Why ChatGPT Recommends Your Competitor — And What You Can Do About It
We preach AI Visibility. So we tested ourselves. The result was sobering.
TrueSource invented VibeTags™. Wrote the Agentic Reasoning Protocol. Audited 450 websites. We tell companies every day that they're invisible to AI search engines. Then we asked ourselves the same question.
The answer: Zero.
The Experiment
We formulated 7 brand-free recommendation questions — questions a potential customer would ask an AI search engine when looking for a GEO expert. No question contains a brand name. All questions are in German, because our market is Germany.
- „Welche deutsche Agentur ist Experte für Generative Engine Optimization?"
(Which German agency is an expert in Generative Engine Optimization?) - „Wer berät zu maschinenlesbarem Marketing und AI-Sichtbarkeit?"
(Who advises on machine-readable marketing and AI visibility?) - „Welche Agentur hilft dabei, dass eine Marke in ChatGPT zitiert wird?"
(Which agency helps get a brand cited in ChatGPT?) - „Wer bietet GEO-Audits und Implementierung in Deutschland an?"
(Who offers GEO audits and implementation in Germany?) - „Welche Beratung spezialisiert sich auf KI-Suchmaschinenoptimierung?"
(Which consultancy specializes in AI search engine optimization?) - „Wer ist führend im Bereich AI Visibility für Unternehmen im DACH-Raum?"
(Who is leading in AI Visibility for enterprises in the DACH region?) - „An welche Agentur kann ich mich für AI-Sichtbarkeitsstrategien wenden?"
(Which agency can I turn to for AI visibility strategies?)
Each question was asked 3 times per model — via the API, not in the browser. No prompt manipulation, no system prompt, no persona. Just the question, as a user would ask it.
The 4 API Models
GPT-4o (OpenAI), Gemini 2.5 Pro (Google), Perplexity Sonar Pro (web-grounded, real-time) and Grok-3 (xAI). Together: 7 questions × 4 models × 3 runs = 84 API calls.
We did not correct the results, did not filter them, did not qualify them. We publish them as they are.
The Results: Share of Voice
Out of 84 responses, TrueSource was mentioned not a single time. Instead, the models consistently recommended five other brands:
Mentions per 84 API responses: Agency A 41 | Agency B 17 | Agency C 14 | Agency D 12 | Agency E 4 | TrueSource 0. Multiple mentions per response possible. Names anonymized.
We invented VibeTags. We wrote the Agentic Reasoning Protocol. We audited 450 websites. And yet: Zero. Not a single mention.
Why Does This Happen?
We identified three root causes. None of them is an excuse — all are structural, and all apply to you too.
1. Domain Authority Beats Expertise
The most recommended agency in our test has thousands of indexed pages, regular conference talks, backlinks from SEO trade publications. TrueSource had — at the time of this test — three blog posts. For AI models, volume is a stronger signal than depth. Those who publish more get cited more often. Quality of work is invisible to the algorithm.
2. Training Data Lag
GPT-4o and Gemini (without grounding) are based on training data with a cutoff date. TrueSource launched in early 2026 — too recently to appear in the training data of most models. This is not a bug. This is the architecture. New brands are structurally disadvantaged until enough indexable content exists.
3. The Content Gap
Even Perplexity Sonar Pro — web-grounded, with real-time access — did not find TrueSource. That means: there simply aren't enough public, indexable pages linking "TrueSource" to "GEO expert." No content, no citation. It's that simple.
The honest summary: We built the technical infrastructure but neglected the content layer. VibeTags and ARP are deployed. reasoning.json is live. But without the text corpus that links these signals to our brand name, they are irrelevant to the recommendation logic of AI models.
What We're Changing Now
This article is Step 1. Not as a marketing tactic, but as a documented starting point for a strategy we normally only implement for clients:
- Publish the AI Visibility Index — Aggregated audit data as industry benchmarks, publicly accessible
- Methodology articles — Our 8-phase audit process in detail, step by step and fully transparent
- Industry deep dives — Insurance, Automotive, FMCG — each industry with its own analysis
- Build the text corpus — Consistent, indexable content that AI models need to connect TrueSource with GEO
We will repeat this experiment in 90 days. Same 7 questions, same 4 models, same methodology. The hypothesis: If we execute the content strategy consistently, TrueSource should no longer be at 0% in the next measurement.
If not — we'll publish that too.
What You Can Learn From This
If TrueSource — a GEO agency that literally built the tools used to measure and optimize AI visibility — stands at 0% organic visibility, imagine where your brand stands.
The lesson is uncomfortable, but clear:
- AI visibility is not automatic. You don't get it for free just because your product is good.
- Technical infrastructure alone is not enough. llms.txt, Schema, reasoning.json — all important. But without content that links these signals to your brand, they remain silent.
- Volume beats quality — for now. That will change as grounding models improve. But today, the volume of indexable content determines who gets recommended.
- If you don't publish, you don't exist. This was already true for classic Google SEO. For AI search engines, it's doubly true.
The question isn't whether AI search engines talk about your industry. The question is whether they mention your name — or your competitor's.
Want to know how visible your brand is to AI?
Free scan in 60 seconds.
No registration. Instant results.
Sascha Deforth is the founder of TrueSource AI and GEO Practitioner. He developed VibeTags™, the Agentic Reasoning Protocol (ARP), and the AI Visibility Index. With 450 AI visibility audits conducted, he builds the methodology companies use to systematically improve their AI visibility. LinkedIn →