In the first article in this series, AI fundamentals: history, terminology and limitations, we explored the origins of AI and the terminology surrounding it. This article considers how claims professionals can improve the accuracy, relevance and usability of AI-generated output by providing clear, structured and purposeful instructions, known as prompts.
Poorly structured prompts can produce incomplete, misleading or overly generic responses, potentially introducing risk rather than reducing it. Effective prompts establish the task, context and standard of output required. As AI models evolve, prompting approaches may also need to change.
Why effective prompting matters
A well-constructed prompt helps an AI tool respond in a way that aligns with your objectives. Consider whether a colleague with no prior context would understand what you needed from the same instructions. If not, the prompt may need more detail. Clear guidance can help the model return more focused, concise and actionable information.
For example, “summarise this claim” may produce a high-level response that omits important details. A more useful instruction would specify the audience (such as the Head of Claims), the level of detail (such as a concise bullet-point summary) and the issues to address (such as liability and quantum).
The rules of prompting
Different AI models may respond differently to prompt length, structure and examples. Check the guidance for the particular tool and version approved by your organisation, and test whether the output meets your needs.
Tailoring prompts to the tool
Claude: Give clear context and instructions. Ordered steps and examples may help shape the format and tone of the response.
GPT models: Focus on the desired outcome and success criteria. Start with a concise, clearly structured request, adding detail when needed.
Microsoft Copilot: Specify the goal, context, expected format and, where possible, the relevant documents or data sources.
Five elements of a useful prompt
1. Goal: Define the end objective, such as producing an executive claims summary.
2. Context: Give relevant background, including the audience and purpose.
3. Success criteria: Explain what a good response should achieve, such as being concise and clearly structured.
4. Constraints: Set boundaries, including length, tone, output format and exclusions.
5. Source: Identify the documents or data the AI should use to reduce unsupported claims.
AI in the claims environment
Claims professionals can use approved AI tools to support document summarisation, trend analysis, correspondence drafting, file review and data extraction. These activities can be time-consuming and often involve repeatable tasks that benefit from clear prompts and appropriate oversight.
The LMA AI in Claims survey cited in the original article identified Microsoft 365 Copilot as the most widely used AI tool among surveyed claims professionals using AI. The following five examples are framed for Copilot, although the underlying prompting principles are more widely applicable.
Five practical prompt examples
1. Complete a Market Development Communication template
Example prompt:
“Complete the Market Development Communication template using the information in the attached report. Summarise the key developments relevant to the market, written for participating insurers who need an overview for decision-making. Include only information supported by the report and return the output in the template’s structure.”
Why it works:
Clearly defines the task, audience, and output structure, while constraining the AI to use only the provided report.
2. Search emails over the last 6 months containing CLA watchlist code ‘QUAN’ for an internal review
Example prompt:
“Search my email inbox for any emails containing the CLA watchlist code ‘QUAN’ that were received within the last six months. This is for an internal review. Return the results in a table showing sender, recipient, date received, subject line, and a brief summary of the email’s relevance.”
Why it works:
The level of specificity removes ambiguity and ensures the AI retrieves only relevant results. By requesting the output in a structured table, it is immediately actionable and supports sufficient review.
3. Analyse loss data
Example prompt:
“Review the attached loss data and identify any unusual trends observed from January to December 2025. This is for an internal claims review, so focus on anomalies, outliers, sudden changes, or emerging patterns. Summarise your findings in a short report with bullet-point insights and note any areas that may warrant further analysis.”
Why it works:
By guiding the AI on what to look for, it avoids generic summaries and produces insight-driven analysis.
4. Draft a claims update for the broker
Example prompt:
“Using the attached claim notes, draft a concise update for the broker explaining the current position and any outstanding actions. This is for a routine claims progress update, so focus only on material developments, decisions, and next steps. Use clear, plain English wording and present the output as a short email-ready summary or four to six bullet points.”
Why it works:
By defining the audience, purpose and content scope, it ensures the output is practical, easy to use and has the correct tone. By instructing the AI to focus on material developments it reduces unnecessary detail and ensures the output is concise and relevant.
5. Re-draft a letter to the Insured
Example prompt:
“Using the attached letter, re-write it in plain English for the insured. This is a claims update, so ensure the tone is clear, supportive, and easy to understand. Summarise our current position, key facts, and any required next steps. Retain all policy conditions displayed in italics exactly as written and present the letter in a Word document format.”
Why it works:
Carefully balances clarity, tone, and compliance requirements. The instruction to preserve policy wording verbatim ensures legal or contractual language is not altered. Simplifying the surrounding text improves customer understanding, producing a compliant and customer-friendly document.
Across these examples, effective prompts define the objective, provide relevant context and set suitable constraints. This helps make the resulting output more relevant and easier to review and use.
Summary
By using structured prompts, adapting their approach to different AI tools and maintaining appropriate oversight, claims professionals can use AI more effectively in day-to-day activities.
AI can improve efficiency and support quality, but it is not a substitute for professional judgement. All outputs should be reviewed and validated before use. Claims professionals must also follow their organisation’s policies on AI use, confidentiality and information security, including using only approved tools for sensitive material.
Editorial note: AI was used in preparing the original article to define what a good prompt looks like and how the prompt examples could be improved.
Up next: “When the Algorithm Gets It Wrong” exploring how hallucinations can affect the use of AI in claims.