From SEO to GEO at Esade: what 150 executives asked about AI and their brands

October 11, 2026 (today)

On September 17 the Esade Alumni Marketing Club filled the Sala Polivalente in Building 3 at Esade Barcelona with one practical question: how do you manage your web traffic when people ask ChatGPT before they ask Google?

There were five of us on the panel. Xavier Arderiu (Mediapro, chair of the club) moderated. Raimon Casals explained what PepsiCo is doing and Mireia Vázquez Molina did the same for Banc Sabadell. Paula Segarra spoke for Trustpilot, one of the review sources the models read when they judge a brand, and I covered measurement with GEORadar. Susana Prado, General Manager of Inetum Catalunya, the session sponsor, gave the closing remarks.

Download the full panel highlights (PDF, 8 pages).

A format that forces you to be specific

There was very little talking from the stage. After a short introduction, each speaker spent eight minutes at every table taking questions, then moved on. The room held 150 people, almost all senior executives from banking, insurance, healthcare, food, public institutions and transport.

That set the tone. Nobody asked what GEO is. People asked where to start on Monday, how many prompts they need, how long it takes to see a change, and how to fix a negative narrative that comes from old content.

What the five of us agreed on

The game has moved from visibility to narrative. Keywords, backlinks and paid reach no longer guarantee a recommendation. A model recommends what it trusts, and that depends on reputation, on the attributes tied to the brand and on the quality of the sources it reads.

You diagnose with data. A single prompt tells you nothing, because answers change with the user, the location and the model's temperature. You need thousands of prompts segmented by persona, funnel stage, product and geography. At GEORadar our studies run from 6,000 to 30,000 prompts.

You compete on attributes. Map which attributes each model links to each brand, look for the free or less contested ones, and skip the battles a competitor has already won.

You feed the sources. Media, reviews, directories and sector blogs. Paula Segarra made the point at every table: models treat review platforms as a proxy for real experience, and volume, recency and response rate end up in how an AI describes a brand.

The bottleneck is inside the company. Mireia Vázquez Molina put it plainly for Banc Sabadell: the hard part is getting legal, comms and product to move together and reviewing all the content that already exists.

Results take months. The first visible changes show up around day 90. Complex programmes scale in 180.

What I would do on Monday

  1. Run a first diagnostic with 6,000 or more segmented prompts across ChatGPT, Gemini and Claude.
  2. List the ten sources the models cite for your category and check, for each one, whether you are there and whether it is accurate. Trustpilot included.
  3. Pick three attributes you can win and drop the ones a competitor already owns.
  4. Name one person who owns the AI narrative, with a say over comms, legal and product.
  5. Measure again on day 90.

My closing line was this: you do not need more content, you need the right content in the places the model reads. What stayed with me was hearing five very different organisations describe the same shift from five different seats.

Sources