AI visibility, explained

The model is free now. Your catalogue is not.

Anthropic has published a blueprint for commerce agents, free and openly licensed. Shopify built the reference implementation for it. Anyone concluding that every shop now gets a shopping adviser has skipped the decisive line: the agent may only say what is in the catalogue.

Twenty websites, two hundred decoys: a study shows who really reads your content

A research team planted marked content and then asked 22 AI systems whether they knew it. The result refutes two assumptions that appear in almost every bot discussion: that blocking after the fact helps, and that the bot doing the reading works for the same company that later answers.

Five rounds of ChatGPT Ads: what we changed after every measurement

We ran the first ChatGPT ads for a small coffee shop and changed exactly one thing after every measurement. Five rounds, five adjustments, written down openly, including the two where the fault was ours. No verdict, a working state.

How ChatGPT picks products and merchants, and why that has nothing to do with ads

Since ChatGPT started showing ads inside its answers, online retail has been treating both as one thing. OpenAI does not: "Product results are selected independently by ChatGPT and are not ads, nor influenced by any OpenAI partnerships." Organic product results and ads are two separate systems with two different levers. This text walks through how the organic selection actually works, what merchants are ranked by, and why the most important point is not in the marketing but in your firewall.

ChatGPT Ads in Germany: how to get in before self-service opens

OpenAI's availability overview still lists Germany as "Coming Soon". That does not mean nobody can advertise. It means the route runs through the OpenAI Ads Solutions team or through agency and technology partners. Anyone taking that route is asked for the same technical things self-service will ask for later. This text sets out what actually applies today, what you have to sort out first, and where the limits run.

Ads in Google AI Mode: why your product feed now writes the ad

At Marketing Live 2026, Google introduced ads for AI Mode in Search. They work differently from classic Shopping ads and differently from ChatGPT Ads: there is no keyword campaign to bid on. Gemini builds the ad from your product feed when it finds a product that fits a customer's question. And on the Search side the same principle applies to your website: Google picks the landing page and writes the ad copy from what is on it. What that means for ecommerce, and where the difference to ChatGPT Ads lies.

Was kauft ein Werbetreibender, wenn kein Mensch die Anzeige je sieht? (Teil 2)

Ein Verlag verkauft Werbeplätze, die ausschließlich KI-Systeme zu sehen bekommen. Wer dort bucht, bekommt ein Format ohne redaktionelles Umfeld und ohne Kontrolle über die Auslieferung. Wir haben geprüft, was es leistet und was es kostet. Teil 2 unserer Messung. Stand: 14. August 2026.

Warum bekommt ein KI-Bot bei TIME einen anderen Artikel als ein Leser? (Teil 1)

TIME liefert KI-Systemen eigene Fassungen seiner Artikel, und darin steht bezahlte Werbung, die auf der Website für Menschen nicht existiert. Wir haben 30 Nachrichtenseiten mit sechs Abrufprofilen getestet, 40 US-Verlage auf Artikelebene geprüft und den entscheidenden Befund am 14. August unter kontrollierten Bedingungen nachgemessen.

ChatGPT Ads launch in Germany: the complete guide for companies

Since 24 August, ads appear in ChatGPT, here too. This is not search advertising with a new logo. Instead of individual keywords, the conversation context is at the center, and in the EEA a narrower frame applies on launch day than the platform's feature descriptions suggest. What companies can really do now, and what not yet.

Reddit loses 86 percent of its ChatGPT citations while a model with no maker takes the market

Two events from the same August week that come down to the same thing: with one source it is unclear why it is disappearing. With one model it is unclear who even runs it. Anyone basing their AI visibility on a tactic or a vendor name is building on sand.

AI search queries are three times as long. The advantage goes to whoever leads with the answer.

Google has measured how search is changing in AI Mode: the questions are on average three times as long as classic search terms, they arrive as conversations rather than single words, and increasingly with an image or a voice. Anyone still building pages for keywords will not be read here. Anyone who knows the question and starts with the answer will be.

Prompt Injection in AI Customer Chat: Why a Good System Prompt Is Not Enough

Many treat prompt injection as a wording problem that a better system prompt will fix. It will not. It is an architecture problem, and anyone running an AI advisor in their shop should understand why a filter alone does not hold the door shut.

An AI is not a search engine, which is why its demand cannot be derived from search volume

If you want to know how often people ask an AI about your topic, you will find numbers for it now. Almost all of them come from the same place: Google. We looked at how that derivation works, and found that we proposed the same idea ourselves in July. It is still sitting undecided with us, and a number of the same build is already showing in our own product.

Wer liest deinen Shop? Auf Shopify beantwortet das die Plattform, nicht du

Die Frage, die für KI-Sichtbarkeit zählt, lautet nicht „wie viel meines Verkehrs ist Bot". Sie lautet „welches KI-System hat welche Seite gelesen". Auf gehosteten Shop-Systemen kann der Betreiber diese Frage nicht selbst beantworten, weil ihm der Ort fehlt, an dem sie beantwortet würde. Wir haben für einen Shop nachgesehen, was dabei herauskommt, wenn man sich den Ort zurückholt.

Jemand benutzt die Namen der KI-Crawler

Vor sechs Wochen haben wir geschrieben, ein User-Agent sei eine Behauptung und kein Ausweis. Inzwischen haben wir gesehen, was passiert, wenn jemand diese Behauptung ausnutzt. Drei Befunde aus dem eigenen Anfrageweg, und keiner davon stand in einer Bot-Liste.

Was Maschinen von deiner Website lesen

Zwischen dem, was ein Kunde auf deiner Produktseite sieht, und dem, was ein automatisierter Abrufer davon bekommt, liegen oft mehrere Schichten: nachgeladener Text, mehrdeutiges Markup, strukturierte Daten ohne Preis. Dieser Text geht sechs Stellen durch, an denen der Unterschied entsteht, und sagt zu jeder, wie du sie mit Browser und Terminal selbst prüfst. Alle Aussagen beschreiben den Stand August 2026.

Warum KI-Assistenten bekannte Marken empfehlen und nicht nach neuen suchen

Die meisten Marken wollen von KI-Assistenten gefunden werden. Aber ein Assistent recherchiert keine neuen Marken: Er schlägt automatisch die vor, die er aus seinem Training kennt, und sucht dann gezielt bei genau diesen. Zwei aktuelle Untersuchungen zeigen, warum die meisten Sichtbarkeits-Maßnahmen daran nichts ändern.

Ein User-Agent ist eine Behauptung: das Bot-Kompendium

36 Bot-Kennungen von 16 Anbietern, sechs Klassen, und die unbequeme Wahrheit, dass ein User-Agent nur eine Behauptung ist. Stand 12. August 2026.

Wenn 96 Prozent der KI-Zitate an der Marke vorbeiführen: Was jetzt zu tun ist

Zweiter Teil zum Datenartikel „Wir haben 113.775 KI-Zitate gezählt". Teil 1 liefert die Befunde, dieser Teil die Maßnahmen.

SEO ist tot. Und niemand kann Ihnen KI-Sichtbarkeit verkaufen.

Dritter Beitrag der Serie. Teil 1 liefert die Daten, Teil 2 die Maßnahmen. Dieser Teil ist die Marktseite: Was Marken jetzt von Dienstleistern erwarten dürfen, und was nicht.

Wir haben 113.775 KI-Zitate gezählt. Nur vier von hundert führen zur Marke selbst.

Messstand 07.08.2026. Erste Auswertung einer fortlaufenden Messreihe, die Zahlen werden mit jedem weiteren Messlauf fortgeschrieben.

Runde 3 entscheidet: Warum deine Marke in der KI nur so sichtbar ist wie ihre Partner

Kaufentscheidungen reifen im KI-Dialog über mehrere Runden. Wer baut, aber nicht selbst verkauft, verliert die letzte Runde an ein Vermittlungsportal. Wir haben fünf Heizungshersteller geprüft und unsere eigenen Messdaten dagegengehalten.

Vorhanden heißt nicht brauchbar: die häufigsten Fehler in strukturierten Daten und was sie kosten

Eine Fehlerkunde aus echten Prüfungen. Was gültiges Markup trotzdem wertlos macht, was jeder einzelne Fehler kostet, und welche Reihenfolge sich beim Beheben bewährt hat.

The advisor who gets paid: when AI trust becomes ad space

The window is still open: why 89 percent of AI search demand belongs to nobody yet

When bots bend the rules: why AI crawlers became a governance question

A public dispute between Cloudflare and Perplexity shows that AI crawlers are no longer a neutral infrastructure detail. For brands and operators that raises a new question: how do I keep control over what agents do with my website, and over what they see at all?

From digital butler to thinking partner

How people really use AI agents, and why that hands brands a new infrastructure problem

Your headless CMS makes you structured. Not visible to AI.

Why a clean CMS setup isn’t AI visibility yet, and how to ship grounded, citable answer touchpoints straight from your CMS, without the auto-FAQ trap.

How AI Collects Data on Your Site, and What Traffic You Really See

Five kinds of AI traffic, the crawl-to-click ratio, and why last-click attribution makes AI’s contribution invisible. From the first line of markup to the attributed purchase.

Perfect Product Data for AI: How Deep Structured Data Puts Your Brand in the Answer

The basic schemas are becoming commodity. The edge is product data that answers real questions: dimensions, water-tank position, maintenance intervals. With full JSON-LD examples and a reference table. As of July 2026.

Cited Is Not Recommended: How the AI Journey Really Works

Being cited as a source is the door-opener, not the purchase decision. On the phases of the AI journey, why monitoring measures the wrong metric, and why long spoken prompts change the game.

Your CMS generates FAQs now. Why that won't save your AI visibility, and can even hurt it.

Modern enterprise CMS platforms turn your content into finished FAQs in minutes. But the answer ChatGPT and Perplexity give your customers depends on something else: whether it is sourced, reviewed and current.

IKEA appears 40% more often in AI answers. Do you know where you stand?

Four principles decide your visibility in ChatGPT, Claude, Perplexity and Gemini. The Harvard Business Manager showed them on IKEA. Here is how to apply them, without an enterprise data team, and with proof of what it earns you.

How to interview your target audience with AI, and what it actually reveals.

Proper market research costs money most small teams do not have. There is a workable middle way: interview an AI that plays your customer. Same seven questions every time, follow-ups until the vague parts break. Here is the exact script, two findings from real projects, and the mistake that quietly ruins the results.

The three phases of a buying visitor, and why trust is decided in phase one.

Every buying visitor goes through the same three phases: research, comparison, decision. Most websites spend their money back to front. The checkout gets polished while the first impression stays vague. That order is wrong, because trust is decided in phase one. And phase one, more and more often, doesn't happen on your website at all.

Why service providers and B2B companies need structured data even more than shops.

Structured data is nothing fancy: facts written down in a format machines can read. A shop's products at least exist that way: feeds, article numbers, prices. A consultancy's services usually exist only as marketing prose, and prose is the one format AI assistants cannot reliably quote. That makes agencies, consultancies and SaaS companies the most invisible businesses in AI answers, and the ones with the most to gain from fixing it.

How users try to break your AI assistant, and what protects you.

Put a chat widget in front of real visitors and a predictable thing happens: some of them stop asking about your products and start testing the assistant itself. No reason to stay offline. But you want to know the five patterns before they show up, and you want protection that keeps working even when the AI model behind the chat has a bad day.

How shops make their products readable for AI: structured data, step by step.

AI assistants and shopping agents do not admire your product photography. They read facts. The format they read most reliably is Schema.org markup in JSON-LD: a small data block in your page's source code that states name, price and availability in a way machines cannot misread. This guide shows what correct Product and FAQ markup looks like, which mistakes actively hurt, and what to do when you have 500 products instead of five.

Why most AI crawlers cannot read your JavaScript content.

Your product page looks complete in the browser. But GPTBot, ClaudeBot and PerplexityBot, the crawler programs that read websites for ChatGPT, Claude and Perplexity, never open a browser. They read the raw HTML your server sends, and whatever JavaScript builds on top of it simply does not exist for them. Here is how to see your site the way AI sees it, in about two minutes.

AI has to trust you before it recommends you.

When an AI assistant names a brand, it vouches for that brand with its own credibility. So before your company appears in an answer, the assistant runs the machine version of a background check. Four things decide the outcome: consistency, verifiability, readability and freshness. All four are yours to control.

How AI uses your own content against you.

You would expect an AI assistant to simply ignore marketing claims it cannot verify. It does something worse: it quotes them with distance. "The best agency in Germany" becomes "describes itself as the best agency in Germany", right next to a competitor who gets recommended with reasons. Here is the mechanism, how to test it on your own brand today, and how to fix it.

What AI actually cites: why solution content wins over problem content.

When an AI assistant answers a question, it has to build that answer out of something. Pages that describe problems, or praise their own company, give it nothing to build with. Pages that answer questions with structure, evidence and steps become the raw material. That difference decides who gets cited, in plain terms: whose website the AI names and links as its source.

What buyers actually compare before choosing an AI answer platform

An AI answer platform is software that answers customer questions with AI: on your own website, and ideally in assistants like ChatGPT too. Most comparison articles about these tools are feature grids written by one of the vendors. This is the other kind. We walk through the three options buyers actually shortlist, what each one is genuinely good at, and the four criteria the decision really turns on. Price comes after those four, on purpose.

How to build the internal business case for AI visibility

You are convinced that AI assistants like ChatGPT matter for your brand. Your CEO, your IT lead and your data protection officer are not convinced yet. A business case is nothing more than the internal argument for spending money on something. This guide is for the person who has to make that argument. The short version: measure first, answer each decision-maker's real concern, and ask for a two-week trial instead of a yearly contract.

The Monday inbox test: how to stop answering the same ten buying questions every week.

Open your inbox on a Monday morning and count how many messages are questions you have already answered before. Most of them are pre-sales questions: things people want to know before they buy, like shipping, sizing or returns. For many owners and small support teams these repeats quietly eat a realistic five to ten hours every week, spread across email, WhatsApp, chat and evenings. Here is the workflow that gets those hours back.

How to stop an AI assistant from making things up in front of customers.

When an AI invents a fact and delivers it with full confidence, that is called a hallucination. The fear of exactly that is the most common reason businesses keep AI away from paying customers, and the fear is rational. The cure just isn't a cleverer prompt: it is four structural properties you can check in any tool before you let it speak for you.

Der teuerste Trust-Fehler im Silicon Valley: Warum OpenAIs Ads-Wette gegen die eigenen Nutzerdaten läuft

Agentic commerce: when your next customer is an AI agent

Shoppers increasingly delegate research, comparison and even purchasing to AI assistants. Those assistants read your shop very differently from a human, and most shops are invisible to them. Here is what changes, and what to do about it.

Is AI visibility worth it for small and mid-size brands?

AI visibility tooling sounds like an enterprise game: dashboards, monitoring, another line item. But the underlying question, what do AI assistants tell buyers about us, hits smaller brands just as hard. An honest look at when measuring pays off and when it does not.

GDPR, DPAs and audit logs: using AI tools without legal headaches

AI tools do not get a compliance discount. The GDPR rules that apply to your CRM apply to your AI stack too. Here is what to check before you connect one to customer-facing workflows, and how Klariton approaches each point.

What is prompt monitoring? Measuring your brand inside AI answers.

More and more buying decisions start with a question to an AI assistant instead of a search box. Prompt monitoring is how you find out what those assistants say when the question touches your market, and whether your brand appears in the answer at all.

GEO vs. SEO: how visibility works when the answer replaces the ranking.

Search taught marketers to think in rankings: ten blue links, a position to climb, a click to win. Generative engines do not work that way. Here is what carries over, what breaks, and how you measure the difference.

What is a BIQ? Buying Intelligence Questions explained

A BIQ is a curated buying question with a source-backed answer: written from your own material, reviewed by a human before it goes live, and monitored after publishing. Here is what that means in practice.