Answer Box
Developer-tool companies that optimize their documentation for AI citation can achieve up to 41% mention rates in LLM answers, while the median company earns only 4.5% of citations from its own domain, according to a six-month audit of 20+ companies by Althea Labs. The audit measured how often models like ChatGPT, Gemini, Perplexity, Copilot, and Google AI Overviews cite companies when developers ask category-specific questions, revealing that documentation — not blogs or press — is the primary source of AI citations.
Documentation as the new marketing front door
Althea Labs' audit of 20+ companies, mostly in developer tools, found that the best-performing firm gets mentioned in 41% of AI answers in its category, with consistent performance across all models. The citations came overwhelmingly from core documentation, SDK hubs, and protocol-level explainers, not from blog posts or media coverage. That company earned 17% of all its category citations from its own domain, far above the 4.5% median. Another company achieved a 33% owned citation rate driven almost entirely by a single honest comparison page — "best X databases compared," including competitors — on its own site. One well-structured, genuinely useful page outperformed years of blog content.
The mechanism is clear: when a developer asks an LLM how to implement something, the model reaches for content that is structured, specific, and extractable. Documentation fits perfectly with clear headings, canonical answers, code blocks, and no marketing fluff. A docs page that states exactly what a thing does, in what versions, with what limitations, is one of the most citable objects on the internet. Currently, Medium (37% citation rate), Reddit (32%), YouTube (18%), and DEV.to (16.5%) are eating citations that could point at a company's own domain. To capture those citations, companies should treat documentation as a discovery surface — ungate white papers, add comparison pages that include competitors honestly, and use spec tables and decision criteria instead of adjectives.
Visibility varies wildly by model — and the gap reveals what's broken
No company has a single AI visibility number; every firm has six, and they often disagree. Althea audited a cloud infrastructure company that had 33% visibility on ChatGPT and Gemini, 16% on Perplexity, and 18% on Copilot. An edtech platform showed 56% on Gemini but just 9.6% on Perplexity. Across 12 companies with model-level splits, Perplexity came in below the other models for 10 of them. The models sit on a spectrum from "training-data-driven" (ChatGPT, Gemini) to "search-driven" (Perplexity). The Perplexity gap is diagnostic: high everywhere but low on Perplexity means a company is coasting on brand equity — the models "remember" it, but the current site isn't producing citable pages, and that equity erodes with each retraining. Low everywhere but alive on Perplexity means the content works but the company has no entity footprint; the most extreme case registered 0% on ChatGPT, Gemini, Copilot, and Google AI surfaces, existing only on Perplexity via a handful of citable pages — a single point of failure. Low on Copilot specifically matters for enterprise sellers: one database company showed 0% on Copilot and 0.5% on ChatGPT while doing fine on Google surfaces, and Copilot is embedded in the environment where enterprise buyers do vendor research. Companies should measure per model, then fix the weak side: search-driven models want citable pages (structure, schema, llms.txt, clean sitemaps), while training-driven models want entity signals (Wikipedia, Crunchbase, consistent third-party coverage).
Rebranding without entity migration erases AI visibility
One company rebranded in mid-2023 with a new name, new domain, and proper redirects — all traditional SEO boxes checked. Yet its old brand name ranks #3 in its category in AI answers, mentioned 36% of the time, while the new brand ranks #28, mentioned 3.7% of the time. The retired brand is still 10 times more visible than the live one. Training cutoffs are the cause: a model trained before the rebrand has never heard of the new company, and it has years of docs, Stack Overflow threads, conference talks, and blog posts about the old name. Redirects fix HTTP, but a 301 does not teach a language model that two entities are the same. Visibility decays slowly — every quarter until major models retrain on a web that consistently connects the old name to the new one, recommendations keep flowing to a brand that no longer exists. For any rebrand, companies should treat entity migration as a launch-critical workstream: update Wikipedia, Wikidata, Crunchbase, PitchBook, G2, and GitHub org metadata on day one; get third-party coverage that explicitly states "X, formerly known as Y"; add Organization schema on the domain declaring the former name; and monitor model by model until the new entity actually takes.
The wide-angle view
The most encouraging stat in the dataset: the #1 most-visible brand in one category appears in only 17% of AI answers about its own space. The "winner" is absent from 4 out of 5 answers. In dev tools, where documentation culture already produces the exact kind of content LLMs want to cite, that is not a threat. It is the most open distribution channel since early Google, and the teams that treat their docs, comparison pages, and community content as citation infrastructure are going to quietly own it.