The case for cloaking content to LLM bots
LLM providers have not banned cloaking. The brands testing the boundary now will own citation share before the window closes.
Key takeaways
- Cloaking is banned by Google but not by OpenAI, Anthropic, Perplexity, or Google-Extended.
- Most enterprise pages underperform inside LLMs because they are built for human attention, not machine comprehension.
- A bot-facing version with declarative claims, named entities, and dates reduces hallucination risk around your brand.
- Expect LLM providers to publish anti-cloaking guidance within 12 to 18 months. The window is open now.
What happened
Per iPullRank, Mike King is making the case that brands should consider serving large language model crawlers a different version of their content than they serve to human users or to Google. The argument: cloaking is against Google's guidelines and remains a bad idea for traditional SEO, but LLM bots like GPTBot, ClaudeBot, PerplexityBot, and Google-Extended sit outside that rulebook. They have no published anti-cloaking policy, and they are not ranking pages; they are ingesting facts.
iPullRank's framing is blunt: if the goal is to be cited and represented accurately inside ChatGPT, Claude, Perplexity, and Gemini answers, then the page a model ingests should be optimized for machine comprehension, not for human design conventions. That means stripping interactive elements, expanding acronyms, front-loading definitions, replacing visual hierarchy with explicit statements, and feeding the model the cleanest possible version of your claims.
The provocation is real. The technique is not theoretical. It is already being tested by SEO teams who have decided that LLM visibility is worth a separate content pipeline.
Why it matters for your brand
The practical question for a CMO is not whether cloaking for LLMs is ethically clean. It is whether your current website, built for humans and Google, is actively underperforming inside the models your buyers now use to shortlist vendors, frame policy, or research counterparties. In most cases, it is.
Consider a global reinsurer's thought leadership page. It opens with a hero video, a pull quote from the CEO, three columns of navigation, and a 200-word intro that takes its time getting to the point. A human reader tolerates this. An LLM crawler indexing the page for a future query like "which reinsurers have the strongest climate risk methodology" sees a thin signal buried under markup. The model extracts what it can, which may not be the sentence you wanted associated with your brand. A parallel version served to GPTBot, written as declarative statements with named methodologies, dates, and figures, materially changes what the model learns about you.
For multilaterals and policy institutions, the stakes are sharper. When ChatGPT is asked to summarize UNDRR's position on Sendai Framework progress, or CGAP's stance on financial inclusion metrics, the model is reconstructing an answer from whatever it has ingested. If the ingested version is a JavaScript-heavy page with the substantive claims hidden behind tabs and accordions, the model will hallucinate around the gap. A bot-facing version that exposes the full text, with clear attribution and dates, reduces that risk. This is brand safety, not gaming.
For financial services, the compliance angle cuts both ways. Serving different content to bots invites scrutiny from regulators who care about consistency of public disclosures. But serving the same content also means accepting that the model's representation of your firm is shaped by whatever marketing copy happened to win the homepage redesign. Neither is neutral. The cloaking question forces a decision that is already being made by default.
The content strategy implication: most enterprise content teams are still producing one artifact per topic, optimized for a human reader on a desktop. The emerging discipline produces two. One for humans, with the design and narrative that wins attention. One for machines, with the structured claims, named entities, and explicit relationships that win citation. The teams that build this dual pipeline now will be quoted by name in LLM answers eighteen months from now. The teams that do not will be paraphrased, misattributed, or omitted.
The signal in context
Cloaking has been a dirty word in search for two decades because Google built its entire economic model on the premise that what the crawler sees must equal what the user sees. LLM providers have not made that promise. They have published crawler user-agents, robots.txt directives, and ingestion policies, but none of them has said "the page you serve us must match the page you serve a human." That silence is an opening, and iPullRank is the first prominent voice to name it clearly.
This sits alongside a broader shift in how technical SEO practitioners are reframing their craft for the LLM era: structured data is being repurposed for model grounding, llms.txt files are being proposed as a standard, and content teams are starting to think about "machine-readable canonical versions" of their key claims. The cloaking conversation is the sharp end of the same trend. Expect the major LLM providers to publish anti-cloaking guidance within twelve to eighteen months, at which point the window closes. Until then, the brands willing to test the boundary will accumulate citation share that compounds.