When discussing prompt engineering and context engineering and how they differ, what better way to illustrate it than asking the AI to do it?

I said yesterday that context engineering was a future topic. Well, the future is now! I decided to do so by asking the AI – namely ChatGPT 5.2 – to “take a crack” at discussing the difference between prompt engineering and context engineering. Then, I fed that result into Google’s NotebookLM to see what kind of infographic it would create.

My initial prompt for GPT 5.2 was as follows:

“Please discuss the difference between prompt engineering and context engineering in plain terms and provide examples of each.”

GPT 5.2 gave me a detailed result, then I asked it a follow-up question, as follows:

“Please map this distinction directly to eDiscovery use cases.”

The result of the two queries are incorporated into this document here. A few quick highlights:

The beginning of the response to the first query is this:

The short version

• Prompt engineering is what you say to the AI in a single interaction.
• Context engineering is what you give the AI to work with before it ever answers.

If prompt engineering is asking a good question, context engineering is setting the room, laying out the files, and explaining the situation before the question is asked.

GPT 5.2 then proceeded to go into more detail about both prompt engineering and context engineering, discussing what you’re telling the AI for each, then providing a simple example of each. For prompt engineering, GPT 5.2 also provided a “Slightly more engineered example”, while for context engineering, it provided a “Legal-tech-style example (very relevant to your world)” – what can I say, ChatGPT knows me well at this point! 😉

Here’s what that example looked like:

You give the AI:

• A discovery order
• The ESI protocol
• Prior court rulings in the case
• Your firm’s standard review workflow
• A list of custodians and data sources

Then you ask:

“Identify discovery risks and recommend next steps.”

That’s context engineering.

Pretty compelling. And that’s before I asked it the follow up eDiscovery question.

GPT 5.2 went on to further illustrate the differences between prompt engineering and context engineering, including this “simple analogy”:

• Prompt engineering = giving directions to a chef

“Make it spicy, serve it hot, plate it nicely.”

• Context engineering = giving the chef the ingredients, the dietary restrictions, the menu, and the guests’ preferences.

Great directions won’t help if the chef doesn’t know who they’re cooking for.

When I asked the follow-up eDiscovery question, it responded with the “eDiscovery translation (one-sentence version)”, as follows:

• Prompt engineering in eDiscovery is how you ask the AI to perform a task (e.g., “summarize,” “classify,” “flag privilege”).
• Context engineering in eDiscovery is how you define the legal, factual, and procedural universe the AI is allowed to reason within.

In discovery, context engineering is the difference between “interesting output” and “defensible output.”

Interesting take. As requested, GPT 5.2 also provided a discussion of several eDiscovery use cases. Here’s one of them:

Custodian interviews

Prompt engineering example

“Generate custodian interview questions for a sales executive related to this litigation.”

You’ll get a generic list—often useful, but shallow.

Context engineering example

Before asking anything, you provide:

• Case theory and claims/defenses
• Custodian role, tenure, and reporting structure
• Known data sources (Teams, Salesforce, personal phone)
• Prior interview responses from other custodians
• Preservation scope and date ranges

Then you prompt:

“Draft a tailored custodian interview for this individual, focusing on communications related to pricing strategy during the relevant period.”

Why this matters in eDiscovery:

The AI avoids irrelevant questions, aligns with legal issues, and produces an interview that could actually survive scrutiny.

GPT 5.2 proceeded to provide similar analysis for Early Case Assessment (ECA), Privilege identification, Review acceleration and GenAI summaries, Technology-Assisted Review (TAR) / CAL workflows, RAG systems in eDiscovery (where context dominates) and Meet-and-confer and motion drafting. All very interesting.

The response concludes with:

One-line takeaway you can reuse

Prompt engineering helps AI answer questions.

Context engineering ensures AI answers the right questions, using the right evidence, in a way that can be defended.

Personally, I would use the words “helps ensure”. Hallucinations and omissions can still happen with either approach – QC of results is still necessary. But I would agree that an approach based on context engineering improves the relevance of the answer that the AI provides and possibly reduces the potential of a hallucinated output (though, no guarantees). The end result is a response that is more targeted and useful to obtain the information you seek.

I then took the resulting document I created (again, available here) and loaded it into NotebookLM and asked it to generate an infographic, which is here:

Would I go so far as to say: “In discovery, context engineering is the difference between ‘interesting output’ and ‘defensible output’”? I’m not sure I’d go that far. But I would say that the result with context engineering is a more useful output.

So, what do you think? Do you agree with the AI’s assessment of prompt engineering and context engineering? Please share any comments you might have or if you’d like to know more about a particular topic.

Disclaimer: The views represented herein are exclusively the views of the authors and speakers themselves, and do not necessarily represent the views held by my employer, my partners or my clients. eDiscovery Today is made available solely for educational purposes to provide general information about general eDiscovery principles and not to provide specific legal advice applicable to any particular circumstance. eDiscovery Today should not be used as a substitute for competent legal advice from a lawyer you have retained and who has agreed to represent you.