Audience measurement systems in Spain, led by the Asociación para la Investigación de Medios de Comunicación (EGM) and Kantar, show a profound disconnect from the emergence of artificial intelligence as a new form of audience. In the case of the EGM, we are dealing with a system that relies on face-to-face interviews, showing logos on a tablet and calls to landlines and mobile phones to measure an ecosystem that today is dominated by bots.
EGM and Kantar face the AI challenge
Judging by what AIMC describes on its website, its methodology is almost artisanal, with face-to-face personal interviews, supervisors who travel to homes to verify answers and a team that shows logos on a tablet to refresh the interviewee's memory. It certainly seems like a system designed to represent the Spanish population with enviable sociological precision, but the problem is that, in 2026, the population is no longer the only consumer of media.
More than 10,000 AI bots visit Observatorio Blockchain in just 24 hours
While the EGM strives to ask a citizen whether they remember having seen a newspaper's logo on their screen, that same newspaper's servers are being hit millions of times by artificial intelligence agents that extract content, analyze it and reuse it to generate answers that users then consume. These agents do not answer surveys, have no physical home and do not remember logos, but they use information constantly.
Media are accessed through AI
The EGM itself acknowledges in its technical characteristics that it is a population study based on recall. It seeks to represent the individual through what they say they have seen, read or heard. But what value does human recall have when information consumption is no longer direct, but mediated by systems that process content before it reaches the user?
Today, a growing share of the millions of Spaniards that Kantar and the EGM say they represent access media through artificial intelligence agents. It is these systems that read, select, process and synthesize the information. The human receives the final result, which may be a data point from the outlet, a summary or an answer, but does not see the outlet behind it.
The key point is that consumption is no longer direct. Previously, the user went to an outlet, saw the masthead and associated the information with a brand. Today, in many cases, that process is carried out first by an AI. The AI agent accesses several outlets, selects the information, synthesizes it and presents it as a single answer.
Disconnect from reality
The user does receive that content. They read it, use it, even make decisions based on it. But they do so in an environment where the original brand has disappeared or been diluted. There is no homepage, no visible logo, no browsing through the outlet. That is why, when the EGM interviewer shows them a newspaper's logo on their tablet, the respondent says they have not seen it. And their answer is correct from their experience, because they have never entered that outlet, never interacted with it directly. However, the information they received does come from there.
The content has been consumed, but the outlet has not been recognized by the respondent. From the EGM's logic, this translates into an audience decline. If there is no recall of the outlet, there is no consumption. However, in real terms, the content has been used, because it has been selected by an artificial intelligence, processed and delivered to the end user. The outlet has fulfilled its role as a source, but it has done so outside the visible circuit that the measurement system is able to record.
The result is an obvious disconnect, since it is the outlet that has fed the machine, but the system is unable to recognize it. For the EGM, the audience falls. However, value has been generated and the content has been consumed. The problem is that the system is still looking for signs of human recall in an environment where consumption is already mediated by machines.
There is no doubt that the EGM makes an evident effort to capture the citizen's voice and understand what they say they consume. But today, a growing share of access to information no longer passes directly through people. It passes through artificial intelligence systems.
These systems act as intermediaries and are the ones that access content, select which information is relevant, process it and deliver it to the end user in the form of an answer, but they leave no trace in traditional measurement systems. If the EGM wants to remain useful, it is not enough to ask a person whether they have read a newspaper. It also needs to know how many times that content has been used by artificial intelligence systems. Because that is where a growing share of consumption is happening. The EGM measures a user's declared intention, based on their memory, but the real use of content by machines is not asked about: it is recorded. And today, that use is not in the system.
The EGM aims to offer a single data point for the advertising market. But that data point, in the current context, is incomplete, because it reflects people's behavior but leaves out the role of artificial intelligences as new consumers and distributors of content. What remains to be known is to what extent an outlet is being used by machines.
That is where new metrics are beginning to appear, such as the SAS (Synthetic Authority Score), which measures neither recall nor awareness among humans, but the probability that a piece of content will be selected and used by an AI. Because in this new environment, it is no longer enough to be remembered. It is equally or more important to be used.
5,920 meters measuring where consumption no longer happens
At the same time, Kantar, formerly Sofres, continues to project audiovisual consumption across an entire country based on fewer than 6,000 meters installed in households. It is a model that starts from the premise that consumption happens in the living room, through a screen watched by a person. The problem is that this premise no longer describes the whole system.
A growing share of consumption no longer happens when someone watches a video, but when an artificial intelligence accesses that content, analyzes it and uses it to generate answers, recommendations or new products. That access does not pass through the home or through a viewer's gaze. It passes through automated systems that operate beyond the reach of meters.
Kantar continues to measure what a person sees, but does not measure what machines use. And today, in many cases, it is those machines that determine which content ultimately reaches the user. The meter measures the screen, but artificial intelligence decides what appears on it.
What Kantar cannot measure with its meters is the use that artificial intelligences make of audiovisual content. AI access to content is the new click, which happens when an AI enters a YouTube video or an outlet's investigation to analyze it and use it. This does not happen in the 5,920 households of Kantar's panel; it happens in data centers. By focusing only on the eye of the viewer on the sofa, Kantar is ignoring a majority share of the automated activity that, according to Cloudflare, is what now consumes much of the bandwidth on the internet.
Radio: content without an antenna
The impact of artificial intelligence is not limited to the written press or television. Radio is also entering this process, although in a less visible way. Traditionally, the value of radio was tied to direct listening, where the listener turned on the dial, stayed on the broadcast and, in that context, received both the content and the advertising. Measurement based on recall made sense because consumption and the channel coincided.
That link is breaking. Today, artificial intelligences access programs, interviews or talk shows, transcribe them, process them and extract structured information from them. That content becomes text, data and answers that are served to the user in other environments. The user no longer needs to listen to the program. They receive the content directly. And in that process, the distribution channel that is essential to the radio business disappears.
There is consumption, but without a broadcast, without an ad spot and without measurement. Radio content continues to generate value, continues to inform and continues to influence, but it does so outside the channel that allows it to be monetized. The user consumes the information without going through the station, without hearing the advertising and, in many cases, without knowing where it comes from.
This leaves radio in a particularly vulnerable position. Unlike other media, its model depends almost entirely on direct listening. When that listening disappears, so does the basis of its business. The measurement system, however, continues to ask about programs listened to. But increasingly, content is not listened to. It is used.
Without monetization
All of this means that, as of today, media outlets are paying for a measurement system that sends supervisors to homes to verify whether a person remembers having listened to a radio program. Meanwhile, a growing share of the value of their content shifts toward AI systems that access that content, process it and use it without that use being measured or monetized.
The EGM tells the outlet who remembers it, but does not tell it who is using its content or where it should capture value. In 2026, managing a media outlet with only meters and surveys is like trying to measure internet use by counting only visits to web pages, ignoring everything that happens inside apps.