Prompting technique

Context-Augmented Prompting

Large Language Models for Toxic Language Detection in Low-Resource Balkan Languages, arXiv:2506.09992 (2025) · read the paper

What it does

The method consists of adding 2-3 sentences of background context to the prompt before the main text for analysis. This allows the LLM to correctly interpret cultural references, sarcasm, and hidden meanings, significantly improving the accuracy of toxicity classification and other tasks requiring situational understanding.

Task typeclassification
Shapesingle shot
Effortsingle step
Addressesanswers that are close but wrong; the model losing what was agreed earlier
Published2025-06-11
AuthorsAmel Muminovic, Amela Kadric Muminovic

When it helps

Use this technique when an LLM needs to analyze content requiring understanding of situational or cultural context: comment moderation, review analysis, message tone determination. Especially effective for low-resource languages, detection of toxicity, sarcasm, and hidden aggression. Apply when you have access to metadata (discussion topic, product description, video context) that can be added to the prompt. Ideal for classification tasks where without conte

A prompt you can paste

Generic by design: it applies the technique without knowing your task. Adapt the marked parts.

Prompt
You are an AI model specialized in detecting toxic language in low-resource Balkan languages. Your task is to analyze comments and determine their sentiment, considering the specific context provided.

### Context for Analysis

**Context Type:** [Specify context type, e.g., Video Topic, Forum Discussion, Social Media Post, Product Review]
**Context Description:** [Provide 2-3 sentences describing the context. For example: "The video is about a popular local musician performing a traditional folk song.", "This is a discussion thread on a forum about upcoming local elections.", "The social media post is an announcement from a local government regarding new traffic regulations.", "This review is for a newly opened restaurant in the city center."]

### Comment to Analyze

**Comment:** [Insert the comment text here]

### Task Instructions

Analyze the provided **Comment** based on the **Context for Analysis**. Determine the following:

1.  **Toxicity Level:** Classify the comment into one of the following categories:
    *   Non-toxic
    *   Mildly toxic
    *   Highly toxic

2.  **Primary Sentiment:** Classify the sentiment as:
    *   Positive
    *   Negative
    *   Neutral
    *   Sarcastic/Irony

3.  **Key Point:** Briefly summarize the main point or the core of the comment.

**Output Format:**
Please structure your response as follows:

If this one does not fit, the two closest alternatives in the corpus are RICE Framework (Role, Instruction, Context, Examples with Rationale) and Goal-Reversed Prompting, which target the same failure from a different angle.

Worked example

The same technique applied to a concrete job: sort inbound support tickets by urgency. Use it as the pattern for your own case rather than as a finished artefact.

Worked example
You are an AI assistant specialized in content moderation and toxic language detection. Your task is to analyze incoming support tickets and classify their urgency.

### Context for Urgency Assessment

- **Company:** [Specify Company Name, e.g., "TechSolutions Inc."]
- **Product/Service:** [Specify Product/Service, e.g., "Cloud-based CRM software"]
- **User Base:** [Specify User Base, e.g., "Small to medium-sized businesses"]
- **Common Issues:** [List typical issues, e.g., "Login problems, data sync errors, feature bugs"]
- **Support Channels:** [List support channels, e.g., "Email, In-app chat, Phone"]

### Ticket for Analysis

**Ticket ID:** [Insert Ticket ID]
**Subject:** [Insert Ticket Subject]
**Body:** [Insert Ticket Body - the full text of the support request]
**Channel:** [Insert Channel where ticket was received, e.g., "Email"]
**User Tier (if applicable):** [Insert User Tier, e.g., "Enterprise", "Standard", "Trial"]

### Task

Analyze the provided ticket using the context above. Classify its urgency into one of the following categories:

- **Critical:** System outage, complete service unavailability, security breach, or issues affecting a large number of users/high-tier clients. Requires immediate attention.
- **High:** Significant feature malfunction, data loss or corruption, or issues preventing core functionality for a single user or small group. Requires prompt attention within hours.
- **Medium:** Minor bugs, usability issues, or feature requests that do not block core functionality. Can be addressed within a standard support timeframe.
- **Low:** General inquiries, feature suggestions, or requests for information that do not impact current operations. Can be handled asynchronously.

Provide your classification and a brief justification based on the ticket's content and the provided context.

**Output Format:**
- **Urgency Level:** <Critical/High/Medium/Low>
- **Justification:** <Brief explanation linking ticket content to urgency criteria and context>

Get this written for your actual task

Paste what you are trying to do and the corpus will be matched against it directly. Free, no account, about ten seconds.

Free · no signup · ~10s
0.00match confidence
single retrieval pass
Prompt for your task

      

That number is low on purpose, and it is real. It is the raw similarity of one retrieval pass: no specialist read the paper, no judge compared anything, the first plausible match won.

6,235techniques in the corpus
one of which is this page

Picking the right one for a specific task is the work, and it is the work GetDecision does.

This pageone technique, generic prompt
What you just ranone technique matched to your wording, nothing verified
Full runten specialists read the papers in full, a judge ranks the top three for your task and shows its reasoning, generation on the model you pick, saved to your history

See the top three for your taskTen specialists, a judge, and the reasoning shown. Free account, first run included.

Run the full analysis

Related techniques

RICE Framework (Role, Instruction, Context, Examples with Rationale)A framework for structuring prompts from four components: Role, Instruction, Context, and Examples with Ration…Goal-Reversed PromptingA method that consists of inverting the query goal: instead of asking the LLM to find the best option, it is a…Multi-Turn Conversational PromptingMethod of multi-step dialog interaction with LLM where complex tasks are broken into a chain of simple steps. …Debugging Prompts FrameworkA systematic framework for diagnosing and preventing typical errors of language models in instruction understa…

All techniques · Failure modes and fixes