Basic AR can use fixed markers, but modern AR relies on AI for recognizing surfaces and objects.

Is AR expensive to build?

Content creation is often the main cost, especially 3D models. Start small.

What is the difference from mixed reality?

Mixed reality usually means digital objects that interact with the real surroundings more closely. Definitions overlap.

AI Glossary / AI and machine learning / Beginner

Augmented reality

Definition

Augmented reality, or AR, overlays computer-generated images, text or information onto a live view of the real world, usually through a phone, tablet or headset.

What it means

AR adds digital content to what you see. A phone camera shows your living room, and a sofa you might buy appears in it at the right size. Navigation arrows float over the street ahead. A repair guide labels parts of the machine in front of you.

It differs from virtual reality, which replaces your surroundings with a fully computer-made world. AR keeps the real world in view and layers information on top.

AI matters to AR because the system must understand what the camera sees: where the floor is, where a table edge sits, which object is a face or a product. Computer vision and tracking models do that work.

Beyond placing objects, AI can add meaning, such as recognizing a plant and showing its name, translating a sign in your view, or highlighting a part that needs attention.

For a business, AR is mainly a way to show products in context, guide training or support, and reduce errors, but it needs content and hardware and is not a fit for every problem.

Marker-based AR anchors content to a known image or code, which is simpler and more reliable; markerless AR understands surfaces and spaces on its own, which is more flexible but harder.

How it works

A device camera captures the scene frame by frame. Software estimates the device's position and the shape of the surroundings, often by tracking visual features and using motion sensors.

Computer vision models detect surfaces, objects or faces, and the system anchors digital items to them so they stay put as you move.

A renderer draws the digital content, matches lighting and scale, and combines it with the live image on screen, many times per second, so it feels attached to the world.

Quality depends on lighting, camera quality, processing power and accurate tracking. Drift or jitter breaks the illusion, which is why good AR often needs careful design and testing.

Three practical examples

Retail: letting customers see how furniture or glasses look in their own space or on their face.
Field service: showing a technician the next step overlaid on the equipment they are repairing.
Education: bringing anatomy models or historic scenes into a classroom through tablets.

Think of it this way

Think of a pair of glasses with labels written on the lenses. You still see the street, but street names, opening hours and directions hover where you need them.

The limit: the labels are only as good as the system's understanding of the scene, and a wrong label hovering confidently on the wrong object is worse than none.

Worked example

a furniture shop trying AR previews

Suppose a small furniture shop adds an option on its website to preview a sofa in the customer's room through their phone.

The shop needs 3D models of its products at true size, which cost money to create, so it starts with its ten best sellers. The customer points the phone at the floor, and the sofa appears.

The shop measures whether previews lead to fewer returns and higher conversions, comparing products with and without the feature over two months. Any figures here would be the shop's own, not general claims.

It also checks that the feature works on common phones and has a plain image fallback for those that do not.

Illustrative case study

A fictional equipment servicing firm trials AR instructions for a complex maintenance task. Technicians see highlighted parts and step prompts on a tablet.

Error rates on the trial task drop, but the firm finds that creating overlays for each machine model is the real cost. It decides to use AR only for its three most complex, high-value repairs.

The invented lesson: AR can help, but content creation and fit with the task decide its value.

Common misconceptions and mistakes

AR and VR are the same. AR keeps the real world in view; VR replaces it.
It needs a headset. Many AR experiences run on an ordinary phone or tablet.
It is only a gimmick. Where context matters, such as previewing size or following steps, it can help; it is wasted where a photo would do.

Business use

Consider AR where seeing something in context changes a decision or reduces errors: product previews, training and guided repair.

Pilot with a few products or tasks, measure returns, errors or training time, and keep a simple fallback for people without suitable devices.

Start where a phone is enough. Headsets add immersion but also cost, hygiene and adoption questions, so they suit shared workplace settings more than consumer use.

Check your understanding

Why might an AR preview fail even with a good 3D model?

Answer: The system may track the room poorly because of lighting or camera limits, so the object drifts or sits at the wrong scale and customers stop trusting it.

Related terms

Computer vision
3D model
virtual reality
mixed reality
object detection
tracking

Learn next

AI for Business Owners helps you judge which emerging technologies are worth a pilot: https://aiforbusinessandlife.com/academy/ai-for-business-owners. [C2]

Sources

[S2] Google, Machine Learning Glossary; [S14] KNIME, Data Science Glossary. Full URLs are in the source ledger. Verified 11 October 2026.

Previous: Augmented Lagrangian Next: Aumann's agreement theorem