Gains:
- It extracts the request and duration from an incoming petition and produces a formally correct response draft with AI.
- It adapts the same content to different interlocutors such as sub-units, citizens and higher authorities in an appropriate style and maintains the integrity of the content.
- It checks formal correspondence format rules and catches any conditions or exceptions that the AI drops when simplifying.
The language of public administration is writing. Incoming petition, outgoing response, consent, approval, minutes, cover letter, notification, circular, internal correspondence... A significant part of a public official's day is spent reading and producing texts. These texts have both a legal value and a strict form: official correspondence (correspondence between public institutions and citizens according to certain procedural and formal rules; regulated in Türkiye by the "Regulation on the Procedures and Principles to be Applied in Official Correspondence") requires certain title, number, date, address, subject, interest and signature elements. Here AI saves you time by taking over this repetitive style work and first draft; But the legal content, justification and final version of the text are your responsibility.
In this unit, you will see how to speed up the entire correspondence flow with AI, from understanding an incoming petition to producing a draft of the response letter, from checking compliance with the official format to language and style, from personal data cleaning to final reading before approval.
Understanding incoming text: summary and demand extraction
When a long and messy petition arrives, the first thing to do is to clearly understand what the citizen demands. Citizens often bury the real demand in the middle of the text when explaining their problem; sometimes there are multiple requests. AI is powerful here: it summarizes the text, bullets the requests, suggests which unit it concerns, and flags missing information if any.
But beware: AI can sometimes "complement" a request that is not in the text or skip a request. Therefore, always compare the issued list of demands with the original petition.
Tip: When asking for a summary, ask the AI for the rule "only remove requests that are clearly written in the text, do not make inferences, add requests that do not exist." This way you reduce the risk of fabricated claims.
Draft of the answer text: form + content
Response writing consists of two layers: form (heading elements, address, interest, language, closing) and content (legal and administrative response to the request). AI can produce the form almost perfectly because this pattern is clear. It also produces the content as a draft; but every legal basis in the content—law, regulation, deadline, fee—requires verification (see Units 1 and 4). Use AI with full confidence as a format generator and in draft-and-verify mode as a content producer.
Step by step correspondence flow
- Summarize the incoming text and extract the request (fabrication prohibited, faithful to the source).
- Identify relevant legislation and internal procedure (to be verified).
- Produce a draft answer (formal format + plain but formal language).
- Check personal data and confidential information (writing unnecessary data).
- Follow the format and language checklist (address, interest, date, place of signature).
- Verify legal context and clarify rationale.
- Submit to supervisor for approval; Notify after approval.
three mini cases
Case 1 — Workload decreased. A civil servant in a provincial directorate was preparing an average of 22 response letters per day; each lasted ~25 minutes. By handing over the format and first draft to the AI and focusing solely on content and verification, time per post was reduced to 9 minutes; The officer devoted the time he saved to the pending objection files.
Case 2 — Skipped request caught. In a long petition, the citizen requested both the zoning status and a fee refund. At first, the officer would only respond to the zoning request; When the list of demands issued by YZ showed the second request, both of them were answered in a single letter, and the citizen did not have to apply a second time.
Case 3 — Privacy bug averted. In a response draft, AI had unnecessarily included the health information of a third party in the file. It was noticed and removed during the pre-approval check. Lesson: the draft may contain non-relevant personal data; Data cleaning is a must before approval.
Four copyable templates
1) Petition summary and claim extraction:
Your role: editorial specialist. Process the petition below: (1) 3-4 sentence summary, (2) list item by item ONLY the requests clearly written in the text (don't make inferences, add requests), (3) suggest which unit/legislation it concerns (to be verified), (4) mark if there is missing/unclear information. PETITION: [paste text]
2) Response draft (official form):
Your role: official correspondence specialist. Produce a DRAFT response to the following request in accordance with the formal correspondence style: interest, subject, body (request - basis - conclusion), closing. Keep the language formal but simple. Write "[to be confirmed]" next to each legal basis. Write actual personal data, replace with [PERSON]/[IDENTITY]. REQUEST: [summary] CONTEXT: [internal procedure/legislative note]
3) Format and language control:
Check the draft of the official letter below and provide a list: (1) missing format elements (interest, subject, date/number place, signature), (2) expressions contrary to official language, (3) ambiguous/contradictory sentences, (4) unnecessary personal data. Don't rewrite the text, just give a list of corrections. DRAFT: [text]
4) Tone softening (negative answer):
Rewrite the following rejection/negative response in a respectful and descriptive language that does not offend the citizen, WITHOUT CHANGEING the legal content and justification. Specify your right to object and its duration with the note [to be confirmed]. TEXT: [negative response draft]
Weak prompt / Strong prompt
Weak: "Reply to this petition."
Güçlü: "The role is the official correspondence expert. First, make out the requests in the petition, item by item, without making any inferences. Then, produce a draft response in accordance with the official correspondence format; establish the interest-subject-body-closing structure, keep the language formal but simple. Add '[to be confirmed]' next to each legal basis, mask the real personal data with [PERSON]/[IDENTITY]. At the end, for the approval of the superior." "Make a list of 5 things I should check before submitting."
Difference: powerful prompt chains request inference, format, validation, privacy, and assertion control into a single flow; The output is a draft that can be directly revised.
How reliable is AI in which correspondence business?
business
AI trust
note
Formal form (concern, topic, closing)
high
The pattern is clear
Language simplification/tone
high
Content should not change
Demand inference
medium
Compare with original text
Legal basis / duration / fee
low
Must be confirmed by official source
Justification and appreciation
low
human responsibility
Style, formality and addressee setting
In official correspondence, the same information is written in different styles depending on the addressee and purpose. The reason for rejection written to a citizen is understandable and respectful; The request for opinion written to a higher authority is measured and justified; Instructions written to a subordinate unit must be clear and imperative. Style (the degree of formality of the text, tone, and distance from the interlocutor) is the dimension that the AI adjusts most easily but requires the most control — because a language that is too formal alienates the citizen, a language that is too friendly discredits the institution. The Regulation on the Procedures and Principles to be Applied in Official Correspondence determines formal obligations such as title, number, date, attachment and signature block; The AI fills out this skeleton, but you control the format fidelity.
Mini case — one text, three interlocutors. A district governorship would announce a legislative change to three separate addressees: 14 municipalities (official instruction), approximately 900 tradesmen (SMS/announcement, plain language) and the governor's office (information note). Officer derived three versions with AI from a single core text; The duration decreased from 2 hours to 25 minutes. But the phrase "urgently" in the tradesmen's announcement was simplified and concretized as "by August 15 at the latest", because the principle of plain language excludes vague official patterns.
Template that adapts the same content to three interlocutors:
Task: Convert the following core information into 3 versions.Core: [amendment/decision text]Version 1 — Instruction to the subordinate unit: clear, imperative, clause by clause.Version 2 — Announcement to the citizen: plain language, short sentence, what to do + deadline.Version 3 — Information note to the higher authority: measured, justified, reference clause stated.Rule: Adding a made-up date/article; If it remains unclear, write [TO BE FILLED].
interlocutor
tone
length
biggest risk
subunit
commanding, clear
short
excess of authority
citizen
Respectful, simple
short
incomprehensibility
higher authority
Measured, reasoned
medium
baseless claim
Caution: Stylistic adaptation should not change the content. When AI says "simplify" it sometimes means dropping an exception or condition; Always compare the simplified text with the original decision/legislation.
Common mistakes
- Leaving the request to the AI's inference. AI can make or break demand; Compare the list with the original petition.
- Putting it in writing without verifying the legal basis. Item number and duration are the areas most susceptible to hallucinations.
- Leaving unnecessary personal data in the draft. Perform data cleansing before approval; Keep only necessary data.
- Changing the content while softening the tone. The reason for rejection and the legal outcome must remain constant; only the tongue softens.
- Forgetting to write down the right of objection and its duration. It is often necessary to specify the remedies in negative proceedings; confirm the time.
- Skipping the confirmation step. The AI draft is never communicated directly; Supervisor approval and proofreading are required.
In summary
AI saves the public official significant time by taking on the burden of formatting and first drafting the correspondence. Almost completely reliable at format generation, strong at language simplification; But the request extraction must be compared with the original text, each legal basis must be verified from the official source, personal data must be cleared and the final text must be approved by the superior. The draft is AI's, the rationale and signature are yours.
Application task
Receive a real (but with masked personal data) petition. Extract requests with the "petition summary and claim extraction" template and compare it with the original text, noting whether there is at least one difference (added/omitted request). Then produce a draft with the "Answer text draft" template and check your own draft with the "Format and language check" template.
checklist
- [ ] I compared the requests with the original petition (no fabrications/omissions).
- [ ] Official correspondence format elements are complete (interest, subject, date/number, place of signature).
- [ ] I have confirmed every legal basis from the official source.
- [ ] I removed unnecessary personal data from the draft.
- [ ] I have specified the right and duration of objection in the negative action (confirmed).
- [ ] I submitted the text to the approval of the superior; I did not notify you directly.