Gains:
- It distinguishes which tasks in the public workflow AI accelerates and which rights-bearing decisions it should only assist in.
- It applies a three-step verification reflex to each AI output, consisting of source attribution, independent verification and appreciation filtering steps.
- It recognizes public equality, transparency, accountability and KVKK red lines and internalizes that the responsibility lies with the public official.
You are at the service desk of a district governor's office. On the one hand, citizens waiting in line at the toll booth, on the other hand, hundreds of petitions waiting to be answered; On the one hand, the activity report requested by the director, and on the other hand, an officer trying to understand how a new regulation changes the old practice. Public administration (the work of the state and local governments in producing services to citizens, distributing resources and implementing legislation by using public power) is a field that inherently operates with large volumes of text, strict rules, expectations of equality and high accountability. Artificial intelligence (AI - software that can extract patterns from historical data and produce or classify text, tables and codes) accelerates you in this abundance of text and workload. But the very beginning of this module is clear: the AI is an assistant, blueprint generator, and triage tool; You are the competent public official who ensures the rights of the citizen, the correct implementation of the legislation and the final say on the administrative decision.
In this first unit we will focus on discipline, not the tool. We will learn where AI saves real time in the government workflow, where it is dangerous, how to verify each output, how to protect citizen data, and how to ensure transparency and fairness. Without laying this foundation, subsequent units are left hanging in the air — because an unverified output in the public is not just a wrong answer; It is a mistake that violates a citizen's rights, misallocates a resource, or puts an institution under legal liability.
Where does AI come in handy in the government workflow?
Let's divide public sector jobs into two large clusters. First cluster: voluminous, repetitive, text-intensive, pattern-extractable works. The first draft of answers to similar citizen questions, the summary of a long regulation, the translation of a petition into official correspondence language, the categorization of applications in an Excel spreadsheet, the extraction of decision items from a meeting minutes, the first skeleton of an activity report. In these jobs, AI reduces hours to minutes and does not get tired.
Second cluster: decisions that use public power, creating rights or restricting rights. Rejection or acceptance of an application, imposition of a penalty, whether a social aid is deserved or not, legality of a zoning decision, application of a legislation to a concrete case, evaluation of a tender. These decisions require discretion, legal responsibility and justification. Administrative action (an action of a public authority that produces legal consequences with its unilateral will; for example, granting or canceling a license) must be justified and auditable. Here, AI multiplies options, produces drafts and checklists — but the signature, justification and responsibility are yours.
Let's clarify the distinction in one sentence: AI is strong on "what's in this pile and what does the first draft look like" questions; The decision is yours when it comes to questions such as "Is this decision legal, fair and can I defend it?"
Tip: Before outsourcing a task to an AI, ask: “What does the citizen lose if this output is wrong?” If the answer is "a few minutes of rewriting", delegate comfortably. If the answer is "an unfair rejection, a wrong punishment, or a loss of rights", let the AI produce a draft; You make the decision, the justification, and the verification.
Verification discipline: three steps
AI produces fluidly and confidently; That doesn't mean it's true. AI occasionally produces hallucinations — that is, it presents a non-existent law, a repealed regulation, a fabricated court decision, or a false number as real. In an administrative letter or information given to citizens, this leads to both loss of rights and distrust of the institution. Apply a three-step reflex to each output:
- Connect it to the source. Every legal claim of AI must be based on an official source: the Official Gazette, Çözüm.gov.tr, the institution's own circular. "Which article of which law is this and is it in force?" Ask and see for yourself in the official text.
- Verify with independent source. Confirm the date, number and rate given by the AI from the institution's own records or official statistics. Don't trust a single source.
- Pass it through the filter of appreciation. Does the output take into account the characteristics of the concrete event? Are there any exceptions? Does it comply with the principle of equal treatment? Your expert and administrative judgment is the final filter.
Attention: "The AI wrote so" is not a justification and does not sustain an administrative action. If there is a false article, a fabricated case law, or a leaked personal data, the responsibility belongs not to the AI, but to the public official and institution that uses that output without verifying it.
Four public red lines
The use of AI in the public sector, unlike the private sector, is framed by four additional principles: equality (similar treatment of a citizen in a similar situation), transparency (the justification of the decision and, if possible, the method should be explained), accountability (each transaction has a responsible person and an audit trail) and protection of personal data (KVKK - Personal Data Protection Law No. 6698). Citizen's T.R. to an AI vehicle. haphazardly pasting identification number, health or criminal information; leaving a decision unjustified because "the algorithm said so"; disadvantaging a group of citizens because of model bias — all of which is both unlawful and erodes public trust.
three mini cases
Case 1 — Triage saved time. A municipality's solution center received approximately 3,200 requests per week. AI-supported classification divided the requests into 12 categories and directed them to the relevant directorate; average first response time dropped from 34 hours to 6 hours. But requests labeled "urgent/security" were always manually checked by an officer.
Case 2 — Verification caught a hallucination. An officer asked YZ about the relevant article of a regulation. YZ said, "30-day period in accordance with temporary article 3." When the civil servant looked at Çözüm.gov.tr, there was no such temporary article; The period was 60 days. The linking step to the source prevented notifying the wrong time to the citizen.
Case 3 — Return from privacy breach. To be quick, a new staff member uploaded a list of 400 welfare applicants (name, ID number, income) directly into a public AI tool and requested a summary. The chief realized: personal data had been transferred to an external system. The process was stopped; The data was anonymized and reworked with only categorical fields, and the incident was recorded in KVKK records.
Four copyable templates
1) Job suitability assessment:
Your role: senior public administration specialist.I will describe the job below. Tell me (1) whether this job is a drafting/triage job that can be delegated to the AI or a rights-creating/limiting administrative decision, (2) the cost of incorrect output to the citizen and the organization, (3) the verification I need to do before and after delegation. Job: [insert job here]
2) Obligation to link to source:
I'll ask you a legal question. In each answer, MUST specify the legislation you rely on: law/regulation name, article number and validity status. Do not make any claims that do not have a source. Mark the places you are not sure of as "must be verified from an official source". Fabricating a non-existent article or decision.
3) Personal data masking:
I will give you a reference text. Before starting the analysis, mask all personal data: name-surname -> [PERSON], TR ID number -> [IDENTITY], address -> [ADDRESS], phone -> [TEL]. Generate categories and summary only from masked text. Never write real personal data on the output.
4) Justification framework (for administrative action):
Your role: legal counsel experienced in administrative action writing. Produce a draft justification for the following event: (1) factual event, (2) applicable legislation (article by article, to be verified), (3) application to the concrete case, (4) conclusion. Add a note that this is a draft and each item must be confirmed by the official source. Incident: [in summary]
Weak prompt / Strong prompt
Weak: "How should I respond to this citizen's petition?"
Güçlü: "Your role is an experienced editor-in-chief. Produce a draft response to the following petition in official correspondence language. Specify the relevant legislation with the article number, but add a note 'to be confirmed from the official source'. Mask the personal data. Explain in a polite and understandable language which right the citizen can exercise and how. At the end of the draft, put a list of 3 items that I need to check before signing."
The difference: the strong prompt role redefines the format, regulatory discipline, confidentiality and authentication; the output directly approximates a usable draft.
Where is AI strong, where is the decision yours?
business
AI contribution
final decision
First answer to the citizen question
Draft, language simplification
Accuracy of information, confirmation
Regulation summary
Quick summary, ingredient list
Enforcement and interpretation
Application classification
Category suggestion
Borderline/exceptional cases
Reason for administrative action
skeleton, tongue
Legality, signature
Report draft
Structure, first text
Data accuracy, conclusion
Common mistakes
- Putting the output in the official post without verifying it. The highest risk of hallucination is in legislation and numbers; Be sure to confirm both.
- Uploading personal data to a public tool. KVKK violation and loss of trust. Mask first or use in-house/approved tool.
- Blame the decision on the "algorithm". The reason and responsibility for administrative action is human; "The system said so" is untenable.
- Losing sight of equality. AI producing different drafts for two citizens in the same situation may lead to discrimination if not manually controlled.
- Making AI the sole source. Do not proceed without confirming with a second official source.
In summary
AI is a great accelerator in the public domain: generating outlines, summarizing text, classifying references, marking patterns. But the justification, verification and signature of every decision that uses public power, creates or restricts rights belongs to the competent public official. Three-step verification (link to source, independent verification, review) and four public principles (equality, transparency, accountability, personal data protection) are the backbone of this module.
Application task
Choose an actual mission from your unit (e.g., answering a common citizen question). Determine whether this job is a transferable or critical decision using the “Job suitability assessment” template above. If transferable, produce a draft, apply three-step verification, and note at least one error you find during verification.
checklist
- [ ] I distinguished whether the work was a transferable draft or a critical decision.
- [ ] I linked each legal claim to the official source and checked its validity.
- [ ] I masked personal data or used an approved tool.
- [ ] I determined the reason for the decision and the person responsible as human.
- [ ] I observed the principle of equal treatment.
- [ ] I confirmed the output with a second source.