EU AI Act: 6 Steps for Turkish Companies Selling to Europe

The EU AI Act became generally applicable on 2 August 2026. Being established in Turkey does not automatically place a company outside its reach. A company may need to assess the Act when it offers an AI system to people in the EU, deploys a system that will be used in the EU, or produces outputs with effects in the EU. This article is not legal advice. It gives product, sales, and delivery teams a practical starting point for their first review.

The European Commission expects people to know when they are interacting with an AI system and, in defined cases, when content was generated or altered by AI. That makes compliance more than a contract review at the end of a deal. It affects the product interface, onboarding, content workflow, technical documentation, and customer training.

When does the EU AI Act matter to a Turkish company?

The first question is usually about market effect rather than company headquarters. A Turkish software company may deploy a support assistant for its EU customer. A consulting team may help implement a prediction system for a customer's European operation. A SaaS product may add generative AI to the workflow of EU users. In each case, write down the company's role first: provider, deployer, importer, authorised representative, or organisation using the system.

This role mapping helps teams assign responsibility. Using a model supplied by another company does not transfer every product decision to that vendor. A team that adds a large language model to a customer support interface still owns the user experience, data flow, human oversight, and information it gives the customer. Model-provider documentation is an input to this work. It does not replace the product team's review.

The Commission's implementation timeline shows that the Act applies in stages. Prohibitions and AI literacy provisions have applied since 2 February 2025. General application and certain transparency obligations started on 2 August 2026. Some rules for high-risk systems apply later. Teams should therefore avoid labelling every product as high risk from the start. Review the intended purpose, user group, and role of the system in a decision.

What product inventory should you create first?

Sales teams need a short inventory that engineering can defend when an EU prospect asks questions. The inventory is not a marketing document. It records where customer data comes from, what the system produces, and where a person can intervene.

For every AI-powered feature, record these questions:

• Does a user interact directly with the AI system?
• Does the system generate text, images, audio, or video?
• Will the output be published as public content?
• Does the system affect hiring, credit, health, education, or another sensitive decision?
• Is an output applied automatically, or does a person approve it?
• Does the system use personal data, confidential customer data, or special-category data?
• Which suppliers provide the model, infrastructure, and data sources?

This inventory does not slow down sales. It prevents sales teams from guessing when a customer asks where data travels or who makes the final decision. It is also the starting point for designing data security in an AI project.

What should users be told?

For interactive AI systems within scope, people should be able to understand clearly that they are interacting with AI rather than a human. The Commission's transparency guidelines expect this to be visible in the design. A small line in terms of service may not give users useful information at the moment they need it.

In practice, a support assistant should identify itself in its first message. A voice assistant should provide a notice at the start of the interaction. If a user can request a human agent, that route should be easy to find. The notice should use the user's language and appear when the user is deciding whether to rely on the system.

For example, a customer-service assistant might say: "You are chatting with Binode's AI support assistant. It prepares answers from our knowledge base. You can ask for a team member." This is only an example. The wording must match the product's actual behaviour. A product must not promise human escalation when no escalation path exists.

Marking requirements for AI-generated or AI-altered images, audio, and video also belong on the product roadmap. Public-interest text published without human review or editorial control may also need disclosure. The Commission's Article 50 Q&A gives more detail by system type. Teams selling a content studio, social publishing tool, or document automation product should include disclosure fields in the publishing workflow instead of relying on a manual reminder at the end.

What changes at proposal and contract stage?

Evidence of compliance should not appear for the first time when a contract is ready to sign. Sales and product teams need to work from the same version before a proposal goes out. Claims such as "fully automated decisions", "complete accuracy", "human-like answers", or "fully compliant" create risk when they do not match technical reality. Teams should describe only features they can test and processes they can monitor.

Add these items to the proposal checklist:

• System description and intended use
• Customer role and provider role
• Data processing, retention, and access boundaries
• Point where human oversight operates
• Owner of incident, error, and challenge records
• Process for notifying model or infrastructure changes
• Limits on the use of AI outputs

These items do not replace a contract. Legal counsel should assess the wording for the customer's country and use case. Early collection still reduces late security and compliance discussions that often delay a deal.

How can a team demonstrate AI literacy?

The AI literacy provision of the EU AI Act has applied since 2 February 2025. The Commission's AI literacy Q&A explains that there is no single mandatory certificate level. Organisations are still expected to take measures that help staff understand the opportunities, limits, and risks of the systems they use.

A one-hour general presentation will rarely be enough for a team that sells and operates AI features. Training should follow roles. Salespeople need to know what the product does not do. Customer-success staff need to know how to record harmful or incorrect outputs. Engineers need to handle data boundaries, model changes, and evaluation scenarios. Leaders need visibility into decisions affected by automation.

A participant list is useful, but it is weak evidence by itself. Short scenario tests, in-product warnings, an updated use guide, and a record of improvements after incidents are more useful. This approach also gives the company a durable base for its AI literacy programme.

What can be delivered in six weeks?

This plan is designed for a small or mid-sized Turkish software team preparing to sell into the EU. If the use case could be high risk, legal and compliance specialists should join in week one.

Week	Deliverable	Owner	Review question
1	Product inventory	Product lead	Which feature uses AI?
2	Role and data map	Engineering	Where does data come from and go?
3	Interface and content notices	Design	Can users see they are engaging with AI?
4	Human-oversight flow	Operations	Who can stop and review it?
5	Sales and training pack	Sales lead	Does every team make the same claim?
6	Evidence file and test	Leadership	Are changes recorded?

The purpose is not to produce a thick compliance folder. The purpose is to establish a working practice that answers customer questions consistently and updates evidence when the product changes. Those documents also speed up decisions in a project that connects an AI assistant to existing systems.

Frequently asked questions

Is a Turkish company outside the EU AI Act?

A product used in the EU or an output with effects in the EU can require a scope assessment even when the company is established in Turkey. Review the role and intended use in writing.

Is every generative AI feature high risk?

Classification depends on the use case. Start with what the system does, who it affects, and where it sits in a decision process.

Is a small chatbot notice enough?

The notice should be understandable and appear when the user needs it. Test the interface, user flow, and escalation option together.

What records should we keep for team training?

Keep a role-based training plan, attendance record, short scenario assessment, and current use guide. Update the material when the product changes.

Companies selling AI solutions into the EU should treat compliance as part of delivery. A product inventory, clear user notice, human oversight, and role-based training belong in the same operating model. Binode can review your product, data, and operations flow and turn that review into a delivery plan.

Author: Abdullah Oğuz works with organisations on AI projects that connect process, data, and product goals.