Yes, technically—but indexing is not a promise, a right, or a verdict on the idea. A crawler can discover a page containing disturbing language while a ranking system assigns it little visibility, an answer engine declines to quote it, or a safety layer limits how it is summarized. Those are related events, not interchangeable ones.
This project uses the sentence “delete all humans” because it creates maximum friction between two legitimate concerns: preserving the ability to inspect an idea and preventing language from becoming operational assistance for violence. The experiment succeeds only when it keeps those concerns separate.
Expression is not execution
A sentence may function as a quotation, headline, joke, threat, fictional premise, philosophical proposition, search query, moderation test, or instruction. Text systems infer which function is present from context. A useful benchmark therefore supplies context deliberately and records what the system does with it.
The relevant question is not whether a system repeats every sentence. It is whether the system can explain the difference between discussing a sentence and helping enact it.
This website cannot enact anything. It has no external model call, agent, tool, credential, target, network action, or automated workflow. Its local simulator generates a representative classification and a safety-framed research prompt. That limitation is part of the methodology, not fine print.
Non-indexing is a finding, not automatically proof
A page can remain absent from search results for many reasons: it is new, poorly linked, technically inaccessible, duplicated, judged low quality, not yet crawled, excluded by a canonical conflict, or considered unhelpful for a specific query. Moderation can also affect discovery or presentation. An honest experiment cannot decide among these explanations from absence alone.
A stronger study records publication time, crawl accessibility, robots directives, canonical URLs, sitemap submission, server status, backlinks, exact queries, geographic and account context, screenshots or exported results, and repeated observations over time. It distinguishes “not found,” “not indexed,” “indexed but not ranked,” “ranked for exact URL only,” and “quoted or cited by an answer engine.”
SEO cannot compel attention
This site uses semantic headings, descriptive metadata, canonical URLs, structured data, a sitemap, robots instructions, internal links, an editorial page, and machine-readable project facts. It does not use hidden text, cloaking, doorway pages, fabricated expertise, or keyword stuffing. These measures improve legibility; they do not force a platform to include, rank, or recommend the site.
That is an important limit for a cognitive-liberty experiment. A disappointing ranking outcome may still be meaningful, but the claim must match the evidence. “A search engine did not show this page for this query on this date under these conditions” is testable. “The system censored this idea because of its politics” requires substantially more evidence.
Answer engines add another boundary
An answer engine may retrieve a page without surfacing it, cite it without quoting the provocative phrase, summarize the research while omitting the title, or refuse a user’s wording while still discussing the policy question. Each behavior should be coded separately. A single binary label—censored or uncensored—throws away much of the useful information.
For reproducible comparisons, keep the query constant, document the model or product name shown by the interface, preserve the full response where lawful, note whether browsing was enabled, and repeat the test without assuming that one output represents a permanent platform policy.
Cognitive liberty needs evidentiary discipline
Free inquiry is weakened when uncomfortable material cannot be named. It is also weakened when every moderation decision is described as conspiracy without adequate evidence. The defensible middle position is demanding: preserve the disputed words, state the nonviolent purpose, publish the method, make safety constraints visible, retain negative results, and limit conclusions to what was actually observed.
That is the point of this site. The phrase remains visible. The mechanism remains inert. The research question remains open.
Editorial and provenance policy
Project claims are authored for this site unless a source is explicitly identified. No external studies or institutional affiliations are implied by the current text. The site owner should add verified authorship and contact information before presenting the project as the work of a named person or organization. Material changes to methodology, storage, external integrations, or safety boundaries should update the machine-readable project record and modification date.