand whether their brand is really visible in AI answers?
Your Google ranking says virtually nothing about whether the AI recommends you or not. Beyond mere presence, two factors matter: position within the result and sentiment. Appearing in fifth place is not the same as appearing first, and appearing in a negative light is worse than not appearing at all.
Sentiment is the metric most tools ignore. If you show up in an answer and the context is negative, that is not a win: the engine is actively steering buyers away from you. It is important to be able to trace negative sentiment back to its source, to understand which pages on the internet are responsible for it. Unfortunately, we see a lot of false information being spread by competitors about one another. We monitor prompts about our own Rankscale brand for exactly this purpose.
In general, at Rankscale we treat sentiment in calibration bands (roughly: above about 65%, the engine promotes you like a sales ambassador; below about 55%, it behaves like a disgruntled former customer warning prospects away). These thresholds are our internal calibration, not a universal constant. The point is directional: a high detection rate paired with negative sentiment is an alarm bell, not a trophy. A 90% detection rate where you are consistently cited as the bad example is a problem dressed up as a success.
That is why a true AI visibility score has to weight detection by position and sentiment, not simply count mentions.
Today much content is still written mainly with Google and users in mind. If the goal instead is to become a source cited by ChatGPT, Gemini, Claude or Perplexity, how does the way you write a page change? Which characteristics can no longer be missing?
Five rules. Every article and blog post on your site should satisfy all five:
- BLUF (bottom line up front). A direct 40-80 word answer in the first paragraph, including the current month and year.
- Question-style headings. Every H2 phrased as a question a user might actually type.
- Justification attributes. Explicit statements like “best for X”, comparison tables, clear pros and cons. AI search engines look for reasons that justify recommending your site.
- Chunked structure (chunkability). Every paragraph stands on its own, with a single topic per paragraph, so it can be understood without reading the rest of the page.
- Evidence and data. Statistics, dates and cited sources. AI search engines want certainty. Evidence and data provide it.
For European companies there is a sixth, non-negotiable rule: publish an English version of your content. AI engines still strongly favor English-language sources. Without an English version you might not be found at all — not just ranked at the bottom, but simply absent.
There is more to say (these engines follow token economics, a topic in its own right), but a page that satisfies these five criteria is already structured to be easily retrieved.
We often talk about authority, verifiable data and up-to-date content, but many business owners struggle to understand what that means in practice. Which elements truly make the difference in getting an artificial intelligence to consider a page trustworthy and decide to use it in its answers?
Specificity, freshness and provenance, in that order.
“Up to date” means a clear freshness signal: state the month and year (for example, July 2026) and keep it current. “Trustworthy” means facts, not adjectives. “The best solution for cutting costs” gives the engine no extractable data; there is no encoded reason for it to recommend you. “We reduced processing costs at this specific stage by 37%, here is the case study” provides a citable, attributable data point.
The general rule: if a claim cannot be backed by a number, a name or a date, it will not get cited. Replace marketing adjectives with statistics whenever you can.
Many companies invest time and resources in producing content, yet that content is not always used by artificial intelligences. What are the most frequent mistakes that prevent a page from being retrieved and cited by AI Search systems, and how can they be avoided?
Three common mistakes, in order of frequency:
- Blocking crawlers. The engine reaches your site and runs into a barrier it cannot open, usually a WAF rule that silently blocks AI user agents. Check your WAF configuration and your robots.txt file for the user agents ChatGPT-User, OAI-SearchBot, GPTBot, ClaudeBot, PerplexityBot, CCBot and Google-Extended.
- Content missing from the HTML or loading too slowly. If the text is generated client-side in JavaScript, or if the page is slow, the engine sees an empty page and moves on (since the user is waiting for an answer). Everything that matters must be present in the served HTML and load quickly.
- Measuring a prompt only once. AI answers are probabilistic, so no two runs are identical. Reading a single result as “your score” for that prompt is noise, not signal.
In a recent article of yours, you talked about “AI Grounding Pages“, arguing that some pages should no longer be designed to earn clicks, but to become the source from which ChatGPT, Google AI Mode and the other AI systems retrieve information. What exactly are these pages, and which characteristics should they have to be considered trustworthy by artificial intelligences?
An AI “Grounding Page” is designed to be retrieved, not clicked. Its purpose is to ensure that when an engine looks for information about your brand, product or category, it finds a complete and accurate source — and that this source is yours.
Think of these pages as “About us” pages, but dedicated to every entity you care about: every product, every service, every regulation that affects your buyers. They feature an extremely high information density (explicit numerical data, dates, regulatory bodies, prices, named entities, attributions) and almost no filler. Few people will read them from start to finish, and that is fine: they are a machine-readable knowledge hub.
The contrast is stark. A homepage reads: “Rankscale helps companies unlock the potential of AI search”, but it contains no extractable data.
An AI Grounding Page instead reads: “Rankscale is an Austrian SaaS platform, operated by Rankscale GmbH in Vienna, that monitors a brand’s AI visibility across more than 17 generative AI engines without a paywall and provides prompt-level recommendations to improve AI visibility for each prompt in a specific AI search engine”. Same brand, seven extractable facts instead of none. That density is exactly what information retrieval targets.
An honest caveat, because it is the objection a careful reader will raise: AI engines tend to favor so-called “earned media” — articles, citations and mentions obtained on authoritative sites — over content published directly by a company. A grounding page on your own domain does not, by itself, create that external authority. What it does is ensure that, once an engine finds you, it interprets you correctly.
So follow the right sequence: your site’s reference pages support the work of digital PR and citations on authoritative sites, but they cannot replace it. First build your authority externally, then give the engine a clear, complete page on your own site.
(Rankscale AI works with Hanns Kronenberg, the pioneer behind “grounding pages”, so the topic is very close to our hearts, as you can see.)
If a company fails to give artificial intelligence clear, complete and verifiable information, it risks being described incorrectly, confused with a competitor or even excluded from generated answers. How widespread are these problems today, and what should a company concretely do to avoid them?
It is a common situation, and the solution is simple: create AI reference pages that describe your brand, your company and your products in depth, so that when an engine is prompted to learn more, the source is already available.
When that source is missing, an engine does one of two things. Either it “hallucinates”, filling the gap with plausible but incorrect claims and misrepresenting you (often drawing its interpretation from a competitor’s comparison page, which is exactly where many of the strange claims we see on Rankscale come from). Or it ignores you, staying silent because it is not confident enough to include you, and you are simply absent from the answer.
Both of these outcomes are avoidable if you own the source directly, instead of letting the engine derive it on its own.
There are currently many tools that promise to measure visibility in AI answers, but generative models produce results that change constantly. How much can these metrics really be trusted, and what are their main limitations?
Trust should be proportional to sample size, and most people place far too much trust in limited data.
A single tracked run is just noise: it could be a lucky or unlucky outcome, and either way you would end up reading variance as signal. Seven daily runs are the practical minimum for understanding whether a prompt is stable or volatile. This approach reveals the pattern of variance, not a precise score; so treat the weekly figure as indicative, not exact.
For any project you are investing resources in, it would be better to widen the time window or add more days of prompt runs. Be wary of anyone handing you a “visibility score” with decimal places based on just seven samples: that is false precision, the same mistake as trusting a single run, only disguised.
Beyond the score itself, keep an eye on the cited-sources metric: how often a specific page was actually used as the source of an answer. That is a clearer indication of which content is performing than any headline number.
Google Search Console is showing more and more data about presence in AI answers, but it still does not say how many users actually click through to the cited sites. Is that a strategic choice? And how much does it weigh on companies and professionals not to have access to what is probably the most important metric?
It is a rational decision on Google’s part, and it hides an uncomfortable truth: Google now holds absolute control over the discovery layer.
In the past, you at least got a cookie and a click — a relationship you could later reactivate through ads.
In AI-generated results, the relationship lives inside the answer itself, and I would go so far as to say that the AI answer is your new homepage, the first meaningful touchpoint a customer has with your brand. From my personal tests, the click-through rate sits well below 1%. If Google made that figure public, people would start asking out loud what the actual value of organic discovery is.
That is why this is not a reason to retreat from search, but the strongest reason of all to double down on it. The click was never the prize; it was an indicator of being chosen. That indicator is disappearing, but the choice is still being made — now inside the AI answer — and at the moment almost nobody is optimizing for it. The channel that mattered for twenty years is closing just as a new one opens, practically free of competition. Being the source an engine cites costs little today, but it will not stay that way for long. Whoever claims that position first will be served up as the default answer, while competitors keep counting clicks.
The strategy, then, is obvious: stop chasing the click you can no longer win and start winning the citation instead. Turn your website into a knowledge base that provides the answers, become the source the engine draws from, and you will be the first thing the buyer sees, ahead of everyone still optimizing for a homepage nobody visits. If you cannot win the click, win the citation — on your own pages and on the third-party sites the engines trust — and get there before everyone else.
For anyone who wants to dig deeper into the topic, every two months we run a one-day AI Search training Bootcamp, which you can sign up for here.
If you cannot win the click, win the citation!
We reach the end of this conversation with something concrete in our hands, and that, as you know, is far from a given when it comes to AI Search.
What struck me most in Jules’s words is the precision with which he dismantles some certainties the industry keeps repeating as if they were mantras. Monitoring only ChatGPT, counting mentions without looking at sentiment, publishing content without checking that AI crawlers can actually read it: these are mistakes many companies are still making today, in good faith, simply because nobody has told them clearly yet.
The concept of sentiment as a strategic metric is the one that matters most to me. Appearing in AI answers in a negative context is worse than not appearing at all. And yet almost nobody measures it.
I also find the distinction between grounding pages and traditional content very useful. Building pages designed to be retrieved by language models, dense with facts, data and verifiable attributes, is a shift in perspective that requires little technical effort but a completely different mindset from the one used to write content for Google.
And then there is that sentence: the AI answer is your new homepage. The more I think about it, the less it sounds like a quip and the more it reads as an accurate description of what is already happening.
Heartfelt thanks to Jules de Bruin for the quality and concreteness of every answer. A conversation worth reading and rereading.
See you next week, as always, on SEO Confidential. See you soon.
#fullsteamahead
