Libelium + iris360: a new web architecture model for search engines and AI

Libelium.com and iris360iot.com launch an architecture where every piece of data lives in one place and reaches people, search engines and AI assistants

Case studies

Libelium, a Spanish end-to-end IoT hardware manufacturer, set out to overhaul its digital presence from the ground up. After reviewing where to focus its engineering and marketing efforts, the company adopted a new web architecture model for Libelium.com and iris360iot.com, built to support a demanding technical catalogue, strengthen organic search visibility and match how people look for information in the age of AI.

A catalogue with a single library index

In a library, every book has one card in the index, and from there you reach the shelf, the loan desk and every bibliography that cites it. Libelium wanted the same for its catalogue: each sensor described once and discoverable from any page, search engine or AI assistant.

Growing across search engines and AI assistants

An expanding industrial catalogue multiplies the number of product sheets, verticals and use cases living on the site. When an engineer looks up a specific sensor or compares solutions, the content has to respond quickly both on Google and inside assistants such as ChatGPT or Claude.

Libelium's marketing team spotted that shift early. Traditional organic traffic now shares the stage with AI-generated answers, where a search engine or language model pulls specifications, product relationships and use cases straight from structured dataGlosarioJSON-LDFormato de datos estructurados basado en JSON que describe el contenido de una página con el vocabulario de Schema.org. Se inserta en un bloque <script type="application/ld+json"> del HTML. Buscadores y asistentes de IA lo leen para identificar entidades como un producto, un artículo o una ruta de navegación sin interpretar el diseño de la página..

Building a solid presence there means working on three fronts together: foundational technical SEO, readability for assistants (AIOGlosarioAIO y GEODos capas de optimización que se suman al SEO clásico. AIO (AI Optimization) busca que el contenido sea comprensible para asistentes de IA mediante estructura clara, entidades definidas y datos legibles por máquina. GEO (Generative Engine Optimization) busca que una marca aparezca y se cite en las respuestas de los motores generativos. Las dos dependen de los mismos cimientos técnicos que el SEO, como el marcado estructurado y un archivo llms.txt.) and accurate citation in generative engines (GEO). That goal shaped the new architecture, where every page is linked directly to the product or solution entityGlosarioEntidad (Entity)El núcleo inamovible de la lógica de negocio en la Arquitectura Limpia. Una Entidad encapsula las reglas y datos empresariales más fundamentales y generales (las reglas de negocio críticas). Pueden ser objetos con métodos o un conjunto de estructuras de datos y funciones, y son completamente ajenas a cualquier cambio externo, como modificaciones en la base de datos, en el sistema de vistas o en el framework de desarrollo. it describes.

Separating content from data

The turning point came from separating editorial copy from the data that powers it. Every piece of data lives in one single place and flows everywhere from there: to the page a visitor reads, to internal search and to any crawler or assistant querying the site. Related content is connected by topic, so someone landing on a case study can reach the exact solution they need without getting lost.

I explain the underlying pattern in separating the public page from the data entity and shared taxonomies and content relationships (in Spanish).

Structured data and a map for AI models

With the product entity connected, every page outputs its own JSON-LDGlosarioJSON-LDFormato de datos estructurados basado en JSON que describe el contenido de una página con el vocabulario de Schema.org. Se inserta en un bloque <script type="application/ld+json"> del HTML. Buscadores y asistentes de IA lo leen para identificar entidades como un producto, un artículo o una ruta de navegación sin interpretar el diseño de la página. markup generated from the exact same data source the visitor sees. A product, a vertical solution and a case study declare their properties, relationships and availability explicitly without duplicate maintenance.

On top of that semantic layer sits an llms.txtGlosariollms.txtArchivo de texto en la raíz de un sitio que ofrece a los modelos de lenguaje un índice resumido de sus páginas más relevantes. Funciona como un mapa legible por IA, comparable a un sitemap pensado para asistentes. Un generador propio puede construirlo a partir del modelo de contenido del sitio en lugar de mantenerlo a mano. Más detalle en WebMCP y llms.txt. index mapping the site's key sections, making it easier for search engines and assistants such as ChatGPT or Claude to identify and cite the catalogue accurately. The layered setup is detailed in SEO, AIO and GEO: layers and structured data (in Spanish).

Custom, auditable components

Relying on third-party page-builder plugins tied the content to external formats and update cycles outside the project's control. To remove that dependency, they were replaced with a custom component library maintained inside the repository, with auditable code designed directly around Libelium's data model.

All existing content was migrated to the new blocks before launch, preserving every page intact while gaining full control over performance, layout and future changes. The criteria behind that call are in custom code vs third-party dependencies (in Spanish).

Performance from the ground up

A full Lighthouse crawl and production measurements pinpointed two clear bottlenecks: listing pages were loading the full body of every post even though they only rendered a card, and server responses lacked object and page caching to absorb repeated requests.

Replacing the listing templates with summary cards and enabling layered caching produced measurable gains across three levels:

  • Payload and visual stability on listings: the heaviest route cut its HTML size by 77%, eliminated layout shift completely (CLSGlosarioCLS (Cumulative Layout Shift)Cumulative Layout Shift (CLS) es una métrica de Core Web Vitals que mide la frecuencia y la magnitud con la que los elementos visibles en la pantalla cambian de posición de forma inesperada durante la carga de la página. from 0.801 down to 0) and more than tripled its Lighthouse performance score.
  • Server response time (TTFB): the homepage dropped from ~0.7–0.9 s to ~0.46–0.6 s after adding object and page caching.
  • Core Web Vitals in production (mobile): in Google Search Console, URLs flagged as "Poor" fell from ~450 in August to 0 in November.

Fixing the shared template improved an entire group of archive pages in one stroke. The full method is in archive template performance with visual builders (in Spanish).

Launching without losing a single signal

Launch is the moment that causes the most nerves in any website migration. Years of search equity depend on every old URL finding its place on the new site. I audit every URL across both brands with an automated, repeatable crawl, and each valuable address is redirected to its new destination in a single 301 hop, the cleanest way to preserve ranking signals.

The inventory, redirect and QA workflow is documented in the URL migration checklist (in Spanish).

Technical specs you can trust

In an industrial hardware company, a single wrong spec can cost a sale or an engineer's trust. That is why I set a source-of-truth rule before rebuilding the catalogue: official technical documentation takes precedence, and anything without backing does not get published. Every figure is verified and written in one place. The principle is covered in source-of-truth rules for resolving data conflicts (in Spanish).

Working with an AI agent under clear rules

I work alongside an AI agent during development, which multiplies what one person can ship in seven weeks. The key is discipline: versioned rules in the repository and a dry-run, backup and single-apply write pattern so every change stays reviewable and reversible. I walk through that setup in AI agent with written rules and safe writes (in Spanish).

Verified results

After launch and the first production audits across Libelium.com and iris360iot.com, the numbers confirm the impact of the new architecture:

  • Mobile Core Web Vitals: URLs flagged as "Poor" in Search Console dropped from ~450 in August to 0 in November.
  • Server response time (TTFB): object and page caching brought homepage TTFB down from ~0.7–0.9 s to ~0.46–0.6 s.
  • Listing templates: 77% reduction in HTML weight on the heaviest route and zero layout shift (CLS from 0.801 to 0).
  • Strategic indexation: 26 out of 26 key URLs for solutions and products indexed cleanly, with a 5-language hreflang setup and segmented sitemaps.
  • Crawl control: effective blocking of noise parameters (?nocache=) and consolidation of variants through clean canonicals, preventing duplicate indexation.
What changedBeforeNow
Core Web Vitals (mobile)~450 URLs flagged "Poor"0 Poor URLs and CLS 0 on listings
Technical informationWritten as plain text on each pageConnected as data and verified against its source
Editorial contentMultiple formats mixed in one spaceEach type structured and linked by topic
Page componentsTied to third-party pluginsCustom library, auditable and maintained in-repo
Search engines and AIIdentical markup across all pagesPage-type structured data and AI index
Crawling and indexationParameter noise and fragile migration26/26 key URLs indexed and single-hop 301 redirects

What comes next

With the architecture in place and 0 Poor URLs in Core Web Vitals, the next phase is already underway: trimming mobile LCP on pages still marked "Needs improvement", finishing indexation of the newly migrated case studies and tracking qualified traffic and conversion trends on the new foundation.

The takeaway is straightforward: when architecture starts from the business's data, every later improvement in content, SEO or AI stands on solid ground.

Dealing with a similar bottleneck?

If your corporate site has outgrown its structure, or you need to prepare it for search engines and AI assistants without losing hard-won rankings, I can help. Take a look at my software architecture design service or tell me about your project.

Frequently asked questions about the project

Related services · Industrial IoT