top of page
Search

Integrating ChatGPT into Public Affairs: A Practical Guide to Smarter Stakeholder Mapping

  • marta2253
  • Jun 10
  • 17 min read


Co-Founder Advocacy Academy, Advocacy Strategy and Owner at Paul Shotton Consulting

June 3, 2025

By Paul Shotton


Disclaimer: This article was developed in collaboration with ChatGPT through an iterative conversation of over 70 back-and-forth prompts. Each section was refined using practical examples, reflection, and revision. The final product reflects my own expertise and methodology, augmented by GPT, but it does not guarantee error-free insights. If you spot anything incorrect or have suggestions for improvement, I'd be grateful to hear from you—this is a living methodology, and I welcome constructive input. Claude was the one who proof read and edited the Chat GPT supported article.


Important Notes on GPT Use and Limitations


Before diving in, it's important to understand a few things about using GPT in stakeholder mapping.

GPT is not an automation tool or web scraper. It can't access the internet, live data, or structured databases unless it's embedded in a larger toolchain. If you need real-time analysis or updates from databases, this requires scripting (e.g., Python), API access, or tools like LangChain or Zapier—not GPT alone.

GPT's memory is session-bound and limited. Even within one chat session, earlier parts of a conversation can be dropped from memory if the conversation is long. Always restate context and avoid assuming persistence unless you are using a custom GPT with memory. This article was too long and the final draft needed to be drafted in a series of 5 prompts.

Advanced GPT use includes Retrieval-Augmented Generation (RAG). For more sophisticated applications, GPT can be combined with retrieval tools to answer questions based on indexed datasets, internal databases, or document repositories. This allows GPT to simulate "knowledge" from structured sources it would otherwise not access.

Use GPT for prompting, not programming. It is a powerful assistant for thinking, writing, and reflecting—but it cannot automate your stakeholder mapping model. You still need to guide the logic, make design choices, and carry out evaluations.

Always ensure data compliance. Do not share confidential, sensitive, or personally identifiable information with GPT unless you are using a secure enterprise solution. Make sure your use of GPT aligns with internal data protection and privacy protocols.

Use structured prompt design. When working with GPT, use this model to get the best results:


  • Role: Define who GPT is (e.g., "You are a policy advisor…")

  • Task: Be clear about the action you want (e.g., "Extract all stakeholders…")

  • Format: Specify output structure (e.g., "Provide a table with 4 columns…")

  • Constraints: Include word limits or tone (e.g., "Max 200 words, keep neutral")

  • Context: Provide background info, if needed


Introduction

In recent months, I've been testing and refining how ChatGPT and similar large language models (LLMs) can be integrated into real public affairs workflows. Not as a shortcut, and not to replace expertise, but to genuinely support strategic thinking, structure complex processes, speed up practical work and deepen my analysis.

Stakeholder mapping is just one of the 11 steps in our broader 11-step methodology for public affairs management. For this article, I've taken that single step—stakeholder mapping—and broken it down into its own dedicated workflow methodology. Within each of the steps of the stakeholder mapping process, I've developed a modular workflow and a detailed reflection on where GPT can help, where it can't, and where more advanced techniques—like big data analysis—could further enhance the work.

This article presents that stakeholder mapping methodology in full detail: a breakdown of each sub-step within stakeholder mapping, a practical sequence of GPT prompts to support each task, pro tips and process notes to ensure effective and strategic use, and cautions about GPT's limitations, with examples of where big data can perform similar or more advanced functions.

This detailed walkthrough is not only a guide for using GPT—it's a way to improve your own understanding of stakeholder mapping. By walking through these structured steps, you clarify your thinking, sharpen your definitions, and improve consistency across your team.


Step 1: Select the Policy Issue or Proposal


What You're Doing

Before you begin stakeholder mapping, you need a clear sense of what you're mapping. The first step is to identify the legislative or policy proposal your organisation is engaging with. In our broader 11-step methodology, this is closely linked to issue prioritisation. For this article, we assume that prioritisation has already been completed, and we begin with a chosen proposal.

Once selected, it is helpful to describe the title and reference number of the proposal, the key articles or chapters relevant to your interests, and your organisation's position on the proposal and its parts. This forms the basis for how you will later judge stakeholder alignment.


Prompt Sequence

Prompt 1: "Help me summarise the key objectives and structure of this legislative proposal based on this document."

Prompt 2: "List the key chapters or articles of the proposal and provide a short explanation for each."

Prompt 3: "I want to assess stakeholder alignment with our position. Our position is [insert here]. Based on this, which parts of the proposal should we focus on?" Consider adding any relevant briefings, position papers other similar documents, if available.

Pro Tips

·         You can ask GPT to help you create a "sub-issue framework"—a breakdown of the proposal into meaningful chunks (e.g., by article, theme, or topic) so that you can assess alignment in detail later on.

·         Treat this step as scoping. It's about building the foundation so that your mapping is anchored in a real, high impact, high priority and well-understood issue.


Step 2: Identify Stakeholders



What You're Doing

Now that you have identified the policy proposal you're focusing on, the next step is to identify the stakeholders relevant to that proposal. This isn't just about creating a list of names—it's about building a structured dataset that serves as the foundation for all subsequent analysis.

Your goal in this step is threefold: extract names and organisations from policy and advocacy documents, categorise each stakeholder by role, interest type, or institutional affiliation, and expand the scope by reflecting on stakeholder categories you might be missing. You are creating your initial stakeholder table here—this will later be extended with additional data points like influence and alignment.

How GPT Can Help


GPT is highly effective in this step when used methodically. It can help extract stakeholder names and organisations from documents, generate structured tables with key fields, reflect on potential missing categories of stakeholders, and suggest categorisation logic based on your policy context.


The Manual Work You Still Do

While GPT can structure and extract from short or medium-length documents, you still need to gather the source material yourself, upload and summarise documents in manageable clusters, review and validate extracted data, and identify strategic gaps in the stakeholder landscape.


Workflow Breakdown


2.1 Gather Source Material

Start by collecting documents from the most relevant platforms: Legislative Observatory (OEIL), Commission's 'Have Your Say' portal, DG websites and newsrooms, EUR-Lex. You want impact assessments, SWDs, COM documents, position papers from associations, companies, NGOs, parliamentary hearing summaries or transcripts, and press releases or news articles.


2.2 Analyze and Extract Stakeholders

Use GPT to extract names, organisations, and roles from each document. Don't upload everything at once. Instead, work in small batches so that GPT can focus its token budget more effectively.

Prompt 1: "You are a policy analyst. Extract all stakeholder names and organisations mentioned in this consultation response. Present the output in a table with columns: Stakeholder First name, Last Name, Job Title, Organisation, Contact Information, Source Document."

Upload the document or paste content into the window. Then repeat for each key file and update the table.


2.3 Categorise Stakeholders

Once you've got a preliminary list, ask GPT to help you categorise them.

Prompt 2: "Now organise this list of stakeholders into categories. Possible categories include: EU Institutions, National Authorities, Trade Associations, Companies, NGOs, Academia, Media, Experts. Add a new column for Category."

Prompt 3 (optional): "Suggest additional stakeholder categories relevant to the EU policymaking environment for this proposal. Are there any types of actors we've likely missed?"

2.4 Review and Consolidate

Once you've built your stakeholder table, you can ask GPT to highlight duplicates or similar entries, suggest possible missing names based on the policy context, and identify high-level summaries.

Prompt 4: "Based on this table, summarise which stakeholder groups are most present or most vocal. Are there gaps in representation?"

Pro Tips

·         Save links and files in a central folder. If your GPT tool allows uploads, you can reference these individually.

·         If you have many documents, maintain a running Excel or Airtable sheet and paste GPT outputs into it. You can ask GPT to continuously update a table by saying "Add these new stakeholders to the previous table."

·         You can also cluster stakeholders by likely position (supportive, neutral, opposing) based on content hints—though this overlaps with alignment, which you'll do in a later step.


Big Data Considerations

This is a step where big data analysis can be applied with great sophistication—but only if you have access to the tools and support to do it. A big data approach would involve crawling structured datasets, automatically extracting and tagging stakeholder names and metadata, and building a stakeholder graph based on interaction patterns or sentiment analysis.

However, this is significantly more complex and resource-intensive. You'd need a structured or semi-structured dataset, support from a data scientist, a tagging algorithm and logic for classification, and validation mechanisms for false positives or weak matches. GPT cannot do this automatically—but it can help you structure and interpret the results once you have them.


Step 3: Define Indicators – Influence and Alignment


What You're Doing

Once you've identified and categorised your stakeholders, the next critical task is to define how you'll evaluate them. In our methodology, we focus on two primary indicators: influence (how much capacity or leverage the stakeholder has to shape the outcome of the policy process) and alignment (how closely the stakeholder's views or interests align with your organisation's position or goals).

This step is about defining what these indicators mean in practice. Without a clear definition, rankings will be inconsistent or subjective, and you'll lose the ability to compare and cluster effectively later on. This is not yet the moment to carry out the evaluation. That comes next. Right now, you're defining the framework that will guide your judgment and potentially that of your team.


Sub-Steps in Step 3


3.1 Select Your Indicators

We use Influence and Alignment as our standard indicators, but depending on your context, you may also consider visibility, access to decision-makers, expertise or technical credibility, financial interest in the outcome, or historical engagement on the issue. These can be useful additions—but they complicate the analysis. We typically recommend sticking with two for clarity unless there's a strong reason to expand.

Prompt 1: "What are the most useful indicators to evaluate stakeholders in an EU legislative stakeholder mapping exercise? Please focus on indicators that support decision-making around strategic engagement."

Follow-up if needed: "Can you compare the pros and cons of using 2 indicators (influence and alignment) versus using 4 or 5?"

3.2 Define the Indicators

Here you want GPT to help clarify what each indicator means—specifically in your policy context. Influence, for example, could mean formal power, informal access to key decision-makers, technical authority, or agenda-setting capacity. Alignment might relate to agreement with your organisation's mission or vision, support for your position on a specific article, or shared values or long-term goals.

Prompt 2: "Help me define 'Influence' as an indicator for stakeholder mapping in the EU policy process. Include examples of high, medium, and low influence."

Prompt 3: "Help me define 'Alignment' as an indicator for a stakeholder mapping exercise. How can we distinguish between alignment on the proposal as a whole, and alignment on specific articles?"

3.3 Define Your Scale

We usually recommend a 1 to 5 scale, where 1 = very low and 5 = very high. But this needs to be more than just numbers. You should define what each number means for each indicator.

Prompt 4: "Using a 1 to 5 scale, define levels of influence for stakeholders in the EU policy process. Be specific about what qualifies as a 1 versus a 5."

Prompt 5: "Now do the same for alignment—define a 1 to 5 scale to evaluate how aligned a stakeholder is with our organisation's position."

3.4 Design the Table Columns

Now that you have your indicators and your scale, you can design the table you'll use for evaluation. At minimum, your table should include: Stakeholder Name, Organisation, Category, Influence (1–5), Alignment (1–5), Notes or Evidence, and Sub-Issue (if relevant). If you're working in Excel or Airtable, GPT can also help with formulas or dropdown formatting.

Prompt 6: "Design a stakeholder mapping table in Excel with columns for stakeholder name, organisation, influence score, alignment score, and notes. Use dropdowns for the scores and predefined formatting."

3.5 Define Means of Verification

A vital but often overlooked part of stakeholder mapping is agreeing how you'll evaluate each score. This step is about defining the types of evidence you'll accept to justify a score. For example, a 5 on influence might be justified by a formal role or repeated authorship of amendments, while a 4 on alignment might be based on a published position paper or speech.

Prompt 7: "Suggest possible means of verification for scoring stakeholder influence in a legislative context."

Prompt 8: "Now list means of verification for alignment—what kind of data or statements can we use to justify a 1–5 alignment score?"

This becomes a reference sheet you can use for internal consistency—especially if multiple people are scoring.


Pro Tips

·         Avoid rushing this step. Sloppy definitions now lead to inconsistent analysis later.

·         GPT is very good at surfacing distinctions—ask it to challenge your definitions, compare indicators, or simulate borderline cases.

·         Use GPT to help craft the guidance that others will follow if multiple colleagues are filling out the table. It can generate practical scoring instructions.

·         Don't forget nuance: A policymaker may be highly influential on some issues, but not others. Use "sub-issues" if needed.


Big Data Comparison

In theory, big data systems could automate parts of this step by crawling public statements to infer alignment and mapping institutional relationships to estimate influence. But in reality, designing these systems takes time, technical skill, and resources. They require robust training data and consistent logic, and human judgment is still essential—especially for interpreting nuance, sentiment, or context. That's why this workflow leans on AI-supported thinking, not AI-driven automation.


Step 4: Evaluate Stakeholders


What You're Doing

This is the step where you apply the framework you built in Step 3. You now need to assess each stakeholder identified in Step 2 against the indicators and scoring scales defined in Step 3—typically influence and alignment, both ranked on a 1–5 scale.

This is one of the most labour-intensive stages in the stakeholder mapping workflow. It requires judgment, familiarity with the legislative process, contextual reading of documents, and—critically—consistency. You must ensure that all stakeholders are evaluated using the same logic, definitions, and evidence standards. GPT can help, but it cannot fully automate this step. In the end, you (or your team) must make the final judgment calls.


Sub-Steps in Step 4


4.1 Set Up the Evaluation Table

You should now have a structured table with the following (at minimum): Stakeholder Name, Organisation, Category, Influence Score (1–5), Alignment Score (1–5), Notes/Evidence, and Sub-Issue (optional). If not already done in Step 3, you can ask GPT to help set up the table structure or improve its formatting in Excel or Airtable.


4.2 Conduct the Evaluation

This is where the real work begins. For each stakeholder, you will read relevant materials, score influence using the predefined scale and supported by evidence, score alignment relative to your position on the overall proposal or specific sub-issues, and document justification in the notes or evidence column.

GPT can assist with analysis by examining text for alignment with your position.

Prompt 1: "Analyse this position paper and extract the stakeholder's stance on [Article X of Proposal Y]."

Prompt 2: "Based on our position that [insert position], how aligned is this stakeholder on Article X? Rate alignment from 1 to 5 and explain."

Repeat this process one stakeholder at a time or in small clusters of documents, rather than trying to load all documents at once.


4.3 Subjective vs Objective Sources

Whereas industry groups and NGOs often publish their views, institutional stakeholders like policymakers or officials may not. In these cases, your scoring may rely on more subjective sources: meeting notes or readouts, speeches, interventions, or interviews, voting records or amendment proposals, and informal insight from team members. GPT can help you structure your judgment, but not replace it.

Prompt 3: "Here is a transcript of a parliamentary discussion. Highlight whether this MEP expresses alignment with our position on [topic]."

Prompt 4: "Based on this input, suggest a preliminary influence and alignment score from 1 to 5 with a short rationale."

4.4 Ensure Consistency

Once you've gone through all stakeholders, you need to check for inconsistencies in your scoring. GPT can help you validate and refine your evaluations by asking probing questions.

Prompt 5: "Based on this stakeholder table, ask me question to identify any inconsistencies in how influence or alignment has been scored?"

You can also cluster stakeholders by category and look for score distributions that appear off or surprising.


4.5 Sub-Issue Level Evaluation

One of the strengths of this methodology is that it allows you to break down alignment by sub-issue, rather than treating the proposal as a whole. You can define sub-issues as specific articles in the legislative text, technical vs political components, or economic, environmental, or ethical dimensions. This is especially valuable when stakeholders have mixed positions—supporting some parts of a proposal but opposing others.

In your table, you can then include a column for sub-issue, score stakeholders multiple times (once per sub-issue), and use GPT to help generate the sub-issue breakdowns and assessments.

Prompt 7: "Here is our list of top sub-issues in the proposal. Based on this stakeholder's input, rate alignment on each sub-issue."

Pro Tips

·         If you're uploading documents to GPT (in Pro or enterprise versions), do so gradually. Overloading the model with too many documents at once will reduce accuracy and clarity. Work iteratively.

·         Define your evidence rules clearly. For each score, you should be able to justify it with a quote, reference, or clear observation.

·         Don't overcomplicate. Too many indicators or too fine a scoring scale can lead to confusion and loss of comparability.

·         Cluster scores for comparability. Use GPT to sort and filter the table by stakeholder type or sub-issue for better validation.

·         Maintain judgment. This process supports decision-making, but you still have to choose how to interpret grey areas.

·         Use GPT to simulate a dialogue. You can ask GPT to play devil's advocate or challenge your scoring with counterarguments.

·         Ask GPT to run a reflection session: "Act as my strategic advisor and help me reflect on these scores. Where do you see gaps, inconsistencies, or potential errors?"

Big Data Comparison

A full big data approach could theoretically scrape and analyse social media, press releases, or legislative databases, use Natural Language Processing (NLP) to infer sentiment or alignment, and build network graphs to estimate influence from citations or institutional roles. But this requires a clearly defined dataset, a programmed model with good training data, and technical implementation (often with Python, R, or commercial platforms). Unless you're working with a dedicated data science team, the hybrid approach we present here—manual evaluation, augmented by GPT—offers a better balance of control, feasibility, and depth.


Step 5a: Analyse Results and Cluster Stakeholders

What You're Doing

Once you've evaluated each stakeholder based on influence and alignment, it's time to analyse the results. This is the step where the data you've collected becomes actionable. You move from individual scores to patterns—using those patterns to shape strategic decisions.

Your goal is to visually map stakeholders (e.g., in a quadrant), group them into engagement clusters, and identify gaps, priorities, and tactical insights. This is also where GPT is especially useful as a strategic reflection partner—not just for generating visualisations, but for interpreting patterns and prompting smarter decisions.

Sub-Steps in Step 5a

5a.1 Format and Clean the Table

Before analysing, make sure your data is clean: no missing scores, all stakeholders have categories and sub-issues (if used), and consistent scale use across columns. If using Excel or Airtable, GPT can help check formatting or generate formulas.

Prompt 1: "Review this stakeholder mapping table. Identify any missing scores or inconsistent entries."

Prompt 2 (for Airtable/Excel users): "Help me create a formula / / use conditional formatting to automatically highlight stakeholders with Influence ≥ 4 and Alignment ≤ 2."

5a.2 Visualise the Map

Next, you want to plot your stakeholders on a 2x2 matrix, using influence and alignment as axes. GPT can help explain how to build this in Excel, simulate example quadrant layouts, and generate descriptions of each cluster.

Prompt 3: "How do I create a quadrant graph in Excel using stakeholder Influence and Alignment scores from 1 to 5?"

Prompt 4: "I've mapped stakeholders into four quadrants: High Influence / High Alignment, High Influence / Low Alignment, Low Influence / High Alignment, Low Influence / Low Alignment. Help me label and describe each group in strategic terms."

5a.3 Cluster Stakeholders

Now assign each stakeholder to a strategic engagement cluster based on their quadrant. You can also use this moment to reflect on engagement readiness, urgency, or the type of relationship needed.

Prompt 5: "Based on this quadrant analysis, cluster stakeholders into categories like 'Core Allies,' 'Priority Opposition,' 'Low Priority Supporters,' and 'Monitor Only'. Suggest criteria for each."

5a.4 Interpret the Patterns

This is where you shift from descriptive to strategic. GPT can help you answer where you have the most support, which stakeholders pose the greatest risk, whether there are influential actors you're ignoring, and where you should concentrate effort.

Prompt 6: "Act as my strategic advisor. Based on the stakeholder quadrant mapping, what are the three most important insights I should take into account for designing an engagement strategy?"

Pro Tips

·         These clusters can be adapted based on campaign needs (e.g., prioritisation, risk, or advocacy capacity).


Step 5b: Design Engagement Strategies per Cluster

What You're Doing

With stakeholders grouped into clusters, you now begin to develop tailored engagement strategies. Rather than treating each stakeholder individually, you define approaches based on group characteristics. This step is a bridge between stakeholder mapping and campaign planning. It connects analysis to action.

Sub-Steps in Step 5b

5b.1 Define Strategy Per Cluster

For each cluster, define engagement objective, recommended tactics, key messages, and frequency and format of interaction.

Prompt 7: "Suggest engagement strategies for the following stakeholder group: High Influence / Low Alignment. Include objectives, preferred channels, and suggested message tone."

Repeat for each quadrant/cluster.

5b.2 Identify Priority Stakeholders

Within each cluster, identify top-priority stakeholders based on decision-making proximity, timeliness, reputational weight, and potential to shift alignment.

Prompt 8: "Based on this cluster of High Influence / Neutral Alignment actors, which 3 stakeholders would you prioritise for immediate engagement and why?"

5b.3 Map to Wider Campaign Tactics

Engagement does not happen in isolation. This is the moment to link stakeholder strategies to message development, coalition-building, timeline and sequencing, and legislative process tracking.

Prompt 9: "Based on this stakeholder cluster and our position on Article 6, suggest a sequence of advocacy tactics for the next 8 weeks."

Pro Tips

·         Ask GPT to help you differentiate between institutional actors, corporate actors, and NGOs, even within a given quadrant.

·         Use your 11-step methodology to reconnect. Stakeholder mapping (Step 5) now links with Strategic Planning (Step 6), Message Development (Step 7), and Engagement Planning (Step 8).

·         Stakeholder clustering is where your analysis becomes strategy. Don't rush this—it's the payoff for everything you've done.

·         Use GPT as a sparring partner. Ask it to test assumptions, simulate challenges, or compare alternative approaches.

·         Revisit the data periodically. This is not a one-off map. As stakeholders shift positions or political dynamics change, so must your map.

·         Keep it visual. Decision-makers and teams understand diagrams faster than tables. GPT can help design and describe your visuals—even if you're building them elsewhere.

Conclusion: From Mapping to Strategy

Stakeholder mapping has long been considered an essential part of public affairs—but in many organisations, it's done informally, inconsistently, or without a shared framework. This article proposes a structured, methodical, and AI-supported way of doing it better.

By integrating ChatGPT into each step of the stakeholder mapping process, you gain structure, even when you're still exploring; speed, especially for repetitive analysis and formatting; and sparring, with a tool that helps you reflect and iterate.

But this isn't just about speed or novelty. GPT is most powerful when it reinforces your thinking, not replaces it. It works best when you break down your tasks into modular prompts, define your assumptions clearly, and use the model to simulate, critique, and help structure—not to make strategic decisions for you.

This hybrid approach—grounded in your own experience and strengthened with smart AI use—is far more powerful than relying on instinct or automation alone.

A Note on Big Data

For those with access to more sophisticated tools and technical support, big data techniques can automate parts of this workflow. You can build dynamic stakeholder datasets, automate position extraction from documents and speeches, run sentiment or influence analysis using NLP, and integrate with dashboards that track policy activity.

But this comes at a cost—both financially and technically. It requires pre-structured datasets, algorithm development, and ongoing validation. For most practitioners, the model presented here—manual expertise supported by GPT—is more realistic and adaptable.


Final Invitation

This article was written through 60+ prompts and reflections with ChatGPT, shaped by my own methodology, and aimed at supporting public affairs professionals as they begin to explore AI use seriously and practically. If you've found gaps, contradictions, or ideas worth improving—I would genuinely love to hear them. This is a living methodology, not a final product.

And if you're working on building your own public affairs AI toolkit—or just trying to figure out where to start—reach out. We're all learning this together.

 
 
 

header.all-comments


Contact Us
+32 (0) 470 95 23 29
hello@advocacystrategy.com
Brussels - The Hague - Madrid
© 2025 Advocacy Strategy. All Rights Reserved.
bottom of page