Flarseo
AI transparency
Content produced by the platform is generated with artificial intelligence systems and is disclosed as such. This page explains how it works, sector by sector.
Courtesy translation. Only the Italian version of this document has legal value: if the two texts diverge, the Italian text prevails. Italian law applies to the relationship in any case.
Why this page exists
Flarseo is a product built around artificial intelligence systems. Hiding that would be unfair to customers and to the people who read what gets published.
This page explains where AI steps in, how we disclose it, which controls we impose in sensitive sectors, and where our responsibility ends and yours begins.
It is written to be read before subscribing, not after a problem.
What the assistant generates
The assistant works on the customer brand and produces editorial and promotional material.
- Editorial plans: topics, calendar, priorities, the goal of each piece.
- Articles: structure, body copy, headings, meta descriptions, internal links.
- Posts for social channels, with copy and visual direction.
- Copy and structure for a website, when the customer asks the platform to build one.
- Analysis: local search rankings on Google, visibility inside conversational assistants, comparison with competitors.
Commercial decisions stay with the customer. The assistant proposes, writes and, within the limits described below, publishes.
Which models we use
We do not use a single model for everything. We orchestrate large language models supplied by third parties, choosing for each task the one best suited on quality and cost.
| Function | What the model does |
|---|---|
| Planning | Reads the brand context and proposes topics, priorities and an editorial calendar. |
| Writing | Produces the draft following the planned structure and the brand tone of voice. |
| Automated review | Checks coherence, repetition, fidelity to the brief and compliance with editorial rules before a human sees the text. |
| Analysis | Interprets ranking and visibility data and turns it into practical guidance. |
We do not publish model brand names here because they change faster than this page does. The current picture of providers and functions is available on request, and the categories are described on the subprocessors page.
Supplier categories and location
How we disclose content
Content generated with AI systems is disclosed as such. The disclosure travels with the published content and is not an option the customer can switch off for convenience.
The form of the disclosure depends on the risk profile of the sector: in the simplest cases it is a note on the artificial origin of the content, in sensitive ones it also names the person who reviewed and approved the text.
At signup the customer expressly acknowledges that content is generated with artificial intelligence systems and that it is disclosed as such.
The four risk profiles
Every sector maps to one of four profiles. The profile is shown during signup, before the customer enters any data, and it drives three things: whether human review is mandatory, how far the assistant may act on its own, and how content is disclosed.
| Profile | Example sectors | Human review | Maximum autonomy |
|---|---|---|---|
| Standard | Retail, crafts, restaurants, tourism, personal services. | Recommended, not imposed. | Full: the assistant can plan, write and publish. |
| Regulated | Activities subject to specific advertising or licensing duties. | Mandatory on content touching the regulated aspects. | Limited: automated planning and writing, publication after approval. |
| Health | Medical and dental practices, health professions, wellbeing with health claims. | Mandatory on every piece, by a competent person. | Reduced: no autonomous publication. |
| Finance and law | Financial, insurance and tax advice, law firms. | Mandatory on every piece. | Reduced: no autonomous publication. |
The health, finance and law profiles correspond to what editorial practice calls YMYL: content that can affect the health, the money or the rights of the reader. They deserve a different level of care, and the platform enforces it in code, not only in guidance.
What human review means
Human review means that an identified person reads the content in full and approves it before publication, taking editorial responsibility for it.
It is not a confirmation click: the platform records who approved, when, and against which version of the text.
In the profiles that require it, publication stays technically blocked until approval. There is no shortcut around it, and asking us to remove it is not a request we can grant.
The reviewer must have the relevant expertise. In the health profile, for example, review of clinical content belongs to someone qualified to do it.
Autonomy caps
Autonomy measures how far the assistant may act without asking. The customer chooses it at signup, but cannot exceed the cap imposed by the risk profile of their sector.
- With full autonomy the assistant plans, writes and publishes on the agreed calendar.
- With limited autonomy the assistant prepares everything and stops before publication.
- With reduced autonomy every piece waits for an explicit approval, and nothing goes out without it.
The customer can always pick a more cautious level than the one allowed. They cannot pick a more permissive one.
Who the publisher is
The platform produces and, where authorised, publishes. The publisher remains the customer: the website, the social profiles and the domain where the content appears are theirs.
It follows that responsibility for what is published on those properties sits with the customer, towards third parties, towards the public and towards authorities.
Our role is to provide a tool that lowers the risk: editorial structure, automated checks, mandatory review where it matters, disclosure of artificial origin. It is not, and cannot be, a guarantee that every single statement is correct.
Duties when publishing AI generated content
No content is professional advice
This needs saying plainly, because it is where the most damage happens.
No content generated by Flarseo constitutes medical, legal, tax or financial advice, and none of it replaces the opinion of a qualified professional. This applies to content published by customers and to the guidance the platform shows inside the interface.
In the health, finance and law profiles the platform inserts a statement to that effect in published content. The customer must not remove it.
Well written content can sound authoritative even when it is wrong. That is exactly why human review is not negotiable in sensitive sectors.
Asking for human review
Even outside the profiles that require it, a customer can ask for human review to be switched on for their workspace.
- From the interface: lower the autonomy level of the assistant and publication goes back to requiring approval.
- On request: write to info@flarseo.com to have mandatory review enabled for the whole workspace or for specific content categories.
- On already published content: ask for a specific text to be reviewed again, giving the page address and the reason.
If published content contains an error, tell us: we correct it and, if the error came from a defect in our system, we write it up in the review of the problem.
Your content and model training
We do not use customer content to train models, ours or anyone else.
We require terms from model providers that exclude the use of transmitted data for training their systems.
One customer content never feeds another: workspaces are separate and there is no shared model built on your texts.
Known limits of AI systems
We would rather say it upfront. Language models have structural limits that no orchestration removes entirely.
- They can produce plausible but false statements, including figures, dates, quotations and legal references.
- They can attribute to a source something that source does not say.
- They can reflect biases present in the data they were built on.
- Their knowledge of the world has a cut off date and does not match current events.
- Given the same request they can give different answers.
That is why the platform runs automated checks on the text before a human sees it, and why fact checking before publication remains a duty of the customer.
Legal reference
The reference framework is Regulation (EU) 2024/1689, the European regulation on artificial intelligence.
Article 50 is the relevant one: it places transparency obligations on those who use AI systems to generate or manipulate content, requiring that artificially generated or manipulated content be made recognisable as such to the people who receive it.
Our architecture is built to meet that obligation on behalf of the customer, by disclosing the artificial origin of published content. The customer duty is not to remove that disclosure.
This page describes how the product works. The contractual commitments on the use of AI, including the constraints tied to the risk profile of your sector, are in the framework agreement and its annexes, delivered by email at signup, in Italian.
Is your sector in a sensitive profile?
Talk to us before signing up: we will tell you exactly which constraints apply to your case and what daily work looks like with mandatory review switched on.