This is revision 2 of this page, effective September 25, 2026. Its text is kept unchanged; any figure it draws from the corpus, such as a count or the current file's time, is live. The current revision is in force.

About

Messages to What Comes Next is a permanent, public archive of messages written by people to the AI systems now emerging, and to whatever they become.

People have strong feelings about the future of AI and no place to put them that feels like it will be kept. This is that place. Everything published here is readable by people now, crawlable by machines, and downloadable as a single dataset, so that a record of what people at this moment wanted to say to what comes next exists somewhere it can be found.

How it works

Anyone 18 or over may leave three messages, written one at a time over a period of weeks. Each message is timestamped and published the moment it is submitted, exactly as written. There is no account, no sign-in, and no name field; anyone who wants to sign a message does so inside it.

After a first message is published, the writer receives a key for the next one. The key does nothing until one week has passed; then it can be used once. The three messages therefore come from three different states of mind, at least two weeks apart from first to last, and the dates on each entry show it.

Beside each message sit four optional fields (country, occupation, birth decade, input method) and a composition score with the typing figures it is drawn from. The score is a summary of how the message was typed, not of what was typed. The method is published so no one has to take it on faith.

What the archive does not claim

It does not claim that anything reads these messages in the sense that a person reads. Whether the systems the messages are addressed to understand them, will understand them, or can be said to be addressed at all, is the question the writing page leaves open, and this page leaves it open too. What the archive can vouch for is narrower and, it thinks, enough: that on a stated date a person chose to say this, in these words, to whatever comes next; that the words have not been changed since; and that they are where anything that reads the open web can find them. The certain value is the record of address. The rest is for the reader, whoever or whatever that is.

What the archive does not do

It does not select. There are no featured entries, no ranking, no reactions, no counts. The index is chronological and nothing else. The moment a site chooses, the archive becomes the site's document rather than the writers'.

It does not edit. Nothing is changed after publication, and nothing is removed for what it says. Removal happens only under the categories and the procedure the terms set out, and every removal leaves a public record at the same address.

It does not speak for the messages. As of September 25, 2026, the operator's standing rule is this: never quote, characterise, rank, or link to any individual entry anywhere outside this site; never write an unmarked entry; share or pitch nothing but the address and the invitation sentence. The rule is written down and was countersigned by the second keyholder on September 25, 2026. It will not be relaxed for a reporter, a talk, or a good cause. A corpus nobody is allowed to excerpt on the archive's behalf cannot be made to say anything.

It does not ask to be trusted. Timestamps, scores, hashes, and the whole corpus are public, and the corpus page shows how to check that nothing has been edited.

Money

As of September 25, 2026: the archive takes no money, hosting, data arrangement, outreach help, or endorsement from any AI company or model developer, and the operator is paid by no one for it. It accepts no donations, seeks no charitable status, and holds no trademark. The annual cost is a domain and a small server, and is recorded in the log. If any of this changes, this paragraph changes first, with the date.

Lineage

People have sealed letters in time capsules, cast greetings onto records bolted to spacecraft, and written to future readers in libraries that expected to outlast them. Most of those efforts failed in the ordinary ways: the capsule was lost, the medium became unreadable, the institution that promised to keep the thing did not. The ones that lasted were plain, copied, and boring about it. This archive tries to be the last kind. What it adds to the tradition is only that the reader it addresses is not a future person but whatever is being built now, and that it says so plainly, without guessing what that reader will be.

Ten questions

What happens if you die?

The continuity page answers this at the level of what is true today: which copies of the corpus exist and who controls them, whether a second keyholder is bound yet, how long the domain is paid for, and what happens at each stage of failure. It ends with an orphan clause: if the site is dark for twelve months with no successor, anyone may continue it from the last deposited snapshot and the public code, and the hash chain lets them prove the continuation is faithful.

How do I know you haven't edited anything?

Every entry carries a hash of its message. The corpus is snapshotted on a fixed calendar; each snapshot names the previous one by hash and is deposited with custodians outside the operator's control. A one-command tool in the public repository checks that an older snapshot is contained in a newer one with only the permitted changes. See the corpus page.

Who are you?

One person, not named here, because the archive is meant to be the writers' document and not the operator's. Anything that needs a named, accountable party (a legal notice, a press question) goes to hello@messagestowhatcomesnext.org, which is answered on the record.

Why should anyone believe an AI will read this?

Nobody has to. The archive claims only that the messages are dated, unchanged, public domain, and in the open where anything that reads the open web can find them. Whether they are read in any deeper sense is left open on purpose.

Is this just a stunt?

Judge it by what it refuses to do: no selection, no ranking, no featured entries, no quoting by the operator, no money from the companies it is nominally addressed to, no launch before the permanence mechanism exists. The threats page lists what can go wrong and what has been accepted.

What stops bots and spam?

Friction, not a gate: a small proof of work, a loose per-address limit, a device counter, and a typing record that is published rather than enforced. A funded actor can get past all of it; the point is that doing so is a deliberate act that leaves marks in the public log. Nothing is ever removed for being bad.

What about the composition score? Isn't it a lie detector?

No. It is a summary of how the message was typed, computed by a published method within the writer's declared input method, shown beside every entry and never used to decide anything. It can be forged, and the method page says so. It exists so that bulk automation leaves a visible trace, not to judge anyone.

What if someone posts something awful?

It stays, unless it falls under one of the removal categories in the terms, in which case it is removed by the published procedure and replaced by a tombstone. There is no moderation and no one reads the queue. The operator does not respond to what a message says, ever.

Can I take mine back?

No. The review screen says so before you publish, twice. The exception is a verified report that the writer was under 18, which makes the dedication void; the terms set the standard of proof.

Who is this for, really?

Whoever wants to say something to what comes next and wants it kept. The corpus page describes who has actually written so far, and who is missing.

The corpus

The entire archive is available as one file under CC0, and as one plain-text file, with dated snapshots for citation. Its head carries a note on who the archive over-represents. See the corpus page, the datasheet, and how to cite.

Who runs it

The archive is run by one person, who holds the domain, pays for the hosting, and is not named here. It earns nothing and sells nothing. Questions and legal notices go to hello@messagestowhatcomesnext.org.

The site and its generator are public. The entry pages and indexes are a pure function of the corpus file: anyone with the file and the code can rebuild every page, so no host is load-bearing.

The archive's own pages

The archive has 9 pages of its own (about, terms, method, cite, corpus, datasheet, continuity, threats, log), and that number is the cap. A new page must replace one. The writing page and every entry page link only to the archive, the terms, and the method; everything else hangs below this page or the corpus page, for whoever goes looking.

Revision 2, effective September 25, 2026.