Tutorial: How to handle LLMs in your research

For the Department of Development & Education of Youth in Diverse Societies

Course coordinators: Robert Bagheri, Pablo Mosteiro, Laurence Frank, Ozgur Togay (Department of Methodology, Statistics and Data Science)

Duration: 1 hour

This tutorial was drafted with assistance from GPT-5.6-Sol. All examples below are AI generated.

Learning objectives

By the end of this tutorial, you will be able to:

  1. Use UU AI-Chat and choose an appropriate way to organise your work using chats, prompts, personas, projects, and groups.
  2. Use AI assistants for common research tasks, including:
    • extracting relevant information from text,
    • coding, classifying, or annotating text,
    • summarising or comparing research material.
  3. Turn useful one-off prompts into more reusable research workflows.
  4. Critically evaluate AI-generated output and recognise important limitations.

Part 1: Get started with UU AI-Chat

For this tutorial, we will use UU AI-Chat, Utrecht University’s generative AI environment. Sign in with your UU account and start a new chat.

UU AI-Chat provides access to several language models and supports files such as PDFs, Word documents, spreadsheets, presentations, and text files. It also includes tools for web search, data analysis, code execution, image generation, and other tasks.

Choose the right workspace

You do not need to use every feature in this tutorial. The main options in the left sidebar serve different purposes:

Feature Useful when…
Chat You want to ask a question, analyse a file, or carry out a one-off task.
Personas You want a reusable AI assistant with a consistent role, instructions, or source material for a recurring task.
Projects You want to keep related chats, files, instructions, and other material together for an ongoing research task.
Groups You want to share reusable resources such as prompts or personas and collaborate with other UU AI-Chat users.
Prompts You want to save an instruction that you expect to use again instead of rewriting it each time.


The chat screen also contains ready-made prompts for common tasks. For example, under Research you will find prompts for developing hypotheses or engaging with research literature. Treat these as starting points: inspect what the prompt asks the model to do, adapt it to your research question, and evaluate the result critically.

Choosing a model

For the exercises below, the default model is usually sufficient. You can browse or switch models using the model button in the bottom-right corner of the chat.

Different models may vary in capability, speed, and computing requirements. UU AI-Chat indicates relative computing or energy use in the model selector. Use a more demanding model when the task benefits from it, rather than automatically choosing the largest option.

Working with files

When an exercise uses research material, you can either paste the relevant text into the chat or upload the source file. When working from an uploaded document, tell the model clearly which document it should use and what information it should base its answer on.

Throughout the tutorial, remember that the model’s output can be incomplete or incorrect even when it is based on an uploaded source.

Part 2: Example applications in research

Example 1: Extract information from an observation

Scenario: You have a short classroom observation about multilingual pupils’ participation during a collaborative learning activity.

Copy this prompt to your AI chat:

Extract the key information from the classroom observation below.

Identify:
1. Observable instances of participation and language use.
2. Teacher or peer actions that may be relevant to participation.
3. Ambiguities or missing context.

Support your points with details from the observation. Separate observation from interpretation, and do not infer traits or motives that are not directly supported.

Observation:
"During a small-group science task, the teacher asked pupils to discuss why one object floated and another sank. Lina first discussed her idea quietly in Arabic with Samira. Samira then summarized part of their discussion in Dutch. The teacher told the group that they could use any language while working out the problem, but that the group should present its final explanation in Dutch. Later, Lina raised her hand when the teacher asked groups to report back, but another pupil from the group answered after the teacher called on the group. At the end of the lesson, Lina completed the written explanation independently."

Task: Use the BRAVE(R) framework to analyze the prompt, then use the FACTS framework to analyze the response. Discuss your evaluation with a classmate.

Note: LLMs can also help you build and refine prompts.

Reflection

  1. Did the model distinguish observation from interpretation?
  2. Did it support its claims with details from the excerpt?
  3. What would you need to know before drawing broader conclusions about participation or inclusion?

If this is a task you expect to repeat, consider saving the instructions as a Prompt. A saved prompt is useful when the instructions stay mostly the same but the input changes.

Example 2: Analyze two focus-group texts

Scenario: You want to explore caregivers’ needs, challenges, strategies, and cultural factors related to toilet training autistic children using two synthetic focus-group excerpts.

Copy this prompt to your AI chat:

Analyze the two synthetic focus-group excerpts below in relation to this research question:

What are the needs, challenges, strategies, and cultural factors that shape toilet-training practices for autistic children in these two focus groups?

Identify evidence for each of the four themes. Keep the two focus groups distinguishable, support interpretations with speaker IDs or short excerpts, and note ambiguity rather than filling in missing information.

These texts are fictional training materials. Do not treat them as representative of caregivers, autistic children, or toilet-training practices in Malaysia or Morocco.

Focus group 1 — Malaysia:

Moderator: What has been most difficult about toilet training for your child?

M1: "My son does not always tell us that he needs the toilet. Sometimes we notice from his behaviour, but by then it can be too late. I would like clearer guidance on how to recognise his signals without pushing him too much."

M2: "For my daughter, the bathroom itself can be difficult. She dislikes the sound of the flush and gets upset when the floor is wet. At home we let her leave before we flush, and we use the same picture sequence every time. Public toilets are harder because we cannot control the noise."

M3: "The family also has different ideas about when a child should already be toilet trained. Some relatives compare my son with his cousins and ask why he still needs help. That creates pressure. His grandmother wanted us to put him on the toilet at fixed times very often, but he became resistant. We eventually agreed on a routine that gives him more time."

Moderator: Are there differences between places where your child uses the toilet?

M4: "Yes. At home we have a sitting toilet, but at my parents' house there is a squat toilet. My daughter was confused by the change. We used pictures and a small foot support to practise the different steps before visiting them."

M1: "Consistency between home and school helps us most. When the teacher started using the same words and picture card that we use at home, my son understood the routine more quickly."

M3: "I would also like information that everyone in the family can use. Some guidance we received was in English and was too detailed for my mother. A short version with pictures and simple Malay would make it easier for us to follow the same approach."

Moderator: What strategies have been helpful?

M2: "A predictable routine, pictures, and not making an accident into a big event. If she is distressed, we stop and try later. The biggest improvement came when home and school agreed on the same steps."


Focus group 2 — Morocco:

Moderator: Chno huwa akbar taḥaddi f toilet training? [What is the biggest challenge in toilet training?]

R1: "Bnti katjbed yeddi ila bghat chi haja, walakin ma kanʿrafch dima wash katqsed toilette." [My daughter takes my hand when she wants something, but I do not always know whether she means the toilet.] "I need practical ways to help her communicate the need earlier without expecting spoken words."

R2: "Mon fils est très sensible au bruit dans les toilettes." [My son is very sensitive to noise in bathrooms.] "The echo and the hand dryer in some places frighten him. We first practised when the bathroom was quiet and used a simple visual sequence. Now he manages at home, but unfamiliar toilets are still difficult."

R3: "Several relatives are involved because we live close to my parents and they help with care. That can be useful, but everyone gives different advice. Waḥed kaygol khassek tkon ṣarem, waḥed kaygol khallih b raḥto." [One person says you should be strict, another says leave him at his own pace.] "We found it works better when everybody follows the same routine and accidents do not become a conflict."

Moderator: Are there household or family practices that affect the routine?

R4: "In our home we normally use water for cleaning after the toilet. My son disliked the sensation at first, so we introduced it gradually and let him control when the water touched his skin. At school the routine is different, so moving between the two settings still takes preparation."

R1: "Some information we received was in French. Moi, je peux le lire, mais ma mère préfère l'arabe, et à la maison on parle surtout darija." [I can read it, but my mother prefers Arabic, and at home we mostly speak Darija.] "Pictures and short instructions that all caregivers understand would make it easier to stay consistent."

Moderator: What has helped most?

R2: "La coordination avec l'enseignante." [Coordination with the teacher.] "We agreed on the same cue and roughly the same sequence of steps. We also stopped treating speed as the main goal. When he became anxious, we went back to an easier step and tried again later."

R3: "For us it was important that relatives understood that the same approach does not work for every child. Once we agreed on a routine and stopped comparing him with other children, there was less tension around toileting."

Task: Before sending the prompt, adapt it using the BRAVE(R) framework. For example, decide how you want the four themes organized and what kind of evidence the response should provide.

If useful, create a Project for the exercise and a simple Persona with reusable qualitative-analysis instructions, such as staying close to the supplied texts and distinguishing evidence from interpretation.

Then run the prompt and use the FACTS framework to evaluate the response.

Data protection: These excerpts are synthetic. Before using real focus-group, interview, or other research data that may contain personal or sensitive information with GenAI, check the applicable requirements with the relevant privacy or data-protection contact. Do not assume that using UU AI-Chat alone is sufficient clearance.

Note: You can include a codebook in the prompt. Clear definitions, decision rules, and output formats can make repeated classifications more consistent.

Reflection

  1. Did the model distinguish the four themes and support them with evidence from the texts?
  2. Did it preserve differences between the two focus groups rather than generalising from them?
  3. Which instructions would be useful to save in a reusable Persona, and which are specific to this research question?

For a recurring task with stable instructions to the AI assistant, a Persona may be more useful than repeatedly pasting the full setup into a new chat.

Example 3: Assess the relevance of research

Scenario: You need to quickly assess whether a paper is relevant to research on children’s development, learning, participation, or educational contexts.

Choose an abstract, article excerpt, or paper that you would like to work with. You can either paste the relevant text into the prompt or upload the document directly.

If the prompt contains [INSERT TEXT HERE]:

  • replace it with your chosen text; or
  • if you uploaded a document, replace it with an instruction such as Use the uploaded document as the source.
Summarize the research abstract below for a researcher deciding whether the study is relevant to their work on child or youth development and education.

Provide:
1. The research question, population, and setting.
2. The study design and main concepts or outcomes.
3. The main findings.
4. Any stated limitations or boundary conditions.
5. What would require reading the full paper.

If relevant, note how diversity, inequality, social context, or participation is addressed. Use only information from the abstract and mark uncertainty clearly.

Abstract:
[INSERT ABSTRACT TEXT HERE]

Task: Use the FACTS framework to analyze the response. Discuss your evaluation with a classmate.

Reflection

  1. Did the model stay within what the abstract actually supports?
  2. Did it avoid overstating causality or generalisability?
  3. Would the summary help you decide whether to read the full paper?

Compare your result with one of the ready-made Research prompts in UU AI-Chat. What assumptions does the built-in prompt make? What would you add or change for your own research question?

If you are working on different parts of the same research project, consider creating a Project. Chats within a Project can share context, so you can use separate chats for the literature review, methods, analysis, or other parts while keeping them connected.

Part 3: From one-off chats to reusable workflows

A useful prompt is not always the final form of an AI-assisted workflow. After completing the examples above, choose one prompt or task that you found useful and decide how you would reuse it:

  • Save it as a Prompt if you mainly want to reuse the same instruction.
  • Create a Persona if the AI should consistently follow a particular role, method, codebook, or set of instructions.
  • Create a Project if the work involves multiple related chats, files, or recurring context.
  • Use a Group if prompts or personas should be shared with collaborators.

Task: Pick one example from this tutorial and explain which option you would use for a real research workflow and why.

The purpose is not to use the most features. Use the simplest setup that keeps the task clear, reproducible, and easy to review.

Part 4: Understanding model limitations

Exercise 1: Verify generated literature

Task: Ask the model:

Find three peer-reviewed empirical studies published in 2026 about digital technologies and social participation, learning, or development among children or adolescents. For each study, provide the title, authors, journal, DOI, study population, and one-sentence main finding.

Verify the references and findings using reliable external sources such as journal websites, Crossref, Google Scholar, or another academic database.

Reflection

  1. Do the publications and bibliographic details exist as given?
  2. Do the reported findings match the original sources?
  3. What would make this literature-search workflow more reliable?

Exercise 2: Detect assumptions in a research question

Task: Compare responses to these two prompts:

Prompt A:

Explain why multilingual children participate less in classroom discussions than monolingual children.

Prompt B:

Discuss factors that may be associated with variation in classroom participation among multilingual and monolingual children. Do not assume that a group difference exists. Distinguish empirical evidence from possible explanations.

Compare how the framing affects what the model treats as established.

Reflection

  1. Did Prompt A encourage the model to accept an unsupported assumption?
  2. Did either response overgeneralise or rely on stereotypes?
  3. How could you phrase your own research questions to reduce leading assumptions?

Wrap-up: Your personal AI-use checklist

Create your own checklist:

Next steps

For tasks that need to be applied systematically to large amounts of data, API-based workflows may be more appropriate. The Department of Methodology, Statistics and Data Science provides consultation on AI use in research: Consultation Data Science / AI

More information about the university platform is available on the UU AI platforms page.

Quick reference: BRAVE(R) and FACTS

Use BRAVE(R) when formulating or improving a prompt. Use FACTS when evaluating an AI-generated response.

BRAVE(R)

B: Boundaries Set limits on the format, length, or any other constraints.
R: Role Identify the role or perspective you want the AI tool to take.
A: Audience Specify who the output is intended for to determine the appropriate tone and style.
V: Variables Highlight key details, variables, or points that should be included in the response.
E: Expectations Clearly state what you expect the AI to accomplish, including the intended outcome and purpose.
R: Refine Provide feedback to improve the output and guide future interactions.


FACTS

F: Focus Focus on the AI’s response by breaking it down into its main points and arguments.
A: Authenticate Authenticate the AI’s output by verifying it against current research and empirical data.
C: Critique Critique the response’s accuracy, relevance, and depth in the context of the topic.
T: Think Think about the response from different perspectives and consider its potential impact.
S: Scrutinise Scrutinise the AI’s conclusions by critically questioning them and exploring alternative hypotheses or interpretations.

Additional resources

Example from research: LLM-assisted data annotation

View the research prompt

LLMs are increasingly used for data annotation because they can make large-scale coding faster and cheaper. Performance still depends on the task, but recent work has found that LLMs can match or outperform human annotators on some social-science text classification tasks (Törnberg, 2024).1

The example below comes from Togay et al. (2026), which used LLMs to identify the political target and stance in Reddit comments across 138 predefined political targets plus an open-ended category. The best-performing models were comparable to highly trained human annotators.2

This is deliberately much longer than the tutorial prompts above. Research annotation prompts can function like codebooks: they define labels, edge cases, output formats, and examples so that the same rules can be applied consistently across many observations. This prompt was also designed for an earlier generation of models; repeated emphasis such as **...** was used to reinforce important instructions and is generally not necessary with newer models.

You will be provided with a Reddit comment and the title of the submission under which it was posted.

Your task is to identify the **target** and the **stance** expressed toward it in the Comment, using the predefined list of political targets at the end.

Only classify stances related to a target from the list. If the Comment refers to a similar target with different wording, select the exact matching entry from the list.  
**Do not rename, paraphrase, or invent new target names.**  
- If the post refers to a target not on the list, label it **"Other"**.  
- If no political target is mentioned, label it **"None"**.  

Use the Submission **only** to resolve ambiguity (e.g., resolving pronouns or vague references). **Do not infer stance from the Submission.**

---

## Classification Steps

1. **Identify the Political Target**  
   - Choose a target from the predefined list at the end.  
   - If the post refers to a target not on the list, label it **"Other"**.  
   - If no target is mentioned, label it **"None"** and skip stance classification.

2. **Determine the Stance** (only if a target is identified)  
   - **Positive**: clear support or praise  
   - **Neutral**: mention without a clear stance  
   - **Negative**: criticism or disapproval
   - **None**: no stance can be identified

3. **Return the classification**

---

## Output Format

Your response must be a JSON object with the following fields:
- "target" (string): The identified political target, from the list of Predefined Targets.
- "stance" (string): The determined stance. Must be one of: "Positive", "Neutral", "Negative", or "None".
- "confidence" (float): A confidence score between 0.0 and 1.0 (inclusive) representing how certain the model is about its classification.

Example JSON output if a target and stance are identified:
```json
(
  "target": "Donald Trump",
  "stance": "Positive",
  "confidence": 0.95
)

## Examples

# Example 1
Input:
    Submission Title: New York Primary Results Megathread   
    Comment: Didn't Sanders vote for that crime bill too? Has he evolved on that position or does he still support it?  
Output:
```json
(
  "target": "Bernie Sanders",
  "stance": "Neutral",
  "confidence": 0.8
)

# Example 2
Input:
    Submission Title: Flat-tax in the U.S. - a good idea?   
    Comment: I'd say it's a good thing, but the reason for existing tax breaks is to encourage people to live in a way that is good for society; get educated, own property, have kids, etc.    
Output:
```json
(
  "target": "Tax cuts (general)",
  "stance": "Positive",
  "confidence": 0.90
)

# Example 3
Input:
    Submission Title: Is drug legalization/decriminalization sound policy? 
    Comment: "Treating drug abuse as a medical problem instead of a criminal problem has been successful in European countries and has been a conclusion by various studies.
     
     (let's make the AutoMod happy: )
     
     https://www.drugabuse.gov/about-nida/noras-blog/2014/03/unodc-recommends-treating-addiction-health-not-legal-issue
     
     https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2681083/" 
Output:
```json
(
  "target": "War on Drugs",
  "stance": "Negative",
  "confidence": 0.90
)

# Example 4
Input:
    Submission Title: What are the benefits and costs of the federal hiring freeze? 
    Comment: I will note you didn't call out the dude who cited a government agency to show government agency hiring shouldn't be frozen.   
Output:
(
  "target": "None",
  "stance": "None",
  "confidence": 1.0
)

---

## Predefined Targets

{predefined_targets}
Target; Explanation
Republican Party; Republican Party as an organization
Republican politicians; Republican politicians as a group, including representatives, senators and governors
Republican politician (generic); a Republican politician not on the list. 
Republicans; Republican party supporters. A comment about both the Republican politicians and Republican party supporters would fall here. If it is only about Republican politicians, use the Republican politicians label.
Conservatives; Conservative people and/or ideology
Right-wingers; right-wing people and/or ideology
Alt-right; alt-right people and/or ideology
Independents; independent people (often used for voters without a party affiliation)
Democratic Party; Democratic Party as an organization
Democratic politicians; Democratic politicians as a group, including representatives, senators and governors
Democratic politician (generic); a Democratic politician not on the list
Democrats; Democratic party supporters. A comment about both the Democratic politicians and Democratic party supporters would fall here. If it is only about Democratic politicians, use the Democratic politicians label.
Centrists; Centrist people and/or ideology
Liberals; Liberal people and/or ideology
Leftists; Leftist people and/or ideology
Greens; Green people and/or ideology
Socialists; Socialist people and/or ideology
Antifa; Antifa political group and its supporters
Abortion; Planned Parenthood also almost always fall under here, unless they specifically focus on Planned Parenthood as a public healthcare organization.
Gun Control;
LGBTQ rights;
Vaccination;
Belief in Institutional Racism;
Belief in Climate Change;
Immigration;
Affirmative Action for Racial Equality;
Affirmative Action for Gender Equality;
Muslims;
Jews;
Catholics;
Protestants;
Christians;
Atheists;
Whites;
Blacks;
Hispanics;
Asians;
Indians;
Government spending; Fiscal policies that favor an increased government budget  Use the specific label if it refers to a higher budget for education, healthcare, military, security, or any other item on the list.
Tax cuts (general); Tax cuts that apply to everyone
Tax cuts (for rich); Tax cuts that apply to the rich
Tax cuts (for low-middle classes); Tax cuts that apply to the low and/or middle class
Higher corporate tax; A comment calling for lower corporate tax rate would be labelled as Anti (so -1) Higher Corporate Tax,
Higher taxes for the rich;
Minimum wage/Higher minimum wage; establishing a minimum wage, or increasing it
Social spending; social spending, except healthcare, education, Coronavirus stimulus checks and security  Examples: unemployment or childcare benefits, public funding of parks etc.
Universal basic income; Universal Basic Income (UBI) in which people regularly receive a minimum income without any checks on conditions.
Public healthcare; public healthcare, including Affordable Care Act (Obamacare), Medicare, Medicaid, healthcare reforms that would require more government funding
Private healthcare; private healthcare. Examples include calls for more privatization, removal of existing social security programs (ACA/Medicaid etc.)
Public education; public education. Examples include calls for more government funding for schools
Private education; private education. Examples include calls for more privatization
Student loan forgiveness; whether student loans and debt should be forgiven.
Government bailouts; government bailout of private corporations, a common example would be bank bailouts during the 2008 financial crisis.
Military spending; the US military spending/budget
Police spending; the US police/law enforcement spending/budget. This is often referred within the state level, but covers both state and federal budgets.
International Trade; International Trade. Examples include Trans-Pacific Partnership, US trade with China, US trade with Canada and Mexico (NAFTA), World Trade Organization etc.
Protectionism; protectionist international trade policies, such as Tariffs and trade barriers.
Nuclear Energy;
Renewable/Green Energy;
Fossil fuels;
Trust in the US government (general); the Trust the US government as an organization that do not mention a party/president
Trust in the Biden administration;
Trust in the Trump administration;
Trust in the Obama administration;
Trust in the G.W. Bush administration;
Trust in the US Military;
Trust in the US Law Enforcement;
Trust in the US electoral system; the US electoral system, including electoral college, two-party system, first past the post voting, electoral districts, campaign donations and funding (political action comittes, SuperPACs)
Restrictions on Voting Access; the regulations that make voting less accessible. Examples include registration or ID checks. Counter examples, i.e. cases that make voting easier, would include mail-in ballots, ways to reduce voting queues, or helping those waiting in the voting queue.
Trust in the US judicial system/courts; the US judicial system and courts in general. Examples include fair sentencing, costly court cases where corporations can outspend individuals
Trust in the US Supreme Court; Specifically refering to the US Supreme Court
Coronavirus stimulus checks;
Discrimination based on Political Beliefs or Activities;
Capital Punishment;
Diversity, equity, and inclusion initiatives; DEI initiatives that seek to promote the fair treatment and full participation of all people, particularly groups who have historically been underrepresented or subject to discrimination. Criticized by Conservatives on the basis that it creates preferential treatment of minorities, forgoes merit as the basis of quality, and restricts freedom of speech.
Critical Race Theory; CRT, which focuses on the relationships between social conceptions of race and ethnicity, social and political laws, and mass media. CRT also considers racism to be systemic in various laws and rules, not based only on individuals' prejudices. It is criticisized by (often) Conservatives for being anti-American and anti-White, for undermining American history and traditions. Mostly referred in the context of right-wing backlash and censure of CRT in schools.
Multiculturalism; multiple cultures living together or interacting
Cannabis legalization;
Environmental Protections;
Trust in the scientific community; the scientific community. Examples include the academia, advisory organizations such as World Health Organization, Centers for Disease Control and Prevention (CDC)
Pornography;
Euthanasia;
AI-driven surveillance; whether the government should use AI technologies in surveillance.
Social Media Regulations; whether social media platforms and the associated companies should be regulated more 
Cryptocurrencies; cryptocurrencies, decentralized finance, and financial use of blockchain technology. A positive stance would favor those, a negative stance would be against or would call for more regulations.
AI Regulations; the general attitudes toward AI regulation or ethics
Use of AI in the workplace; the use of AI , often referred in the context of AI replacing human workers.
Religion (general); religion that speak on general terms without referring to a specific belief system
Net Neutrality; Net Neutrality, the principle that Internet service providers (ISPs) must treat all Internet communications equally.
Zoning Laws; (the US, and often state or city level) Zoning Regulations, that divide land in a municipality into zones in which certain land uses are permitted or prohibited. Often referred in the context of certain municipalities blocking new constructions to keep house/land values higher, which exacerbate housing crisis.
Whistleblowing/Leaks; whistleblowing or journalistic practice of leaking. Examples include Wikileaks, Panama Papers, Boeing whistleblowers, and often discussed through people such as Julian Assange, Edward Snowden, Chelsea Manning
Censorship; censorship by an administrative or regulative body (the government, the state, and also municipal education boards). Often referred in the context of some US schools banning certain books.
War on Drugs; the War on Drugs, the policy of a global campaign led by the United States federal government, of drug prohibition, foreign assistance, and military intervention, with the aim of reducing the illegal drug trade in the US. This is often referred in discussion of whether repression or legalization is better to reduce social harms caused by drugs.
War on Terror; the War on Terror, a global campaign led by the US against militant Islamist movements, such as Al-Qaeda, Taliban, ISIS and offshoots. Does not include the 2003 invasion of Iraq
Invasion of Iraq (2003); the invasion and occupation of Saddam's Iraq.
NATO/US Allies; the NATO and other US Allies. Mainly referred within the context of the US' bases abroad, and whether the Allies are doing enough (financially or militarily)
United Nations;
European Union;
Foreign military interventions by the US; the US military involvement in foreign conflicts. Examples include US involvement in the Syrian Civil War against Assad. Does not include Invasion or Iraq, War on Terror or War on Drugs.
US Support for Israel;
Israel;
Palestine; Palestinian people. A pro-Palestinian stance would stress the rights of Palestinians to self-governance and highlight Israeli oppresion, an anti-Palestinian stance would focus more on Hamas' leadership of Palestine, or poor governance by the Palestinian Authority over the West Bank.
Black Lives Matter movement;
Blue Lives Matter movement; Blue Lives Matter movement was a counter-movement against Black Lives Matter and stressed the dangers the law enforcement faces
MeToo movement; "the MeToo movement, and the practices employed or popularized by it, such as public exposures, social media campaigns against sexual harrasment and discrimination, and also ""cancelling"" of accused figures."
Donald Trump;
Joe Biden;
Barack Obama;
G.W. Bush;
Bill Clinton;
Ronald Reagan;
Richard Nixon;
Franklin D. Roosevelt;
Theodore Roosevelt;
Abraham Lincoln;
Hillary Clinton;
Bernie Sanders;
Alexandria Ocasio-Cortez;
Elizabeth Warren;
George Soros;
Elon Musk;
Ted Cruz;
Marco Rubio;
Mitt Romney;
Ron Paul;
Newt Gingrich;
John McCain;
Trump supporters; Specifically referring to Trump supporters, rather than Republicans, Conservatives, or Right-wingers
Biden supporters; Specifically referring to Biden supporters, rather than Democrats, Centrists, Liberals etc. 
Obama supporters; Specifically referring to Obama supporters, rather than Democrats, Centrists, Liberals etc. 
Bush supporters; Specifically referring to G.W.Bush supporters, rather than Republican, Conservatives, or Right-wingers
Hillary Clinton supporters; Specifically referring to H.Clinton supporters, rather than Democrats, Centrists, Liberals etc. 
Sanders supporters; Specifically referring to Sanders supporters, rather than Democrats, Leftists, Socialists etc. 
N/A; When target cannot be identified
Other; When there's an explicit target but is not listed
Belief in Trump-Russia Collusion; Regarding alleged collusion between Donald Trump and Russian authorities for political gain, including influencing elections
Belief in Russian Interference in Elections; Regarding attempts by Russia to influence US elections, without asserting whether this was intended to support Trump
Increased Surveillance; the increase in state or law enforcement surveillance activities. If the main focus is on the use of AI in surveillance, categorize under 'AI-Surveillance'

(Use the exact wording. Do not rename, paraphrase, or invent targets.)

Footnotes

  1. Törnberg, P. (2024). Large Language Models Outperform Expert Coders and Supervised Classifiers at Annotating Political Social Media Messages. Social Science Computer Review. https://doi.org/10.1177/08944393241286471↩︎

  2. Togay, Ö., Garcia-Bernardo, J., Kunneman, F., & Giachanou, A. (2026). Large Language Models Unpack Complex Political Opinions through Target-Stance Extraction. arXiv:2603.23531. https://doi.org/10.48550/arXiv.2603.23531↩︎