--- title: "Publishing Is Not Distribution: An AI Retrieval Experiment" description: "I gave ChatGPT, Claude and Codex exact words from my own published article. None found it. One other system did. Here's what that reveals about the six-stage AI visibility chain." canonical: https://trychad.ai/blog/publishing-is-not-distribution-ai-retrieval-experiment/ --- Founder NoteFOUNDER NOTE · Oct 20, 2026 · 7 min read · By Alexander Shoucair # Publishing Is Not Distribution TL;DR I gave ChatGPT, Claude, and Codex exact quotes from an article I had published on two platforms. None of them found it. A different retrieval system found both versions immediately. The article wasn't invisible — it was selectively invisible. Visibility is not binary; it depends on which system is doing the looking, and passing one stage of the chain doesn't guarantee the next. A few weeks ago I ran a simple experiment. I took exact sentences from a published article and asked AI systems to find where they came from. The result was more uncomfortable than I expected. ## The setup I had written an article about using coding agents to evaluate a startup's technical SEO and AI visibility — what each model found, what they missed, and what it meant for how founders should think about AI tools. I published the original on the ZeroLayer website and syndicated it to Medium. Then I ran a simpler experiment on top of it. I copied two passages from the published article verbatim and searched for them using ChatGPT, Claude, and Codex. I also ran the same search through Google. I wasn't asking these systems to recommend ZeroLayer. I wasn't testing brand authority or AI visibility in the commercial sense. I was giving them the exact words from a public document and asking them to find where it came from. None of them returned the article. Neither the ZeroLayer version nor the Medium version surfaced. ## Then another system found it When I ran the same passages through a different retrieval system, it found the ZeroLayer article. For one of the passages, it also surfaced the Medium version. That changed what the result meant. The finding wasn't: The article is invisible to AI systems. It was: The article was retrievable by one system and effectively nonexistent to three others — using the same input. That distinction matters more than most AI visibility conversations acknowledge. ## The six-stage chain most founders collapse into one Founders tend to think about online visibility as a simple sequence: publish a page, a crawler finds it, a search engine indexes it, someone searches and the page appears. AI systems make that sequence much less predictable. A page can be publicly available without being indexed by a particular system. It can be indexed without being selected for a result. It can appear in conventional search while being absent from an AI answer. It can be retrieved by one AI system and ignored by another. And even if a system retrieves it, that doesn't mean it will cite it — let alone recommend the company behind it. The AI Visibility Chain Published page exists → Crawlable bots allowed → Indexed in the corpus → Retrievable selected by query → Cited included in answer → Recommended trusted + preferred Passing one stage does not guarantee the next. Each transition can fail independently — and each failure requires a different intervention. Most conversations about AI visibility collapse all six stages into one vague idea of being "visible." A visibility score that records whether your company appeared in a set of AI answers hides everything that happened — or didn't — upstream of that answer. ## Medium didn't automatically solve the problem I had assumed that syndicating to Medium would make the article easier to find. Medium is an established domain with significant authority, internal distribution, and a large volume of indexed content. The article had also generated more engagement than my previous Medium posts. But it still wasn't consistently retrievable. The platform improved human discovery — new readers found the article through Medium's recommendation system who would never have visited ZeroLayer. It didn't improve machine retrieval in the systems I tested. Platform distribution and machine retrieval are different things. An article can reach thousands of human readers through an established platform while remaining effectively invisible to the AI systems those same readers use to answer questions. ## The score is the symptom. The chain is the system. When a company doesn't appear in an AI answer, there are at least five distinct reasons why — and they require completely different responses: - **Crawl failure:** the page was never seen. Adding schema won't fix this. - **Index failure:** the page was crawled but not retained. Publishing more content won't fix this. - **Retrieval failure:** the page is indexed but the query doesn't surface it. Authority building won't fix a technical rendering problem. - **Citation failure:** the page is retrieved but judged less authoritative than competing sources. More backlinks won't fix a content structure problem. - **Recommendation failure:** the company is cited but never recommended when a buyer asks a direct question. This is the hardest gap — it lives in brand trust and answer-layer positioning, not technical fixes. The audit that produces the largest list of recommendations isn't the most useful one. The useful one is the one that identifies which stage is failing and picks the next action accordingly. ## What I'm testing next I'm repeating this across a larger set of articles — some on Medium only, some on a company website only, some published on both with the company site marked as canonical. For each article, I want to answer three questions: 1. Can a system retrieve it from an exact passage? 2. Can it find the article from its title and subject? 3. Will it cite the article when answering the natural question the article was written to address? Those are three different tests. Passing the first doesn't predict the second. Passing both doesn't predict the third. That gap — between technically retrievable and actually useful to a buyer — is where the next generation of visibility tools needs to operate. "Found it once under ideal conditions" is not the same as discoverable. ## Frequently asked questions ### Why can't AI systems find content that's been published and indexed? Publishing and indexing are only two stages in a six-stage visibility chain: published → crawlable → indexed → retrievable → cited → recommended. A page can clear every stage up to indexing and still fail the retrieval step. Each AI platform runs its own retrieval system with different crawl schedules, content weighting, and selection criteria. Clearing one stage does not guarantee the next. ### Does publishing on Medium or a high-authority platform guarantee AI retrieval? No. Platform authority improves the odds of human discovery through feeds and internal recommendation, but it does not guarantee machine retrieval. In Alexander Shoucair's experiment, an article published on both ZeroLayer and Medium was undetectable by ChatGPT, Claude, and Codex when searched by exact passage — despite one retrieval system (Perplexity) finding both versions. The same content was simultaneously discoverable and invisible depending on which system was doing the looking. ### What's the difference between a visibility score and a retrieval chain diagnosis? A visibility score is a snapshot: it records whether your company appeared in a set of AI answers at a point in time. A retrieval chain diagnosis traces why you did or didn't appear. Was the page never crawled? Indexed but not retrieved? Retrieved but judged irrelevant? Cited but not recommended? Those failures require completely different interventions. A score is the symptom; the retrieval chain is the system. ## Sources - **TryChad.ai** — first-person retrieval experiment: exact passages from a published article searched across ChatGPT, Claude, Codex and one additional retrieval system, October 2026. Primary source. Visibility is a chain, not a score Chad diagnoses where in the retrieval chain your content is breaking — and turns that diagnosis into completed work. Not a report. Not a list of recommendations. A fix, verified against the live system. [Run a free visibility scan →](https://trychad.ai/book) ## More field notes [All posts →](https://trychad.ai/blog/) ###[FounderThey Already Had an Agency, SEO Content and Automation. It Still Wasn't Working.The problem wasn't a lack of marketing activity. It was that nobody owned the system. Between June and September 2026, Rocket Room's sessions grew 147%. Here's what that taught us about marketing operating systems.7 Oct 2026 · 8 min read](https://trychad.ai/blog/rocket-room-147-percent-sessions-operating-system-not-more-content/)###[BenchmarkPerplexity Has 1.9% Market Share But Cites You 17.5% of the TimePerplexity's citation rate is 9x its market share. Claude's is less than 1x. Market share doesn't predict where citations come from — here's what that means for your GEO platform priority.19 Oct 2026 · 6 min read](https://trychad.ai/blog/perplexity-citation-rate-market-share-inversion/)###[BenchmarkPerplexity's Photon Engine: FAQPage Schema Doubles Citation FrequencyPerplexity published its Photon retrieval engine's ranking signals. FAQPage JSON-LD doubled citation snippet frequency in A/B tests. Here's what Photon weights and what to do about it.16 Oct 2026 · 6 min read](https://trychad.ai/blog/perplexity-photon-faqpage-2x-citation-frequency/) ## How does your site look to Chad? Drop your domain. In 11 seconds you get your GEO score, your biggest gap, and the first thing Chad would fix.