AI prompt generator for public sector teams: a prompt builder for ChatGPT prompts that follow AI use policies

This free, AI-powered tool turns a task in your agency or department into a structured instruction for generative AI: a notice for residents, a report or memo, an RFP or procurement file, a reply to residents or a public project plan. It’s built for public servants, analysts and program managers who want to use tools like ChatGPT or Claude without putting personal data or case files at risk. It runs in your browser and asks for your email to show you the result. It’s intuitive, but below you’ll find instructions, a real example and tips to get more out of it. If you want to go further, the Founderz Certificate Program in AI and Innovation covers AI tools, automation and responsible AI leadership.

AI prompt generator for public administration

Prepare drafts and work structures for administration without replacing the case file, applicable rules, competent review, or administrative judgment.

Plain languageReportPublic projectPublic policyOpen dataTraining

Brief or context

Public taskAuto / not yet definedDrafting in plain languageStructuring report or memoPreparing terms or case fileResponding to citizensAnalyzing public policyDeveloping strategic planTransparency and open dataManaging public projectTraining in digital and AIPreparing institutional communication

Type of administrationAuto / not yet definedCity councilCounty or island councilRegional governmentNational administrationAutonomous agencyPublic companyMinor local entity

Area or departmentAuto / not yet definedSocial servicesEmployment and trainingUrban planning and housingEducationCulture and sportsEnvironmentCitizen servicesDigital transformationPublic procurement

RecipientAuto / not yet definedGeneral publicPerson with accessibility needsInternal staffManagement teamCollaborating entityNews media

Stage in procedureAuto / not yet definedPrior informationInternal draftProcessing underwayConsultation or participationResult notificationMonitoring and evaluation

ToolMicrosoft Copilot in authorized environmentChatGPT Enterprise / authorized environmentClaude in authorized environmentGoogle Gemini in authorized environmentOther approved tool

⚙ Advanced settings (7)

Legal framework and currencyAuto / not yet definedIdentify rules and date to verifySort provided materials onlySeparate legal requirement from recommendationDon’t cite regulations without verificationPending competent review

Data and privacyAuto / not yet definedNo personal data or case filesAggregated or anonymized dataVerified public informationSensitive data: don’t processEvaluate permissions first

Language and accessAuto / not yet definedPlain language and easy readingMultichannel versionAccessible for screen readerTranslation pending reviewFormal administrative record

Review and signatureAuto / not yet definedReview by competent unitLegal validationData protection validationPolitical or executive reviewPending responsible party

DeliveryAuto / not yet definedClear draft and checklistIndex, draft, and pending itemsProject plan and risk matrixAnalysis framework and sourcesPublication plan and controlsSession, cases, and protocol

Limits, available materials, or elements to avoid (optional)

Quality controls

Do not introduce personal data, case files, or confidential informationDo not invent regulations, deadlines, competencies, figures, or requirementsDistinguish confirmed information, drafts, and pending questionsVerify currency, competent authority, and official channelEnsure plain language, accessibility, and non-digital alternativeDo not present output as an administrative act or final legal advice

Prompt quality, format, and compatibility

Configure the AI depth and work mode, language, attachments, and the format or visual system to follow. All useful context you provide will be used in any mode.

Depth of developmentEssential: fast, without losing contextDeveloped: recommended for most casesProfessional: decisions, exceptions, and reviewSpecialist: maximum precision and scenarios

Final output language (priority)English (US)English (UK)English (Canada)English (Australia)Spanish (Mexico)Spanish (Colombia)Spanish (Argentina / Rio de la Plata)Neutral Latin American SpanishSpanish (Spain)Portuguese (Brazil)Portuguese (Portugal)French (France)French (Canada)GermanItalianCatalanDutchPolishTurkishArabic (Modern Standard)HindiIndonesianJapaneseKoreanChinese (Simplified)Chinese (Traditional)Bilingual: Spanish and English

AI work modeDirect and agile responseDeep and balanced analysisResearch with authorized sourcesPlan for agent or supervised automation

Presentation, format, and brandNo special visual requirementFollow existing template or documentProfessional report or documentPresentation or slidesTable, dashboard, or spreadsheetEmail or communicationVisual piece or publishable contentChecklist, process, or work space

Material you’ll provide to the AINo files or external referencesReference document or briefImage, design, or screenshotSpreadsheet or structured dataMultiple documents or sources

Attachment reference (optional)

Brand template or reference (optional)

Generate my promptClear selection

Unlock your complete prompt

Enter your email to see and copy it.

Show my prompt

I have read and agree to the Privacy Policy so Founderz can process my email and deliver the prompt.I want to receive communications from Founderz about training, resources, and news. This is optional.

Your prompt is ready

Copy prompt
Discover the program

How to create prompts and customize them: a five-step workflow

1. In “Brief or context”, describe what you need to prepare and what material you have, with no personal data or case files.
2. Set “Public task”, “Type of administration”, “Area or department”, “Recipient” and “Stage in procedure”. Fields left on “Auto / not yet defined” are left out of the prompt.
3. Choose the “Tool”. Every option is listed “in authorized environment”, because that is the starting condition for using an assistant at work.
4. Open “Advanced settings” to configure the legal framework and currency, data and privacy, language and access, review and signature, and delivery, then check the six quality controls, which are on by default.
5. In the quality block, pick the depth, final language, work mode, presentation and attached material. Click “Generate my prompt”, enter your email, then “Show my prompt” and copy the result.

The “Public task” field covers ten common jobs: drafting in plain language, structuring a report or memo, preparing terms or a case file, responding to citizens, analyzing public policy, developing a strategic plan, transparency and open data, managing a public project, training in digital and AI, and preparing institutional communication.

Example: a plain language notice from a city council

The input was the brief “We need to explain the new bulky waste collection schedule to residents: which days, how to book a pickup and what is not collected. The current text is long and confusing”, with Public task: Drafting in plain language, Type of administration: City council and Recipient: General public. Here is an excerpt of the output:

OBJECTIVE AND PROVIDED CONTEXT:
Brief or context: We need to explain the new bulky waste collection schedule to residents: which days, how to book a pickup and what is not collected. The current text is long and confusing
Public task: Drafting in plain language
Type of administration: City council
Recipient: General public
Tool: Microsoft Copilot in authorized environment

DELIVERY INSTRUCTIONS:
1. Use all useful provided information and distinguish facts, assumptions, examples, and open questions.
2. Do not invent data, sources, permissions, results, or access to documents.
3. Deliver an actionable result in the requested format, with limits and a final human verification.

The full prompt adds the six selected controls and the quality configuration (depth, language, work mode, presentation, brand and material). Compared with a generic request like “rewrite this notice”, the model knows it can’t invent deadlines or requirements, and that the text must offer a non-digital alternative for residents who don’t use online services.

Best practices for AI use and responsible use of AI in government

• Don’t paste case files or personal data into AI tools your agency hasn’t approved. The first control repeats this in every prompt.
• Ask the assistant to separate legal requirements from recommendations, and check every rule and its effective date at the official source before you cite it.
• Use a recognized reference for risk. The NIST AI Risk Management Framework is a voluntary reference many US organizations use to map and manage AI risks.
• Use “Language and access” to ask for easy reading or a screen-reader-friendly version when the text goes to the public.
• Mark in “Review and signature” who validates: the competent unit, legal or data protection. What you get is a working version, never an administrative act.
• Iterate: refine prompts by changing one setting at a time, try different prompts for the same task, and keep the versions that work as standard prompts for your team.

Use cases: prompts for public sector teams, from reports to AI chatbots

• Citizen services: clear answers to frequent questions, including scripts for AI chatbots on your website.
• Tenders: an index and first version of the terms or an RFP, with pending data clearly marked.
• Reports: structure a report, memo or executive summary from the material provided.
• Projects: a public project plan with a risk matrix and a stakeholder map.
• Transparency: an open data publication plan with quality controls.
• Training: a session with cases and a protocol to train staff in digital skills and AI.

Prompt engineering for any AI model or LLM: what makes better prompts

A good prompt gives the assistant what you’d give a colleague from another department: a role, a specific task, the context of the procedure, the material it can use, the constraints and the output format. In government work, three more pieces matter: the applicable rule and its date, data protection and human oversight. If the rule is missing, the model invents one or cites an outdated version; if the recipient is missing, the text comes out in bureaucratic language for a resident who only wants to know which day to put out a sofa. Customizing the recipient and the stage in the procedure changes the result more than adding adjectives, and it is the safest way to integrate AI into an existing workflow. The same structure works if you create AI prompts for custom GPTs or a Copilot agent.

More prompt templates and prompts to get started with artificial intelligence, like ChatGPT

If you’d rather start from written examples, our guide to AI prompts for public officials and public administration collects prompts by task. The Founderz AI prompt library has templates for other roles and tools, and these generators cover related work:

• AI prompt generator for project managers, for public project plans and status reports.
• Copilot Excel prompt generator, for tracking tables and aggregated data.
• AI prompt generator for customer service, for multichannel reply templates.

Why I built these generators

I built these generators because I kept seeing people get generic answers from AI when their prompts were vague. I’m Paul Delaney, I’ve been writing AI prompts since the GPT-2 era, and I work at Founderz, a school that teaches people to use AI at work. So I put the structure I use myself into every generator: role, task, context, material, constraints and format. You add what you know about your procedure, and the generator puts it in order so the AI doesn’t have to guess.

Frequently asked questions

Is the generator free?

Yes. You can create as many prompts as you like at no cost. The tool builds the prompt in your browser and, to show and copy it, asks for your email and your acceptance of the privacy policy. Receiving communications from Founderz is optional and checked separately.

Can I use ChatGPT in government work?

It depends on what your agency has authorized. That’s why the “Tool” field only offers options in an authorized environment: Microsoft Copilot, ChatGPT Enterprise, Claude, Google Gemini or another approved tool. Check your organization’s AI policy before you upload any document to an AI assistant.

What data should I never enter?

Personal data, case files, confidential information and any sensitive data. One control forbids it, and the “Sensitive data: don’t process” option tells the assistant to refuse to work with it.

Is the output legally valid?

No. It’s a working draft. One of the controls stops the assistant from presenting the result as an administrative act or final legal advice, and review by the competent unit is required before anything is signed or published.

Can I write in other languages?

Yes. In the final language setting you can choose US, UK, Canadian or Australian English, several Spanish variants, Portuguese, French, German, Italian and other languages.

The generator gives you the prompt; the change in your team comes when people learn to use AI with judgment and within the rules. The Founderz Certificate Program in AI and Innovation is fully online, takes 3 to 4 months and leads to a certification from Founderz and Microsoft, with modules on prompting, automation and AI agents, cybersecurity risks and responsible AI leadership.