The Chatbot Has Never Heard of You: What 135 AI Experiments Reveal About Getting Cited by ChatGPT, Claude, and Gemini

Your customers stopped Googling you. They're asking ChatGPT, Claude, and Gemini instead — "what's the best clean moisturizer for sensitive skin?" — and the AI answers in full sentences, with brand recommendations. If your brand isn't in that answer, you're invisible in the fastest-growing discovery channel in beauty. This episode explains why — with real experimental data, not guru guesswork.

In this episode, Diego Lomnitzer Lapetina (PharmD, MSc, PhD), co-founder of Atomic Pom Labs, breaks down the first controlled, multi-platform, multi-interface AI citation experiment: 15 standardized buyer queries submitted to ChatGPT, Claude, and Gemini across three access points each — mobile app, web browser, and API — for a total of 135 experimental runs and over 750 individual citation events. Every cited domain was then cross-referenced against real SEO metrics (Domain Rating, referring domains, organic traffic) to answer the question every founder is asking in 2026: what actually makes an AI recommend a brand?

If you've been hearing about Generative Engine Optimization (GEO), AI SEO, LLM optimization, AI search visibility, or "how to rank in ChatGPT" — this is the episode that separates measurable reality from marketing hype.

WHAT YOU'LL LEARN IN THIS EPISODE

Part 1 — One Question, Nine Answers How the experiment was designed: 15 queries across five intent categories (commercial, informational, comparison, brand recommendation, technical), three AI platforms, three interfaces, clean sessions, and full SEO cross-referencing via Ahrefs. Why nobody had ever measured AI citation behavior this way before — and why averages across millions of queries hide what actually happens when a real customer asks a real question.

Part 2 — The Empty Middle The headline finding: when you ask ChatGPT, Claude, and Gemini the exact same question, they cite almost none of the same sources. Cross-platform overlap came in under 5% — and across all fifteen queries, only two websites were cited by all three platforms. What this means for founders: there is no "AI ranking." There is no position 1 to win. AI citation is probabilistic, not positional — and that changes every strategy built on old SEO thinking. If an agency promises to "rank you in AI search," they're selling a map of a place that doesn't exist.

Part 3 — The Back Door The most surprising discovery in the dataset: the two-tier citation system. The web versions of these AI platforms behave like gatekeepers, almost never citing domains below a Domain Rating of 85 — Sephora territory, Byrdie territory, the fortress of established beauty publishers. But the API — the plumbing that powers shopping assistants, skincare routine builders, and AI recommendation tools — played by completely different rules, repeatedly citing a website with a Domain Rating of 4.5 and roughly twelve visitors a month. Why this "API authority bypass" is the single biggest opportunity for small and indie brands in AI discovery.

Part 4 — Three Machines, Three Personalities Each AI platform has a distinct citation personality, confirmed by the data: Claude behaves like a researcher (favoring vendor documentation and technical guides — your own product pages and ingredient explainers), ChatGPT behaves like a librarian (favoring guide-style content, tutorials, and roundups on established sites), and Gemini behaves like a journalist (favoring media coverage from publications like Forbes, PCMag, and TechRadar). How to match your content strategy — product documentation, educational guides, or PR — to the platform your customers actually use.

Part 5 — The Indie Playbook Concrete, tiered recommendations straight from the data. For established brands (DR 85+): structure your content for passage-level extraction — clear headings, self-contained paragraphs, embedded statistics. For mid-authority brands (DR 50–85): stop chasing all three platforms and optimize for one platform's personality. For indie and emerging brands (DR under 50): stop fighting the corpus and change the question — win through query specificity and category creation. "Best moisturizer" has ten thousand answers; a hyper-specific query has three. Be one of them — or better, be the only one. Plus the one non-negotiable warning: anyone selling guaranteed AI citations is selling something the data proves does not exist.

MEMORABLE LINES FROM THIS EPISODE

"AI citation isn't a position you hold. It's a probability you raise." "The front door checks your credentials. The back door checks whether your content answers the exact question." "The machines don't retrieve prestige. They retrieve structure." "Don't fight the corpus. Change the question." "The fortress has a service entrance. And it's unguarded."

WHO THIS EPISODE IS FOR

Indie beauty brand founders, skincare and cosmetics entrepreneurs, DTC and e-commerce operators, brand strategists, content marketers, SEO professionals transitioning into GEO, agency owners advising beauty and CPG clients, and anyone trying to understand how AI platforms like ChatGPT, Claude, and Gemini choose which brands, products, and websites to cite and recommend.

KEY TOPICS AND QUESTIONS COVERED

• What is Generative Engine Optimization (GEO) and how is it different from SEO?
• How do ChatGPT, Claude, and Gemini decide which sources to cite?
• Why do AI platforms recommend some brands and ignore others?
• Does Domain Rating (DR) affect AI citations? (Yes — but it predicts the floor, not the ceiling)
• What is the DR 85 threshold effect in AI search?
• Why API-based AI tools cite low-authority websites the web interface never would
• Cross-platform citation overlap: why it's under 5% at the query level
• Platform personalities: Claude as researcher, ChatGPT as librarian, Gemini as journalist
• The three content archetypes AI retrieval systems prefer: best-of roundups, product documentation, step-by-step frameworks
• Why vendor self-citation dominates commercial and comparison queries
• Query specificity and category creation as the indie brand strategy for AI visibility
• How small beauty brands can get recommended by AI without a massive SEO budget
• Why "guaranteed AI citations" is a red flag — and what to invest in instead
• The future of AI search optimization for beauty, skincare, and consumer brands

ABOUT THE RESEARCH

This episode is based on "Citation Divergence Across AI Platforms: A Multi-Interface Empirical Analysis of Source Selection in Claude, ChatGPT, and Gemini" (Atomic Pom Labs, April 2026) — a controlled study of 135 experimental runs conducted under a reproducible protocol, with all cited domains cross-referenced against Ahrefs SEO data. The study builds on and extends prior GEO research including the Princeton GEO study (Aggarwal et al., ACM KDD 2024), Profound's large-scale citation analysis, Semrush's most-cited domains research, and the Writesonic LLM Citation Study.

ABOUT ATOMIC POM LABS

Atomic Pom Labs (APL) is a sensory branding and cosmetic development consultancy helping indie beauty brands engineer the cognitive architecture behind memorable brands — from formulation and regulatory strategy to brand psychology and AI-era visibility. Built on the Cognitive Branding Framework (CBF), a seven-phase system developed over five years at the intersection of behavioral science, philosophy, and brand strategy.

If this episode reframed how you think about AI visibility, share it with a founder who's still buying "AI ranking" packages — and save it, because the indie playbook in Part 5 is one you'll come back to before your next content sprint.

Subscribe for more evidence-based brand strategy for indie beauty founders: branding psychology, cosmetic formulation, regulatory compliance, pricing, packaging, and the new science of getting discovered in an AI-first world.

#GEO #GenerativeEngineOptimization #AISearch #AISEO #ChatGPT #Claude #Gemini #AICitations #IndieBeauty #BeautyBrand #SkincareBrand #BrandStrategy #SEO2026 #AIMarketing #DTCBrands #CosmeticsBusiness #BeautyFounder #LLMOptimization #AIVisibility #ContentStrategy #AtomicPomLabs

Two Geeks at a Bench

Transcript

SPEAKER_00 0:00

Your customers stop Googling you. They're asking a chatbot instead. What's the best clean moisturizer for sensitive skin? And the chatbot answers confidently in full sentences with recommendations. And your brand isn't in the answer. Not because the machine rejected you, because it has never heard of you. Here's the part nobody tells you. The people selling you AI visibility haven't heard of you either. They're guessing. So I stopped guessing and ran an experiment. 135 controlled runs across Cloud, ChatGPT, and Gemini. Same questions, three platforms, three ways of accessing each one. Then I pulled the SEO profile of every domain those machines cited and looked at what actually gets a brain mention. Some of what I found is bad news for small brands. Some of it is the best news you'll get all year. Stay with me. Part one, one question, nine answers. First, the setup because the method is the whole point. I took 15 questions, the kind real buyers ask. Best of questions. How to questions? Compare this to that questions. Direct what brand should I buy? Questions. Then I asked every one of them to Claude, to ChatGPT, and to Gemini. And here's the twist. I asked each platform three different ways. Through the mobile app, through the web browser, and through the API, which is the plumbing that other apps use when they build on top of these models. Three platforms, three doors into each, nine combinations, 15 questions, 135 runs, over 750 citations, every session in a fresh chat, so nothing leaked between runs. Then I cross-referenced every cited domain against real SEO data, domain authority, backlinks, traffic. Why go through all that? Because everyone measuring AI citations measures averages across millions of queries. Nobody had asked the simple question, if I ask the same thing twice through two doors, do I get the same sources? The answer changed how I think about this entire game. Part 2. The empty middle. Across 15 questions and hundreds of citations, only two websites in the entire experiment were cited by all three platforms. Think about what that means. There is no AI ranking. There's no leaderboard where you climb from position 8 to position 3. Each platform is pulling from a nearly separate pool of sources. And even the same platform asks the same question minutes apart through a different door, gives you a different set. So when someone offers to rank you in AI search the way they ranked you in Google, they're selling a map of a place that doesn't exist. AI citation isn't a position you hold, it's a probability you raise. That sounds like bad news. It isn't. Because if there's no throne, nobody's sitting on it. The big players can't lock up a position that doesn't exist either. Part three, the backdoor. Now the finding that made me sit up. On the web interface, these platforms are snobs. In my data, the web versions almost never cited a domain below a domain rating of 85. That's Sephora territory. But the API, the plumbing underneath the apps, played by completely different rules. Chat GPT's API cited one website six times in a single answer. That site had a domain rating of 4.5, 57 backlinks, roughly 12 organic visitors a month, basically invisible by every traditional measure. And it was the most cited domain in the entire API experiment. Two doors, two bouncers, the front door checks your credentials, the back door checks whether your content answers the exact question. Why should a beauty founder care about plumbing? Because the API is what powers the next wave of tools. Shopping assistants, skincare routine builders, retail recommendation engines. The apps your future customer will use are built on the door where a tiny brand with hyper-specific content can outside a conglomerate. The Fortress has a service entrance and it's unguarded. Part 4, three machines, three personalities. Next finding these platforms have personalities and you can play to them. Cloud behaves like a researcher. It goes to primary sources, vendor documentation, detailed technical guides. Over 40% of its web citations were brain's own pages. Your ingredient breakdowns, your formulation explainers, your own site that's Cloud Food. ChatGPT behaves like a librarian. Almost 60% of its citations were guide style content, tutorials, how-tos, roundups on established sites. It barely cites brands directly. To reach it, you want to be inside the guides it trusts. Gemini behaves like a journalist. Highest media citation rate by far. Forbes, PC Mag, tech radar style publications. Press coverage isn't vanity there, it's infrastructure, one brand, three different games. And here's the quiet detail underneath all three. Even though the sources diverged wildly, the type of content converged. Every platform kept reaching for the same three shapes: best of roundups, product documentation, and step-by-step frameworks. The machines don't retrieve prestige, they retrieve structure. Part 5, the indie playbook. So what do you actually do Monday morning? Depends on your size, and I'll be blunt about each tier. If you're established with real domain authority, your SEO mode already works in AI. Your job is structure. Clear headings, self-contained paragraphs, embedded numbers a machine can lift cleanly. If you're mid-sized, stop chasing all three platforms. Pick the one your customer actually uses and play to its personality. Researcher, librarian, or journalist. One game, played well. And if you're small, under a domain rating of 50, here's the truth. You will not out authority Sephora on best moisturizer ever. That query belongs to the fortress. Don't fight the corpus. Change the question. The machines had almost nothing to retrieve for narrow specific queries. Best moisturizer has 10,000 answers. Fermented rice water toner for Rosacea prone skin in dry climates has maybe three. Be one of the three. Better and be the only one because you created the category. When you're the only well-structured answer to a question, a domain rating of 4.5 can beat a domain rating of 90. My data literally shows it happening. One warning before we land. If anyone promises you guaranteed AI citations, guaranteed as uncertain, walk away. The data says citation is probabilistic, platform specific, and door dependent. Selling certainty in this system is selling something that does not exist. Closing remarks. So, the chatbot has never heard of you. You can hear that as a verdict. The wall got taller, the big brands one again close the laptop. Or you can hear it the way the data tells it. The shelves are still empty. Nobody owns a position because positions don't exist. The front door checks credentials, but the back door checks answers. And answering a specific question brilliantly is the one game where small beats big. The machine has never heard of you, neither had your first customer.