From prompt tracking to off-page GEO: AI search according to Malte Landwehr
Today on SEO Confidential, the column where I talk with the world’s leading SEO and GEO experts, I’m bringing you an interview you truly can’t miss.
I spoke with Malte Landwehr, CPO and CMO of Peec AI, recognized by Business Insider as one of the fastest-growing companies in Europe in the AI search visibility space. Malte has over 20 years of SEO experience, has held roles as VP SEO and VP Product, and is now one of the most influential professionals in the industry, known for always backing up his AI search analysis with hard data.
The starting point is a Peec AI study from June 15, 2026, that analyzed how ChatGPT, Gemini, and Google AI Overviews actually decide what to show and to whom. Pay close attention, because the findings overturn some assumptions that many companies are still building their strategies on.
In the interview you’ll find concrete, no-frills answers to the questions you’re probably already asking yourself: why the intent behind a question matters far more than the exact words used, why measuring everything in clicks gives you a partial view of reality, how a practically unknown site managed to become a source ChatGPT cites with great frequency, and what you can do right now to increase the odds that your brand shows up in AI engine answers.
Whatever your industry and level of experience, what Malte told us affects you directly.
Enough with the pleasantries, then: let’s get started!
Getting cited by ChatGPT, Gemini, and AI Overviews: GEO and brand visibility, strategies and data for anyone who wants to matter in AI search
In the Peec AI study presented in the article “What Matters In An AI Prompt? Intent or Keywords?”, you argue that in the answers of AIs like ChatGPT, Gemini, or Google AI Overviews, the intent of the question matters more than the exact words used. If you had to explain it in simple terms to a business owner or to someone who isn’t an expert in SEO and artificial intelligence, what does it mean in concrete terms? And why is this finding so important today?
Traditional SEO worked with precise keywords and well-defined search volumes. For example, you could know that every day hundreds of people typed the exact search “AirPods Pro” into Google.
With AEO and GEO, the scenario changes completely. The prompts used in AI tools are much longer than traditional search queries and, in most cases, are worded in a unique way. The Peec AI study you mention demonstrated exactly that: the exact words in a prompt matter relatively little, and what really makes the difference is the user’s intent. If two people ask the same question with different wording, the AI still tends to interpret them as equivalent requests.
This finding is particularly significant because it confirms the usefulness of so-called prompt tracking: you don’t need to monitor every possible variant of a prompt, nor know exactly which phrases users type. It’s enough to identify and analyze the search intents behind their requests.
In fact, one of the most interesting findings is that questions worded in different ways often produce very similar results. But when does the result actually change? In which cases can a small change in the question bring completely different brands into the AI’s answers?
As long as the intent of the request stays the same, the results also tend to stay the same.
There are cases, however, where changing a single word is enough to radically change the intent of the prompt and, as a result, the answers the AI generates.
One example is searching for a rental car. If the prompt describes the desired vehicle in detail but you simply change the airport where you pick it up, the intent of the request changes substantially. As a consequence, the results will be different too.
The same phenomenon shows up in areas like skincare or fashion. Adding terms like “men,” “women,” or “kids” changes the search intent, because each category corresponds to different needs and products.
The user’s level of expertise also influences the results. Someone asking about “cables with advanced dielectric insulation and reduced skin-effect distortion” is expressing a much more technical intent than someone who simply asks for “a cable to connect the speakers.” Even though the topic is the same, the AI will interpret the two requests differently and suggest different brands, products, and content.
Your research shows that the style of the question matters too: a ranking, a list, or a more conversational request can change which brands are shown. What’s the most practical piece of advice you would give today to a company or a marketer who wants to figure out how to be more visible in the answers of ChatGPT, Gemini, or Google AI Overviews?
Even small differences in how a prompt is worded can influence the number and type of brands the AI mentions. For example, asking “what are the best CRMs?” often leads to an answer with several brands, while asking “what is the best CRM?” tends to return fewer names, focusing on a single solution or a few alternatives.
This aspect is particularly useful in prompt tracking. Lesser-known brands can use it to assess their visibility in answers generated by language models more precisely, while market leaders can get a broader overview of the competitors that are starting to show up more frequently.
From a practical standpoint, one of the most effective ways to increase your brand’s presence in AI answers is to describe it consistently across all your online channels. When the company name, products, services, and positioning are presented with clear, uniform messaging, language models can better understand who the brand is, what it does, and in which contexts it should be considered when generating answers.
Many companies still measure marketing mostly in clicks and traffic. But with AIs like ChatGPT, people often get an answer without visiting any website. How risky is it to keep reading the market with metrics designed for traditional SEO? And which signals should companies and marketers really start watching?
In traditional web search, click-through rate (CTR) generally hovered between 30% and 40%. In AI-based search, CTR often drops to somewhere between 1% and 5%. This means that evaluating AI search performance based solely on clicks leads you to significantly underestimate the importance and impact of this channel.
When users make their decisions directly within the conversation with a chatbot, what happens on your website is no longer enough to measure the effectiveness of your presence. It becomes necessary to rely on indirect metrics that better represent the real impact of AI. Among the most useful are:
- Self-reported attribution: asking new customers or new leads how they heard about the brand helps you understand whether the discovery happened through an AI assistant.
- Share of Voice via prompt tracking: measuring how often your brand is recommended by language models compared to your competitors gives you a concrete indicator of the visibility gained in AI-generated answers.
- Log file analysis: monitoring bot visits lets you identify which URLs on your site are used by AI models as reference sources (grounding) to build their answers.
AIs are changing how people discover brands, products, and services. What is the biggest mistake you see today from companies that still think of AI search as a simple evolution of traditional SEO?
Treating AI-based search as a simple evolution of traditional SEO can lead to several strategic mistakes.
The wrong KPIs. As we saw earlier, in SEO the number of clicks is one of the most important performance indicators. In AI search, however, this metric loses much of its value, because many users get their answer directly from the chatbot without visiting any website.
The wrong goals. In SEO, the main goal was to rank a web page among the top results on Google. In AI search, the equivalent is getting your URL cited in the answers generated by the language model. But even that shouldn’t be the real goal. The most important outcome is getting the AI to directly recommend your brand as an authoritative solution, regardless of whether a link is shown.
Underestimating off-page GEO is another critical mistake. SEO has always focused mainly on the website and on external activities aimed at earning backlinks. In AI search, however, much of the information used by language models also comes from external sources that have no direct connection to the company’s website.
A Reddit discussion that correctly describes a company’s pricing, an interview with the CEO, an in-depth review, or a citation in an authoritative publication can significantly influence how the brand is represented in AI answers, even without generating a single backlink. That’s why the reputation and consistency of the information spread across the web now play an even more important role than in the past.
Today many companies are investing more and more in AI-generated content to increase their online visibility. From your point of view, which signals distinguish a truly effective strategy from one that risks not working over the medium to long term?
I usually recommend asking yourself two questions before publishing AI-generated content.
- If you were a consumer, would you really visit your website to read an answer written by AI?
- If you were a product manager at Google or OpenAI, aiming to reduce processing costs while also avoiding showing users incorrect information, would you choose to crawl, index, and distribute the AI-generated content on your site?
If the answer to both questions is yes, you’re following a solid strategy.
If, on the other hand, the answer to one or both is no, you’re facing a much tougher challenge. How acceptable that choice is depends on your risk tolerance and the time horizon of your strategy. If the goal is short-term results, it can still turn out to be a worthwhile choice.
To better understand the context, consider one telling data point: a recent analysis of the customers and case studies of one of the leading automated content generation platforms found that 60% of the companies examined experienced a loss of visibility on Google.
In recent months there’s been more and more talk about MCP, the Model Context Protocol, as a system that allows artificial intelligence to connect directly to work tools and data. From your point of view, how much can it concretely change the day-to-day work of SEOs and marketers in the coming years?
I know SEO professionals who no longer open Excel or PowerPoint directly. They prefer to do everything through Claude Code, using various MCPs (Model Context Protocol).
For this reason, it’s fair to say that MCPs are already transforming how SEOs and marketers do their daily work. The main benefit is increased efficiency: they make it possible to automate many operations and streamline workflows.
One fundamental limitation remains, though: MCPs cannot replace human reasoning, hands-on experience, or the intuition needed to make strategic decisions. They can speed up the work and support analysis, but the ability to interpret data and choose the best direction still depends on people.
In your daily work, do you use tools or workflows based on MCP or similar systems to connect data and run analyses faster? Are there tasks you now finish in a few minutes thanks to AI that used to take you hours?
Absolutely. Instead of logging into two or three different tools to apply filters or start searches, I do all of that directly through Claude. I keep it open on my second monitor and only step in when it tells me my attention is needed.
Many SEO and marketing professionals talk a lot about ChatGPT and Perplexity, but much less about Google AI Overviews. In your opinion, are we underestimating the weight Google’s AI ecosystem could have in the coming years on traffic and on users’ purchasing decisions?
I think Perplexity still gets mentioned mostly out of habit. For a long time, people talked about “ChatGPT and Perplexity” as the main AI-powered answer engines. Today, though, the landscape has changed. Perplexity is no longer even among the five most-used platforms. ChatGPT, Gemini, Grok, Google’s AI Overviews, and Claude now have greater reach. Even Rufus, Amazon’s AI assistant, could soon overtake Perplexity in usage.
Yes, in my view, AI Overviews is the most underrated AI platform today. That’s probably because it doesn’t come in the form of a classic conversational interface. And yet, in terms of reach, AI Overviews reaches an enormously larger audience: its adoption is probably at least double that of ChatGPT.
There’s another language model many companies tend to overlook: Microsoft Copilot. For anyone operating in the enterprise market or working with public institutions and government agencies, Copilot is particularly important. Many large organizations only allow their employees to use this platform, banning access to other AI-based assistants.
If you were the owner of an SMB today that already invests in SEO but finds out it’s almost never cited in Google AI Overviews, where would you concretely start? What would be the first three practical actions you’d recommend to increase the odds of showing up in Google’s AI answers?
To start, it’s useful to run a few searches with the keywords or prompts you want your brand to be cited for in AI answers. Then analyze carefully which sources the model uses.
If the sources are external sites, the strategy is to try to be present on those platforms too. If they’re news outlets or editorial sites, it’s worth investing in digital PR. If the sources come from social networks, it makes sense to strengthen your presence with official channels or partner with creators and influencers. If, instead, the models reference review sites, it’s important to encourage satisfied customers to leave positive reviews, preferably with top scores.
When the main sources are your competitors’ content, consider whether you can create better content on the same topics. If you already have the right content, make sure it’s easy to cite. The simplest way to increase its “citability” is to add a short summary at the top of the page: two or three self-contained sentences, written in an authoritative, declarative tone, that explicitly mention the main entities.
Also make sure language models can actually crawl and render your website. Don’t hide your most important content behind JavaScript code!
On LinkedIn you told the story of GummySearch, a small, almost unknown site that managed to become a heavily cited source for ChatGPT by aggregating content from Reddit. What struck you most about this case, and what does it tell us, concretely, about how ChatGPT chooses the sources to show in its answers?
What struck me most was seeing that a website that had already announced its shutdown is still getting so many citations from ChatGPT.
The practical lesson is that once you understand what LLMs look for during the grounding process, it makes sense to offer exactly that kind of content and make it extremely clear to crawlers and agents what value and what information the page provides.
This case seems to suggest that even small or little-known players can become highly visible sources for ChatGPT. From your point of view, what practical lessons should a business owner draw from it today if they want to increase their brand’s presence in AI answers? Are there content characteristics that currently seem more likely to be considered trustworthy and citable?
The truly transferable aspect is that GummySearch created landing pages that present LLMs with exactly the information they look for during the grounding process, organized in a way that’s clear and easy for crawlers and AI agents to interpret.
The main lesson I take from it is that precisely indicating your sources, citing them explicitly, and explaining in detail where the information comes from can increase the value of your content in the eyes of LLMs too.
Make sure the AI understands who your company is!
As you’ve seen, Malte has the rare talent of bringing clarity to concepts the industry tends to complicate unnecessarily.
What struck me most is the consistency of the reasoning that runs through the entire interview. AI search demands a deep shift in mindset, even before a change of tools. Continuing to think in terms of keywords, rankings, and clicks means looking at the present through the eyes of the past.
And the market, as we well know, rewards those who manage to anticipate change.
The concept of “intent” as the true unit of measurement in AI search will, in my view, be a cornerstone of the discipline in the coming years. So will the idea that reputation distributed across the web, external mentions, interviews, and reviews now carry a strategic weight that goes well beyond the logic of backlinks.
Another point that keeps running through my head is the reflection on Google AI Overviews. A tool with probably twice the reach of ChatGPT, yet still underrated by most professionals in the industry.
Heartfelt thanks to Malte Landwehr for so generously making time for this and for the clarity with which he answered every question. A real, concrete conversation, with no beating around the bush, the kind that makes you want to open a document and start working right away.
I bet it will prove very useful in your daily work and inspire you to rethink your priorities.
See you next week on SEO Confidential with another GEO specialist!
