Graphic design guide
Scope of the guide
Short answer: Yes, a draft of an AI-powered application, or even a simple first version, can be made; But for a mobile application that users can use safely, receive approval from the store and can be maintained for months, an expert still needs to make design, security and testing decisions.
I am Selçuk Aker; I am an Istanbul-based graphic and UI/UX designer who has been designing brands, catalogs and digital interfaces for over 19 years. The sentence I heard most frequently recently was: "We had the application done by artificial intelligence, the job is done." Most of the time it is not finished. The screen turns on, the buttons work, the demo looks good; But when the real user arrives, the store review begins, or the first update is required, deficiencies emerge. Then the promotion is prepared with artificial intelligence, the expected return does not come, months pass and eventually it starts over with an agency or an expert. This is exactly the cycle that small business goes through. In this guide, I distinguish where implementing with AI actually works and where it quietly creates problems.
What does implementing with artificial intelligence mean today?
"Applying with artificial intelligence" is not a single method. Today, three different approaches are described in the same sentence, and the results produced by each are different.
| Approach | What does it produce? | Where is it useful? | Where is it difficult? |
|---|---|---|---|
| Platforms that produce applications via chat | Described screens, simple database and a working preview | Making the idea clickable in a few days, collecting initial feedback | Complex business rules, authorization, platform dependency, ownership of code |
| No-code tools + AI plugins | Forms, lists, panel screens built with ready-made components | Internal process tools, appointment and registration forms, small team applications | Custom design, performance, store-specific requirements, scale |
| Code assistants (developer side) | Function, component, test and documentation drafts | Speeding up repetitive work for experienced developer | If there is no one to read and check the code, errors accumulate unnoticed |
The distinction in the table is important: Code assistant provides speed in the hands of someone who knows his job. In the hands of someone who cannot read the output, the same tool produces software that "seems to work" but no one knows the inside of. The problem is not with the tool itself, but with who controls the output and with what criteria.
Things AI does really well in app development
Scaring is not the purpose of this guide. The benefits of artificial intelligence tools are real:
- Speed: It used to take weeks to turn an idea into a clickable prototype; Today, the first version can be seen in a much shorter time.
- Draft and variation: Creating several alternatives of the same screen eliminates the fear of the blank page.
- Idea validation: It helps to understand early whether a business idea remains on paper or whether it really attracts attention.
- Repetitive code: Works such as form validation, list display, simple data transformations are accelerated in the hands of an experienced developer.
- Learning: It helps the founder understand technical concepts and communicate better with the developer.
What these benefits have in common is that they all occur at the beginning of the process. The problems grow at the end of the process, when it is said "let's go on air".
Difference between "working" and "ready to go live"
Artificial intelligence often produces the ideal scenario: The user enters the correct information, the internet connection is not interrupted, the list is always full. In real use, the real burden of the design is the situations that fall outside this ideal scenario.
Missing screen states
Situations such as empty list, loading error, unauthorized access, connection broken, operation interrupted etc. either do not exist in most produced applications or are glossed over with a general error message. When the user encounters the "Something went wrong" message, he does not know what to do and deletes the application. As mobile app designer, I determine these details in the flow map before drawing the screens; because adding it later would break both the design and the code.
Accessibility and usability
The problems I frequently see in manufactured interfaces are: low-contrast gray texts, small touch areas that are difficult to hit with the finger, situations explained only with color (such as red = error), icon buttons that read meaningless in the screen reader. Apple and Google's interface guides recommend minimum sizes for touch areas; WCAG defines a contrast ratio of at least 4.5:1 for normal text. These are not aesthetic preferences; determines whether the app is usable for an elderly user, someone looking at their phone in the sun, or someone with poor vision.
Errors specific to Turkish
Small but annoying errors occur in applications with Turkish content: the word "istanbul" entered in lowercase letters becomes "ISTANBUL" instead of "ISTANBUL" when converted to uppercase letters, "ı/i" order disrupts alphabetical lists, long Turkish words overflow from buttons, error messages that smell like machine translation. Artificial intelligence thinks mostly in English examples; Turkish localization requires a separate round of checks.
Security, personal data and KVKK
The most expensive problems are the invisible ones. Typical risks that may be encountered in applications produced with artificial intelligence:
- Data access left open: If database rules are not defined, a user can see another user's record.
- Keys embedded in the code: If the secret keys of payment, map or email services are written inside the application, someone examining the application can access these keys.
- Untested edge cases: Scenarios such as two reservations at the same time, canceled payment, deleted account have not been tested.
- Personal data obligations: An application that processes personal data in Türkiye is obliged to inform the user and keep the data secure within the scope of KVKK. Issues such as information text, situations requiring explicit consent, and data retention period are not automatically established correctly in the artificial intelligence output.
These articles are not legal advice; You'll need to talk to a lawyer and a developer who can do security testing for your particular situation. My responsibility on the design side is to set up the permission and approval screens in a way that does not mislead the user and to clearly show in the interface which data is requested and why.
Store approval and publication process
Most of those who want to make mobile applications with artificial intelligence underestimate the App Store and Google Play process. Stores store the application not only by whether it works or not; It also evaluates it according to criteria such as privacy policy, declaration of collected data, possibility of account deletion, content rules and minimum functionality. Applications that package only one website and have limited functionality are likely to be rejected. Store images, app description and screenshots are also a separate design work; These are the showcase of the application and the first step in the promotion phase.
Maintenance, updates and scaling
The application does not end the day it goes live. Operating system updates, new screen sizes, version changes of the libraries used and error notifications from the user require regular maintenance. A small change in a code base that was produced in one go by artificial intelligence and that no one has read thoroughly can break another place. Queries slowing down when the number of users increases, growing file storage costs and server settings require special expertise. Since exporting code is limited on some platforms, moving to another team in the future may mean rewriting the application.
Loop: What happens when "The application is finished"?
Example scenario: A neighborhood gym wants an app where members can book classes. The process goes in the order I often see:
- Production with artificial intelligence: An application with a course list and a reservation button will be released in a few days. The demo gets everyone excited.
- Publication attempt: Store review is getting longer due to lack of privacy statement and account deletion; Corrections are again made with artificial intelligence.
- Promotion with artificial intelligence: Social media images and advertising texts are produced rapidly; The visuals are similar to those of competitors, the brand does not have its own language.
- No returns: The number of downloads remains low, some of those who download get stuck on the first recording and quit. No measurement has been established to indicate at which step they are lost.
- Months are lost: Each problem is patched separately; a fix breaks another screen.
- Applying to an expert: Finally, an agreement is reached with an agency or designer-developer team; Most of the time, a significant part of the existing structure is reconstructed.
The real loss here is not money, but time and trust. A bad experience from first users leaves a lasting impression on the brand. I explained the entire cycle step by step in the guide thinking the job is done with artificial intelligence.
Who is it suitable for and who is it not for?
| Status | Does it make sense to start with artificial intelligence? | Note |
|---|---|---|
| Showing the idea to the investor or partner | Yes | It should be clearly stated that it is a prototype |
| Demand testing (MVP) with first users | Mostly yes | With measurement and feedback plan |
| Simple team tracking tool | Yes | If there is personal data, access rules should be checked |
| Application that processes payment, health or child data | Alone no | Security and legal review is a must |
| The application that will be the main customer channel of the brand | Alone no | Design system, test and maintenance plan required |
| Product expected to grow in the long term | Only in prototype stage | The actual structure must be built scalable |
Using artificial intelligence correctly with a designer: hybrid approach
It is wrong to reject artificial intelligence and to blindly trust it. The method I recommend is to use the tool where it is fast and leave the decisions to the human:
- Write the problem first: Whose job will the application make easier? Choose one main mission.
- Set up the flow with the designer: Draw the main task, alternative paths and error conditions. Let the screens come out of this flow.
- Accelerate the prototype with artificial intelligence: Once the flow is determined, the screens produced serve a purpose.
- Try it with real users: Have five-six people do the main task and note where they get stuck.
- Inspect and install the actual build: Let an experienced developer make security, data and performance decisions.
- Receive the design system: Document color, typography, components and situations; Let each subsequent screen speak the same language.
In the Yatcim yacht rental marketplace study, there were interconnected screens such as boat profiles, calendar and payment flow. In this type of product, rather than beautiful individual screens, it is important that every step between the user finding a boat, selecting a date, and completing the payment is consistent. AI can speed up the drafting of these steps; But designing connections and exceptions between steps is a matter of decision. I explained the details of this approach on my page UI/UX designer and in the article web and UI UX design with artificial intelligence. If you are considering the web side, the can a website be made with artificial intelligence guide, and for the visual side of the interface, the can a website be designed with artificial intelligence guide may be useful to you.
Checklist before going live
- Is the main task of the application written in one sentence?
- Are empty, loading, error and unauthorized states designed?
- Have text contrast and touch areas been checked?
- Have Turkish characters, capital letter conversions and long words been tried?
- Have the database access rules and secret keys been reviewed by a developer?
- Are the information text, privacy policy and account deletion facility ready?
- Are store screenshots and descriptions prepared in the brand's language?
- Are registration, first use and purchase steps measured?
- Do you own the code and design files and can you export them?
- Who will do the maintenance and updates after publication?
If you say "I don't know" to most of the items on this list, your application is not finished yet. This isn't bad news; Spotting it at the right point can save months. If you have a prototype produced with artificial intelligence or want to start from scratch, leave a short note on contact page; Let's evaluate the current situation together and talk about which part can be preserved and which needs to be rebuilt.
Frequently asked questions about building apps with artificial intelligence
Can applications be developed with artificial intelligence?
Yes, it can be developed, especially for prototypes and simple first versions. However, for a product that requires security, personal data, store rules and maintenance, the output needs to be overseen by an experienced developer and designer.
Do I need to know code to make a mobile application with artificial intelligence?
It may not be necessary for a simple preview. However, if you cannot read the generated code, it becomes difficult to detect errors, security vulnerabilities and unnecessary complexity. If you do not know the code, I recommend that you get support from an expert, at least during the review phase.
Will an application made with artificial intelligence be approved by the App Store or Google Play?
It can, but the fact that it is produced with artificial intelligence does not make approval easier. Stores look at criteria such as privacy statement, data usage, account deletion, content guidelines and functionality. These requirements are often missing in the first version produced.
Is the application made with artificial intelligence safe?
It is neither inherently safe nor unsafe; It depends on whether it is inspected or not. Do not collect user data until the data access rules, storage of secret keys, and authentication have been reviewed by a developer.
Does it make sense to use AI for MVP?
It often makes sense to test demand. The important thing is to accept that the MVP is an experiment, establish measurements, and re-plan the actual product on a solid structure according to the test results.
Is it possible to have an application made with artificial intelligence corrected by a designer?
Most of the time it is possible. First, the existing flow and screens are examined; In some cases, only the interface and situations are corrected, in some cases, re-configuring the flow is less troublesome. It wouldn't be right to make this decision without looking at the screens.
Will AI replace the application designer?
Changed the display rendering speed; But understanding which task the user abandoned and why, establishing the flow, and maintaining the design system still requires human judgment. The designer's job shifts to decision and control rather than drawing.
Does using artificial intelligence reduce the cost when making applications?
It can save time at the prototype stage. However, later correcting or rewriting an unaudited structure can undo the time originally saved. What reduces the cost is not the tool itself, but its use at the right stage.
Can I make promotions with artificial intelligence after making an application with artificial intelligence?
You can use it for sketches and variations; But advertisements with similar visuals and without a strategy can quickly consume the budget. I covered this topic in the advertising and social media design with artificial intelligence guide.
Do you have a design project?
This guide provides information about education and profession. You can share your existing content and delivery expectations for your design project.
- Communication purpose of the project
- Mediums to be used
- Needed deliveries
