AI in digital signage is software that analyses data about screens, content and viewers, and uses it to decide what plays where and when, or to produce new versions of the content.

This article is for digital signage network operators, retail and restaurant marketing teams, and the integrators who build CMS software for them.

Two years ago, AI in this industry was mostly a forecast. At ISE 2026, most large CMS vendors showed AI features. Generating an image or a video is now the easy part. Deciding which content to show needs data that most networks keep in separate systems, and cameras on screens come with legal duties in the EU.

AI content selection pays off when it starts with data the network already has – proof of play, sales and context – and adds cameras only with anonymous processing on the device and the notice the AI Act requires.

Questions this article answers:

  • What is AI in digital signage?
  • How does AI select content for digital signage screens?
  • What data does AI content analysis need?
  • What does the AI Act change for cameras on screens?
  • How can you start with AI in digital signage?

AI in digital signage in 2026

AI in digital signage covers three jobs: creating content, deciding which content plays, and running the CMS on the operator’s behalf. The first and the third are now standard features of commercial platforms. The second, analysis and selection, depends on each network’s data, so it is still built project by project.

Generative AI in digital signage CMS platforms

Content generation reached mainstream CMS platforms in 2026. Samsung added AI content generators to VXT, and other vendors at ISE showed text, image and video generators built into their editors. For a marketing team, this means a new promotion can get several layouts and copy variants in minutes, in the brand template, without a designer for each version.

Generated variants are useful mainly as material for selection. When a CMS has four versions of the same offer, it can test which one works on which screen. The next step, AI assistants that create campaigns and schedules on request, is covered in our article on digital signage AI agents.

Content analysis and selection: where AI makes decisions

Content selection means choosing, for each screen and time slot, the item from the playlist that has the best chance of reaching its goal: a sale, a scan of a QR code, a shorter queue. Content analysis is the part that measures what happened after each item played. The two work as a loop: the system plays, measures, and changes the playlist.

How AI selects content for screens

Selection methods range from simple rules to models that learn from results. Most networks need a mix, and the order below is also a sensible order of implementation.

Rules and triggers: the starting point

Rules link content to conditions the system can read: time of day, weather, stock levels, prices from the point of sale (POS) system, or local events. How a CMS pulls these values in is described in our article on real-time data feeds for digital signage. A burger chain shows hot drinks below 10 °C, a pharmacy hides a product that is out of stock, and a cinema switches the screen to the next showing. Rules are predictable and easy to audit, and they produce the first clean data about what played under which conditions.

Do you want to add AI content selection to your digital signage CMS?
Leave your e-mail and we will reach out to you!

A/B tests and self-optimising playlists

An A/B test shows two versions of the same content on comparable screens and compares the results. A self-optimising playlist goes one step further and uses a multi-armed bandit, an algorithm that gradually gives more airtime to the versions that perform better while still testing the others. On a self-service kiosk in a restaurant, this could mean showing three photos of the same burger and moving towards the one that gets tapped most often.

The result has to be measurable on the screen or close to it: a tap, a QR scan, a sale of the promoted product in that store and hour. Without that signal, the algorithm has nothing to learn from.

Recommendations on kiosks and menu boards

Recommendation models suggest the next item based on what is already in the basket or what other customers ordered with it. They work best on interactive devices, such as ordering kiosks and self-service checkouts, because the device knows the current order. A customer with a burger and no drink gets a meal offer, and one who already has fries gets a dessert.

Audience-triggered content from sensors

Sensors add information about the people in front of the screen: how many there are, how long they look and, with cameras, their approximate age group. The CMS can switch content when the audience changes, for example to a family offer when a group with children approaches. How this measurement works, and how accurate it is, is described in our article on sensor-based audience measurement. Of all the methods, this one comes with the most legal requirements, covered further below.

Data behind AI in digital signage

The main limit of AI in digital signage is data. Netflix can personalise covers because it knows what each user watched and skipped. A digital signage screen in a shop does not know who is looking at it, so it has to rely on data about places, times and sales.

Data the network already has

Most networks already collect more than they use:

  • Proof of play – a log of what played on which screen and when.
  • Sales data from the POS system – per store, product and hour.
  • Interaction logs – taps on kiosks, searches on wayfinding maps, QR code scans.
  • CMS metadata – content tags, campaign dates, screen locations and formats.
  • Context – weather, opening hours, local events, stock levels.

Proof of play joined with sales data answers the basic question of content analysis: did sales of the promoted product change in the hours when it was on screen? This needs no camera and no personal data.

Comparing screens with control groups

A rise in sales during a campaign does not prove that the screen caused it. Prices, weather and season change at the same time. A reliable analysis compares stores or screens that showed the content with similar ones that did not, over the same period. AI helps here by matching comparable locations and by finding patterns in thousands of plays, but the test design is a business decision that has to be agreed before the campaign starts.

Top 6 digital signage trends for 2026 – read the article
Top 6 digital signage trends for 2026 – Read more

AI Act and GDPR: cameras, emotions and audience data

Camera-based analytics is where screen software meets EU privacy law. Two regulations apply at once: the GDPR, because a face in a camera image is personal data, and the AI Act, which regulates specific uses of AI.

Age and gender estimation: a notice obligation since 2 August 2026

The AI Act treats sorting people into groups such as age or sex based on their biometric data as biometric categorisation. Under Article 50(3), which has applied since 2 August 2026, whoever uses such a system or an emotion recognition system has to inform the people exposed to it. For a screen with a camera, the notice belongs at the screen itself, where people see it before they walk into the camera's view; a line in the privacy policy is not enough.

Emotion recognition: banned in workplaces and schools

Reading emotions from facial expressions, the "sentiment analysis" that many articles promised for digital signage, is the most restricted use. Article 5 of the AI Act has prohibited it in workplaces and educational institutions since 2 February 2025. Elsewhere, it is listed as a high-risk system, and after the AI Omnibus amendment the high-risk obligations apply from 2 December 2027. A retail network would have to meet risk management, documentation and human oversight requirements, which is rarely worth it for choosing an advert.

Anonymous counting and processing on the device

The safer design processes the camera image on the media player or the sensor itself and sends only counts to the CMS: how many people, how long they looked, an approximate age group if needed. Images are never stored or sent. The GDPR risk drops, but it stays. Systematic monitoring of a publicly accessible area on a large scale requires a data protection impact assessment under Article 35(3)(c) of the GDPR, so the legal review should start before the hardware is ordered.

How to start with AI in digital signage: a pilot

The quickest results come from the data described above: proof of play, sales and context. A rules engine with A/B testing on part of the network shows within a few weeks whether content changes move sales, and it builds the clean data that later models need.

profile_image
Book your consultation
Choose a time slot and schedule a 30 min free consultation with Slawomir Wilusz
Calendly right-arrow

A pilot on one group of screens

We add AI content selection to existing CMS platforms, and we start with a pilot on one group of screens, because the results depend on each network's data. For the pilot, we need:

  1. A proof of play export from your CMS for the last few months.
  2. Sales or interaction data for the same screens and period.
  3. The content library with tags, or access to it in the CMS.

In return, you get rules and tests running on the pilot screens, a comparison with a control group, and an estimate of what it would take to cover the whole network.

Digital signage software development for integrators and network operators – see the page
Check out: digital signage software development for integrators and network operators

FAQ about AI in digital signage

What is AI in digital signage?

AI in digital signage is software that analyses data about screens, content and viewers to decide what plays where and when, or to generate content variants. In most networks it is a module of the CMS that schedules the screens, and it does not need cameras to work.

How does AI select content for digital signage screens?

It combines rules (time, weather, stock, prices), tests that compare content versions, recommendations based on the current order on kiosks, and, optionally, audience data from sensors. A self-optimising playlist gives more airtime to the versions that perform better while still testing the others.

What data does AI content analysis in digital signage need?

The basis is proof of play joined with sales or interaction data for the same screens and hours. Context data such as weather, opening hours and stock levels improves selection. Camera data is optional and comes with legal obligations.

What does the AI Act change for cameras on digital signage screens?

It adds duties on top of the GDPR. Since 2 August 2026, people in front of a screen that estimates age or gender or recognises emotions must be informed about it (AI Act, Article 50(3)). Emotion recognition is banned in workplaces and schools. Large-scale monitoring of public spaces also requires a data protection impact assessment under the GDPR.

Can AI in digital signage generate content?

Yes. Several commercial CMS platforms, including Samsung VXT, have built-in generators for text, images and video. Check two things before you use them on a large network: who approves generated content before it goes on air, and whether the brand's fonts and colours are locked in the templates.

How can you start with AI in digital signage?

Start with a pilot on one group of screens, using data you already have: proof of play, sales and context. Rules and A/B tests on that group show whether content changes affect results, before you invest in sensors or custom models.