This case study presents an external project developed by Recursos en la Red for SPORT (Grupo Prensa Ibérica). It is included in the CARL Library as an example of data-driven innovation in the media industry and was not developed as part of the CARL project.
In a newspaper such as SPORT, used to working with intense news rhythms, constantly evolving sports news, and an increasingly diverse content offering, this coexistence between digital and print requires well-coordinated processes. The print edition requires selection, hierarchy, adjustment, and layout, while always preserving the outlet’s editorial identity and the final quality of the product.
This is the context for ESCRIBA, a project developed by Ediciones Deportivas Catalanas, S.A.U., publisher of SPORT, aimed at exploring how artificial intelligence can support certain tasks related to the composition of the print edition. The initiative is publicly funded through grants for the integration of artificial intelligence into the value chains of media outlets.
Use case on measurement, optimization, and observability in an artificial intelligence-assisted automated layout process
The closing of a print edition concentrates technical and editorial decisions into a very short period of time: content selection, hierarchy, text adjustment, images, advertising, templates, and final validation. ESCRIBA applies artificial intelligence to this process, but its value depends on something prior: using data to understand where friction occurs and how to measure whether automation truly improves the closing process.
The challenge
The digital edition is usually the starting point for news production, but the print version requires adapting that content to a medium with physical constraints: pages, modules, boxes, images, advertising, and design criteria.
ESCRIBA addresses this challenge through a layout tool connected to the editorial system, an automated layout algorithm, and generative AI capabilities to adjust headlines, subtitles, standfirsts, paragraphs, quotes, highlights, or translations when space requires it.
Data analysis makes it possible to turn a traditionally manual process into a measurable system: it identifies which decisions are repeated, which content requires the most adjustments, which templates work best, and how AI behaves under closing-deadline pressure.
The role of data analysis
1. Understanding the process before automating it
ESCRIBA starts from an analysis of the real closing workflow: how pieces from the digital edition are selected, what requirements print needs, which points generate the most manual intervention, and which editorial rules condition the composition.
This analysis makes it possible to build a catalogue of parameterized layouts, with boxes, styles, tolerances, content types, and clearly defined usage criteria. Without this structured database, the algorithm would not have a reliable solution space to work with.
Process data helps separate what can be automated from what must remain under editorial decision-making: AI proposes, adjusts, and optimizes, but the closing process retains human review.
2. Modeling layout as an optimization problem
Automated layout can be understood as a fitting problem: digital content must be placed on pages with limited space, design rules, advertising, and editorial priorities.
Data analysis makes it possible to quantify the variables: headline length, text volume, image proportions, sections, story types, page areas, reduction or expansion tolerances, and placement restrictions.
With this information, the algorithm can search for feasible solutions and optimize them iteratively. Generative AI provides additional flexibility by adapting content when a piece does not naturally fit into the available layout.
3. Calibrating cost, time, and quality
In the closing of a print edition, processing time is critical. A more powerful model may produce higher-quality adjustments, but if the response arrives too late, it is no longer useful for real operations.
That is why ESCRIBA needs to measure how long each type of intervention takes, how many iterations a page requires, which adjustments produce better results, and when it is advisable to limit the depth of the process in order to meet closing deadlines.
The cost-benefit analysis is not limited to the cost of the model: it also includes hours saved, reduction of repetitive adjustments, closing stability, less friction between digital and print, and the ability to maintain visual and editorial quality.
4. Visually and editorially evaluating the result
The quality of ESCRIBA’s output has two dimensions: the page must be well composed, and the adapted content must preserve meaning, tone, accuracy, and information hierarchy.
Tests must combine technical evaluation, visual review, and editorial review. Gaps, overflows, image adjustments, suitability of crops, headline coherence, and the need for subsequent intervention can all be measured.
AI evals and the use of LLM-as-a-judge can help compare versions of adapted headlines, subtitles, or paragraphs, generating quality metrics on representative samples. Final validation, in any case, belongs to the editorial and production teams.
5. Monitoring the closing process once the system is in use
Once deployed, ESCRIBA must be observed as an operational product. It is not enough to know whether it generates pages: it is necessary to measure whether it reduces iterations, whether it maintains quality, whether it is naturally adopted, and whether those responsible for closing trust its proposals.
Continuous monitoring makes it possible to detect operational drift: changes in content, templates, editing habits, or the quality of adjustments may degrade the system’s performance over time.
Analysis of feedback and interactions reveals opportunities for improvement: new rules, templates that should be reviewed, types of stories that require special treatment, or features that can be incorporated to facilitate the closing process.
Measurement map for the use case
| Analyzed dimension | What it makes it possible to measure | Decision it facilitates |
| Closing workflow | Identifying manual steps, friction points, and repetitive decisions. | Prioritization of automation and interface requirements. |
| Layout catalogue | Parameterizing boxes, styles, tolerances, areas, and restrictions. | Solution space for the automated layout algorithm. |
| Content adjustment | Measuring cuts, rewrites, headlines, images, and graphic elements. | Editorial and visual quality control. |
| Real operations | Observing closing times, iterations, acceptance, and corrections. | Continuous improvement and detection of degradation. |
Transferable learning
The main lesson from this case is that an artificial intelligence application in media must be managed as a measurable system. Quality cannot be presumed: it must be observed, evaluated, compared, and corrected. Adoption cannot be taken for granted either: it must be analyzed based on users’ real behavior, their feedback, and the usefulness that the tool demonstrates in the workflow.
This approach allows AI to evolve with the product. Historical data helps design the first version; usage data helps adjust the experience; quality evaluations help calibrate models and instructions; and observability makes it possible to detect degradation, opportunities for improvement, or new use cases.
Conclusion
ESCRIBA shows that AI applied to print edition requires much more than text generation: it requires data on design, process, content, time, and quality. Data analysis makes it possible to turn automated layout into a reliable, observable, and adjustable form of assistance, capable of reducing mechanical tasks without losing the editorial supervision that sustains the newspaper’s quality.
