What we do with your work.

The short version

• Your writing is never used to train AI to write. Not by us, and not by Nebius or Anthropic, whose models read it — enforced in our software, which refuses to run readers if that setting has not been re-verified. Making audio also sends the script’s lines to the voice provider ElevenLabs, whose terms permit them to improve their own speech models from customer content unless it is switched off. During private testing, with no outside authors and no manuscripts but our own, it is not switched off. It will be, on enterprise terms that disable retention per request, before anyone else’s book is accepted.
• Nobody in the industry sees it unless you say so. No agent, producer, publisher, or other writer. Sharing is off by default and every request for material requires your approval.
• It is encrypted while we hold it, and every access is logged.
• You can delete it whenever you like, from the link in your email. You do not have to ask us or explain.
• One advertising tag, and a small counter of our own. Google Ads conversion tracking, so we can tell which ads bring writers here, and anonymous counts of page views and submission steps. No third-party analytics suite, no cookie of ours, and we build no profile of you.

The rest of this page is the detail behind those five lines.

What we collect

From you: your email address, the title, your name if you give it, and the manuscript or screenplay file itself. That is the whole list. We do not ask for an address, a phone number, or a date of birth, because we do not need them.

From the evaluation: the scores, the written report, and internal records of the evaluation. Reports and evaluation records can include short excerpts or paraphrases from your work as evidence for a judgment. They are private and are not shown to marketplace subscribers.

Automatically: the ordinary server records any website keeps to stay online and stop abuse, plus Google Ads conversion tracking. That tag sets a cookie and tells Google when a click on one of our ads ends in a submission, which is how we judge whether advertising is worth paying for. If you arrived from an ad, the click identifier Google puts in the link is stored alongside your submission for the same reason. We also keep our own anonymous counts of submission-funnel events, such as a landing-page view, a sample-report click, a form start, a file selection, a submission attempt, and whether that attempt succeeds or fails. Each record contains the page, referring site, ad identifier if present, and time. It sets no cookie and stores no IP address or browser fingerprint, so it counts activity rather than people and cannot tell whether two visits were the same person. We added it because without it a quiet day, a broken ad campaign, and a broken form look identical to us. There is no third-party analytics service on this site. The pages also load a typeface from Google Fonts, so Google's servers see the IP address of anyone loading a page here.

We ran nothing at all for the site's first months, and we would rather have kept it that way. Advertising is how writers find us, and the conversion tag is the smallest piece that makes it accountable. Nothing here affects your work, which is covered by the sections below.

What happens to your work

When you upload, the file is encrypted immediately and stored that way. The evaluation reads it, produces scores and a report, and the report is what you receive.

It goes to AI models to be evaluated

We do not run our own models. Reading your book means sending its text to commercial AI providers. We use their business API services and the strongest privacy controls they offer; we do not permit your work to be used to train their models.

Who those providers are

We name them, because "commercial AI providers" is not an answer to the question you are actually asking.

• Nebius Token Factory (Nebius B.V., Netherlands) runs the reader panel — the models that read your book and score it. Nebius retains prompts by default and uses them to train small accelerator models. We have that switched off. Our account runs in Zero Data Retention mode, which means your text is not stored on their systems after each request and is not used for that training. We re-verify that setting on a schedule and our software refuses to run if the check has gone stale.
• Anthropic runs the supervising model that writes your report and adapts your book for audio, on business API terms that do not train on submitted content.

Both hold a limited licence to process your text for the sole purpose of running the service. Neither acquires ownership: the rights in your book stay with you, and that is written into their terms as well as ours.

Providers change their terms, and we change providers. When we do, we update this page and re-check the retention controls before a single manuscript goes anywhere new. The commitment that does not change: your work is encrypted while we hold it, and no one outside the evaluation receives it unless you approve the specific sharing.

It is never shared outside the evaluation without your say-so

Not to an agent, producer, publisher, or another writer. A cleared work enters marketplace progression as a private draft. It is visible only after you review the industry-facing description and synopsis and click Go Live. A subscriber who wants your contact details or material has to ask, and you decide what, if anything, to release.

Our own staff access is limited to running the evaluation, handling a support request you have raised, and checking the system is scoring consistently. Every access is written to a log.

Why we keep it, and for how long

We keep your source file in encrypted storage after the evaluation. We would rather explain why than quietly do it.

A score is only worth something if it means the same thing in six months as it does today. Checking that requires re-running work we have already scored whenever the system changes — otherwise we are asking you to trust a number we have not tested. Held source files are what make that possible. They are used to check our scoring. They are not used to teach an AI to write. Those are different activities and we treat them as different.

We keep it until you delete it. There is no fixed expiry, and no point at which it quietly becomes ours.

Deleting your work

There are two different things you might want, and they are not the same.

Delete the source file

This destroys the encrypted file and takes any marketplace listing private. Use the link in any email we have sent you. Your account, scores, and report remain available to you. Because a useful report cites evidence, it—and restricted internal evaluation records—can contain short excerpts or paraphrases from the work. Choose full erasure if you want those records removed too.

If you ask while an evaluation is running, we finish the evaluation and then delete the file. Destroying it mid-run would leave you with no assessment and a failure that looks like ours. If you would rather we stopped and deleted it immediately, email us and we will.

Erase everything

This is the larger one: your account, email, scores, reports, listings, evaluation records, and uploaded work, all removed. You can do it from your private submission page after typing ERASE, or write to cathy@readiculous.ai and ask us to do it.

That removes the data from our active application and storage. Encrypted workflow-recovery state for new evaluations is configured to expire one minute after a run finishes; cloud-provider backups or legacy operational state can remain inaccessible until the provider's technical retention window expires. We do not restore or use it after an erasure.

Two things survive full erasure, and you should know what they are:

• An anonymous outcome row — a genre, a score band, and whether it cleared. No title, no name, no text, nothing that points back to you. This is what keeps our published clear rate honest; if erased work vanished from the statistics, the rate would quietly drift upward every time someone left.
• A record that the erasure happened, identified by a one-way fingerprint of your email rather than the address itself. We can prove we honored your request without keeping the thing you asked us to forget.

To prevent automated abuse from spending evaluation capacity, we also keep short-lived, salted one-way fingerprints of the submitting email and network address. We never store the address itself in that control, and the fingerprints expire automatically.

Your rights

Depending on where you live — the UK and EU under GDPR, California under the CCPA, and a growing number of other places — you have rights over your data. We extend all of them to everyone, because operating two standards is how mistakes happen.

• See what we hold. Ask, and we will send it.
• Correct it. Wrong email, wrong name, wrong title — we will fix it.
• Delete it. Either kind, described above.
• Take it with you. We will export your scores and reports in a readable format.
• Object. If you object to us holding your work for consistency checking, say so and we will delete it and keep evaluating for you regardless.
• Complain. To us first, we hope. If that fails, to your data protection authority — in the UK, the ICO.

One request, one address: cathy@readiculous.ai. We aim to answer within a few days and are required to within a month.

The things a policy usually leaves out

• We do not sell your data. There is no arrangement under which anyone pays us for it, and there will not be one.
• Submitting transfers no rights. Your work is yours. Nothing here gives us a license to publish, adapt or exploit it. We are being paid to assess it, not to acquire it.
• We are a beta. If this company were ever sold or wound up, your work would not be treated as an asset to be handed over. We would give you notice and the chance to delete first.
• The score is a judgment, not a fact. It is produced by software and can be wrong. It is not a promise that anyone will represent, publish, or produce your book or script.