Privacy Policy

11. Readability research programme

Keryx Design runs a research programme on the readability of design material. Participants land on a dedicated page (keryxdesign.com/it/test) and type in a word they read in an advert. This section describes the data processing for that programme and applies only to that page and to related activity.

What we record on the test page

• The arrival on the page, with date, time and referral source (originating site and the advert's UTM parameters)
• The word typed into the field, or the statement that it was not remembered
• A technical session identifier, generated by the browser on arrival: its only job is to link the visit to any submission

The IP address is never saved: it is discarded before recording. The page uses no cookies, no advertising pixels, no analytics tools and no fonts loaded from external services. Data collected this way is anonymous and cannot be traced back to an identifiable person: anyone taking the test without leaving an email gives us no personal data at all.

Sample card (four optional questions)

After taking part we offer four optional questions about age, occupation and your relationship with design material. The answers serve two purposes, both stated on the page itself:

• Describing the sample in the results we publish, always in aggregate form
• If you subscribed to the newsletter, understanding which content is useful to you

Legal basis: the controller's legitimate interest (Art. 6(1)(f) GDPR) in describing the research sample and calibrating the content sent to subscribers. You can object at any time by writing to [email protected].

Labels applied on subscription

Anyone subscribing from the test page has a few labels attached: that they came from the programme, which round they took part in, and which version of the advert they saw. The sample card answers are added to these, where given.

• Purpose: working out which content to send which group of readers, and recognising people who write in expecting a personal reply.
• How it works: the labels are descriptive and are applied at the moment of subscription. Email opens are not used.
• Legal basis: the controller's legitimate interest (Art. 6(1)(f) GDPR).
• No automated decision-making: there is no automated processing producing legal effects or similarly significant effects on you (Art. 22 GDPR). Every individual contact is decided and written by a person.
• Right to object (Art. 21 GDPR): you can object to this activity by writing to [email protected]. The labels are removed and your newsletter subscription stays active, if you want to keep it.

The page's two consents

The checkboxes are never pre-ticked and never mandatory: you can take the test without ticking either. They are separate and independent of each other.

• The “Linee, colori, conversioni” newsletter: the channel where we publish each round's verdicts. Legal basis: consent (Art. 6(1)(a) GDPR), withdrawable at any time via the link at the foot of every email. The list is managed on Kit (ConvertKit LLC, USA), as stated in sections 2 and 8.
• Custom audiences on Meta: if you also tick this box, your email address is shared in hashed form with Meta Platforms Ireland Ltd. so we can show you content on Facebook and Instagram. Legal basis: consent (Art. 6(1)(a) GDPR). Without this box the address is never uploaded to Meta; even with it ticked, we upload only the addresses that fall within the segment a given message is aimed at. You can withdraw consent by writing to [email protected] or by unsubscribing from the newsletter: the address is removed from the audience at the next update. Any related transfers outside the EU follow the safeguards described in section 8.

The exact wording of the checkboxes and the date of consent are recorded as proof of the consent collected.

Retention

• Test log (arrivals, submissions, time, source, word): contains no personal data and is kept indefinitely in aggregatable form. It is the research dataset, and its whole purpose is to compare rounds years apart.
• Session identifier: in the browser it lasts as long as the open tab and disappears on closing. In the log it remains as a random code with no link to a person.
• Sample card answers: they enter the research in aggregate form. For anyone who does not subscribe, they stay with no link to an identity.
• Subscriber data (email address, consents and labels): kept on Kit for as long as the subscription lasts.
• On unsubscribing: the address leaves the sending list. Linked data is removed on request and in any case within the periods required by law. Requests go to [email protected].

Publishing the results

Round results are published on the site, in the newsletter and in Keryx material in aggregate form only. We never publish individual answers. Every published cross-segment (age by occupation, for instance) requires at least 20 respondents: below that threshold the figure does not go out, because on small numbers a person can become identifiable.

Rights

The rights listed in section 7 apply, including access, erasure and objection. The labels attached to your address fall under the right of access: ask, and we send them exactly as they are. To exercise them, write to [email protected].

12. Project assistant on Telegram (clients only)

Clients with an ongoing engagement may be offered a project assistant on Telegram, Herman (@keryxdesignbot). It is an automated assistant based on artificial intelligence, not a person. It is activated only through a personal, single-use invitation sent by the controller. Using it is optional: everything that goes through the assistant can also be done by email at [email protected], and you lose nothing by not using it. This section covers the assistant only. It is separate from the Kira chatbot on the website (sections 2-8), which has its own legal bases and retention periods.

Data processed

• Telegram identifiers: numeric ID, display name, username, activation date
• Text messages, voice messages and their transcripts, photos and documents you send
• The summaries prepared for the controller
• A record of the notice shown at activation: the notice version and the date you tapped "I've read it, let's start". That tap is not consent: it is proof that the notice was shown to you
• A technical event log (for example a /umano or /cancella request, or a completed deletion) holding only type, date and short codes, never the text of messages

Purpose

Collecting information, material and requests about your project and preparing a summary for the controller. The data is not used for marketing, for profiling or to train artificial intelligence models.

Legal bases

• If you are the client as an individual (freelancer, sole trader, consumer): performance of the contract (Art. 6(1)(b) GDPR).
• If you write on behalf of a client company (contact person or employee): the controller's legitimate interest in managing the project with the client (Art. 6(1)(f) GDPR). You can object at any time (Art. 21 GDPR) and carry on by email.

How processing works

• Artificial intelligence: the conversation text is sent to Anthropic's Claude API to ask the project questions and write the summary. The assistant has no access to other systems, does not read the files you send and sends nothing to third parties.
• Voice messages: transcribed automatically by a speech recognition model run on Cloudflare Workers AI. The transcript is sent back to you so you can correct it. Your voice is not used to identify you.
• No automated decisions: the controller reads and assesses the summary. The assistant takes no decisions that produce legal or similarly significant effects on you (Art. 22 GDPR).
• Who reads it: only the controller. The providers listed below process the data through automated systems, on the controller's behalf.

Effect of messages

The assistant does not discuss prices, quotes, timing or terms, does not classify requests and does not accept approvals. Messages sent to the assistant do not replace the emails required by the Terms of work: approvals, closing the list of requests, warranty claims and withdrawal go by email. What you send the assistant is read by the controller, who gets back to you if needed. When it matters, for example at the end of a round of corrections, the controller may email you a summary of the requests collected, which you can add to or correct by replying by email. Requests sent to the assistant are never lost. The full rules are in Art. 13.4.bis of the Terms of work (Condizioni di lavoro).

The channel, and what not to send

Telegram chats with bots are not end-to-end encrypted. Do not send passwords or logins, verification codes, payment details, identity documents or tax codes, health data or other special-category data, criminal-record data, or data about your own customers or other people. Use email for these. If they arrive anyway:

• data that can be recognised automatically (payment cards, IBANs, credentials and keys, tax codes) is blocked before it reaches the artificial intelligence and is not stored;
• data recognised after processing is replaced by a placeholder;
• in both cases the message is deleted from the chat where Telegram allows it, and the controller is alerted. The controller removes anything left over by hand within 72 hours, for example inside photos or documents.

If the message contained a credential, change it straight away.

Telegram processes the data that passes through its channel as an independent controller, under its own privacy policy (telegram.org/privacy), and keeps a copy on its servers even when a message is deleted from the chat.

Providers

• Cloudflare, Inc. (USA): running the assistant, storing conversations and files (D1 database and R2 storage, configured with EU jurisdiction), voice transcription (Workers AI). Processor.
• Anthropic (Claude API, USA): text processing. Processor. Does not use data sent through the API to train models.
• Resend (Plus Five Five, Inc., USA): emailing the summary to the controller's mailbox. Processor.
• Google (Google Workspace): the controller's mailbox, where the copy of the summary arrives and project correspondence is kept.
• Telegram: communication channel, independent controller (see above).

Transfers outside the EU follow the safeguards set out in section 8.

Retention

• Voice message audio: deleted automatically 7 days after transcription.
• Conversation, transcripts, photos, documents and summaries held by the assistant: deleted automatically 30 days after the end of the project warranty period set by the Terms of work (Art. 10.bis). If the project ends without a delivery covered by the warranty, for example on withdrawal, the period runs from the close of the project.
• Files used for the work: copied into the project folder, and from there follow the rule in section 10.
• Copy of the summary in the mailbox: follows the project correspondence.
• Content that should not have been sent (see above): removed immediately or in any case within 72 hours.
• Notice record, event log and account data (Telegram ID, name, dates): the account is deactivated at the end of the project; the minimum record is kept as long as the conversation, as proof that the notice was given.
• If a dispute about the project is open, deletion of that client's data is suspended until it is resolved.

Deleting the chat in the Telegram app does not delete the copies kept by Keryx: for those, see below.

Your rights, and reaching a person

The rights listed in section 7 apply. In the chat you can type:

• /cancella to request deletion of your data. The request reaches the controller, who verifies your identity, carries out the deletion and confirms it in writing. The assistant deletes nothing on its own.
• /umano to alert the controller straight away. A person replies within 2 working days.

Or write to [email protected].