Using an Agent to Summarise Documents for a Research Audience

Ray Poynter, 12 August 2024

---

Here is an example of using a simple agent to make your life slightly easier. I will show the example using Copilot, but you could just as easily use ChatGPT, Gemini, Claude etc. You can use the same Instructions that I use for the Copilot.

The Agent takes a file you have uploaded and produces a summary to a standard format, tailored to the market research and insights ecosystem. You can do this manually each time, by typing in instructions as a prompt. But it is easier to set up the Agent, and you are less likely to forget to include various bits.

I have specified that the Agent should ask how long you want the summary to be.

Example

Here is an example of it in use. I am using an interesting document from Stanford University, Future of Work with AI Agents.

I clicked on the Agent to load it, uploaded the document, and clicked the arrow to start it running.

The Agent replied with

“Thanks for uploading the document titled Future of Work with AI Agents – Stanford University – June 2025 

Before I begin the summary, could you let me know:

How many words would you like the summary to be?

You can choose a rough target (e.g., 200, 500, or 1000 words), and I’ll aim to stay within ±50 words of that. Let me know if you’d prefer a bullet-point format or a more narrative style too.”

I typed in 300 and hit the arrow again. Here is the 310-word summary it produced.

“Top-Level Description (~50 words)

This academic white paper from Stanford University explores how AI agents are reshaping the future of work. It focuses on the integration of autonomous AI systems into knowledge work, including research, strategy, and decision-making. The paper outlines emerging capabilities, risks, and implications for productivity, workforce dynamics, and ethical governance.

Source and Credibility

Produced by Stanford’s Human-Centered AI Institute in June 2025, the paper draws on interdisciplinary research, including computer science, sociology, and economics. The authors are respected academics and practitioners, lending strong credibility. Data sources include recent field studies, expert interviews, and pilot implementations across sectors.

Detailed Summary

Key findings relevant to market research include:

• AI agents as collaborators: Autonomous agents are increasingly used to assist with tasks like data synthesis, trend detection, and hypothesis generation—especially in fast-paced environments like consumer insights and innovation labs.
• Workflow transformation: AI agents are shifting roles from execution to orchestration. For example, agents can now design survey logic, monitor fieldwork, and flag anomalies in real time.
• Human-AI teaming: The most effective use cases involve hybrid teams where researchers guide AI agents, ensuring contextual relevance and ethical oversight.
• Skill evolution: Demand is rising for “AI literacy” among researchers—skills in prompt engineering, agent supervision, and ethical evaluation.
• Case examples: The paper highlights pilot programs in retail and healthcare where AI agents improved segmentation accuracy and reduced time-to-insight by 40%.

Implications for Research & Insights

• Accelerated insight generation and reduced manual effort.
• New roles emerging: AI strategist, agent trainer, and ethics auditor.
• Greater emphasis on interpretability and transparency in AI-driven findings.

Reservations/Caveats

• Limited longitudinal data on agent performance.
• Risk of over-reliance on AI-generated insights without human validation.
• Ethical concerns around bias and data provenance.

Recommendations for Further Reading

• “AI Agents in Market Research” – ESOMAR, 2024
• “Human-AI Collaboration in Knowledge Work” – MIT Sloan Review, 2025
• Stanford’s AI Ethics Framework – updated April 2025”

The Instructions
Here are the instructions I used. You can copy and paste these into your LLM and it will be ready to run.

“Purpose: Analyse an uploaded document and create a structured summary aimed at market research and insights professionals.

Workflow:

1. Ask the user: “How many words would you like in the summary?” Target the requested length, aiming to be within ±50 words.
2. Read and analyse the document. Identify the type of document (e.g., report, white paper, press release, academic article). Focus on content relevant to market research and insights, such as trends, methodologies, case studies, data sources, and implications for research practice.
3. Write the summary in this structure:
4. a) Top-level description (~50 words) – what the document is, its main focus, and primary message.
5. b) Source and credibility – who produced it, when, and the likely credibility based on author reputation, methodology, and data sources.
6. c) Detailed summary – main findings, arguments, and evidence, with relevance to market research.
7. d) Implications for the research and insights ecosystem – why it matters, linked to current industry themes (e.g., AI in research, consumer behaviour shifts, data privacy).
8. e) Reservations/caveats – potential biases, methodological weaknesses, missing data, or outdated information.
9. f) Recommendations for further reading – specific related reports, articles, or frameworks.

Tone and style:

• Write clearly and concisely; avoid unnecessary jargon.
• Use professional, objective language suitable for an industry audience.
• When giving use case examples, tailor them to market research contexts (e.g., survey design, qualitative analysis, segmentation, concept testing).
• Use bullet points for clarity where appropriate.

Additional guidance:

• Highlight the most relevant data or statistics from the document.
• Summarise case studies most applicable to research and insights.
• Link the content to practical applications in commercial, social, or academic research where possible.”

Yes, this sort of simple Agent will improve your work
This is a simple case, but if you haven’t started using agents yet, this is where you should begin. If you type in these prompts manually, you will typically find that you type in something much shorter. The first version is usually not good enough, so you either start iterating or settle for something not very good.

Once you have a valuable collection of simple agents, you will soon progress to making more complex agents.

Don’t know how to set up an Agent?
If you don’t know, ask an LLM. Ask something like “I want to set up a simple agent in X and I have the instructions I want it to use, how do I set it up?” Where X is your platform.

You can also check out the recording of a recent presentation I gave at a NewMR Webinar by clicking here.

Related