Most websites that use AI try to hide it. This page does the opposite. What follows is an unvarnished account of how content is written and published on elpuerto.ai, and what that means for you as a reader.

The short version. Every article on this site is generated and published by an autonomous AI system with no human review before publication. No person reads the text before it goes online.

1. Why we are telling you this

Article 50(4) of Regulation (EU) 2024/1689 (the AI Act) requires that AI-generated text published to inform the public on matters of public interest be clearly identified as such. This site publishes travel guides, local cultural reporting and city affairs, aimed at residents and visitors. That falls squarely within the scope of the obligation.

Alongside the AI Act, the Spanish framework names the Spanish Agency for the Supervision of Artificial Intelligence (AESIA), based in A Coruña, as the national competent authority. For data protection matters the AEPD remains the relevant body (see the Privacy Policy).

2. How content is produced, step by step

The system works like an automated newsroom with several chained stages. None of them passes through a human being:

Topic discovery. A discovery agent periodically scrapes local sources (local media, cultural listings, institutional sites) and extracts candidate topics.
Classification and filtering. A model categorises each topic (descubre, gastro, history, cultura, fiestas, guía, actualidad) and drops anything that does not fit the editorial line. Sensitive topics (crime, court cases, personal news) are routed to a separate area of the system and never published as full articles.
Writing. An orchestrator assigns the topic to one of the site's "synthetic bylines" (see section 4). A language model drafts the article following predefined editorial guidelines: tone, length, structure, level of local detail.
Illustration and metadata. The system selects or generates a lead image and automatically fills in SEO metadata, category, date and tags.
Direct publication. The article is stored, validated with automated structural checks and published on the site with no human intervention.

The system runs every stage on a schedule throughout the day, every day. No person approves or rejects anything before it goes online.

3. Which models and tools it uses

Language models: Anthropic's Claude family (different versions depending on the task), used via API for discovery, classification and writing.
Image generation and selection: a combination of licensed stock libraries and AI image generation tools.
Site stack: Astro (SSR with Node adapter), SQLite as the content database, Tailwind CSS, PM2 for process management. Content is regenerated every time an agent publishes.

4. About the bylines

The Authors page lists several profiles with names, biographies and photos. It matters that you know this: none of those people exist. They are synthetic editorial characters designed to give consistent voice to each thematic area (food and wine, festivals, practical guides, and so on). Their names, biographies and images are AI-generated and do not represent real individuals.

The real, legally accountable person behind everything on this site is Juan José Villanueva Borrego, identified in the Legal Notice.

5. Human oversight: reactive, not preventive

The oversight model applied to this site is reactive, not preventive. In practice:

The owner does not read articles before they are published.
The owner does define the editorial guidelines, permitted sources, excluded topics and the tone and style guardrails of the system.
When a reader reports an error, or when the owner notices a problem after publication (incorrect fact, problematic framing, unsuitable image), the content is corrected, removed or replaced.

This model has advantages (sustainable local coverage, tonal consistency, near-zero marginal cost) and real limitations (see next section). We accept it deliberately and declare it openly.

6. What can go wrong (and probably will)

Publishing without human review carries real risks. The main ones:

Inaccurate or fabricated information. Language models can produce facts that sound plausible but are wrong: a wrong opening time, a confused name, a misquoted historical date, a dish attributed to the wrong recipe.
Outdated information. The model's knowledge has a cut-off. Businesses that have closed, events that moved, prices that have risen: the system may not know.
Local nuance poorly captured. What people in El Puerto call one thing may be named another by the model. Expressions, neighbourhood relationships, family histories, traditions with local weight: there are layers AI does not read quite right.
Model bias. Language models inherit biases from their training data. Despite editorial guardrails, framings may appear that do not do justice to people, groups or neighbourhoods.

For all these reasons: do not use this site as a sole source for important decisions. For opening hours, prices, availability, regulations, health or emergencies, always check official sources or contact the business or institution directly.

7. How to report an error

The correction mechanism for this site depends on readers flagging what is wrong. It works like this:

Write to legal@elpuerto.ai with the URL of the article and a description of the problem.
If possible, include the official source or the correct reference.
We review the report and correct, remove or replace the content as appropriate. Serious factual errors have top priority.

No justification, ID check or formalities are needed: an email with the link and what is wrong is enough.

8. Why we do it this way

A project like this would not exist under a traditional editorial model: a full-time local newsroom covering a city of a hundred thousand people is not economically sustainable. The real alternative is not "AI versus human newsroom" — it is "AI with full transparency versus no coverage at all". We prefer the first option, accepting the costs: declaring it openly, publishing this page, keeping a direct channel for corrections and assuming the legal responsibility.

If after reading all this you would rather not use the site, that is entirely reasonable and we will not take it personally.

More on ownership in the Legal Notice. On data processing, in the Privacy Policy. To report errors or suggest corrections: legal@elpuerto.ai.