***A few friends... ***

Gemini’s conception of [hyperrealistic image of scholar doing deep research]. Not sure it’s hyperrealistic, but definitely interesting.

... have recently written posts of their own about using AI for deep research. Since they’ve got some great nuggets, I’m going to leverage their writings and give a quick summary of the top methods for doing high quality deep research with LLMs.

In this post, I’m drawing extensively on a post written by Maryam Maleki (UX Researcher at Microsoft) for people doing product research: How to Do High-Quality AI Deep Research for Product Development Here, I’ve generalized it a bit and given it my own flavor.

Here are the top few tips about getting Deep Research mode to work well for you:

Be clear about what you want.

Keep in mind: You want credible content. Prompt it that way.

In order for the AI to work, you need to tell it what kind of sources you think are reliable and credible. If you can, give it a list of several resources as guidance.

In these patterns below, items in { } and italics are variables. You need to pop in the values you need to get the effect you want.

Pattern:

[ Do deep research on {TOPIC}. Generate {n} credible sources with links that can be used for this research.

Prioritize: {BOOKS / ACADEMIC PAPERS / CASE STUDIES}

For each source, provide: the Title, the URL, a short snippet about why it’s relevant, tell me the Source type. ]

Example:

[ Do deep research on Rocky Mountain locusts. Generate 10 credible sources with links that can be used for this research.

Prioritize: academic papers

For each source, provide: the Title, the URL, a short snippet about why it’s relevant ]

Doing this in Gemini will create a 4,000 word essay about Rocky Mountain Locusts. It will ALSO give you section VII, which has Ten Credible Sources for Rocky Mountain Locust Research. It also creates a reference list for the entire document, with section VII containing the best of the entire list.

By contrast, doing this in ChatGPT 5/Thinking or Claude Sonnet 4.5 gives you exactly what you asked for–they give you the list-of-ten.

Review the AI-generated results for quality

I note in passing that the Gemini-created document is pretty good, but the list of 10 papers was a little mixed in quality. (One paper was very tangential, one paper was just a link to Wikipedia, and one paper wasn’t accessible at all.) I clicked through all of the links to verify that they were real and on-target.

If the results aren’t what you want, feel free to iterate until you get the result quality you need.

Press enter or click to view image in full

Ask for contrary points of view (don’t just confirm!)

Research isn’t just about collecting references — it’s also about understanding the space, both in terms of what you know and what counterarguments you might want to consider.

In reading through the Rocky Mountain Locust collection, you’ll notice that one of the main hypotheses about the disappearance of the locust is that the rangeland where it lived and bred was increasingly plowed up for farmland.

You should ask about other opinions:

Pattern:

**[ GIve me different explanations for {TOPIC}. Are there other points of view that have been considered in the literature? **

**For each source, provide: the Title, the URL, a short snippet about why it’s relevant. ] **

Example:

**[ Give me different explanations for why the Rocky Mountain Locusts disappeared. Are there other points of view that have been considered in the literature? ****For each source, provide: the Title, the URL, a short snippet about why it’s relevant. ] **

Interestingly, Gemini merely did an okay job of this step–ChatGPT was reasonably good, but Claude did a spectacular job of highlighting 11 different hypotheses about what happened. (To see Claude’s output, [here’s the document](https://searchresearch1.blogspot.com/2025/11/GIve me different explanations for why the Rocky Mountain Locusts disappeared. Are there other points of view that have been considered in the literature? For each source, provide: the Title, the URL, a short snippet about why it’s relevant.).) This also suggests that you should get multiple AI opinions to improve the quality of your research!

Double Check Everything

We still live in a hallucinatory world. As great as AI generated content is, I still double check everything. In her post, Maryam has a great set of questions (below). This is what is on my mind as I read through EVERY claim and EVERY linked document. You should too.

• Source Quality — Is it recent, reputable, and methodologically sound?
• Fact Containment — Only use approved notes/sources.
• Triangulation — Every claim needs at least two independent sources.
• Original-Source Tracing — Don’t rely on LinkedIn slides, Twitter posts, or a quote in a blog. Find the earliest credible publication.
• Hallucination Sweep — Audit the final draft. Remove or qualify any claim not directly supported.

Search Research Summary

When using AI for deep research, keep in mind 3 heuristics:

***1. Be clear about what you want. *** Not just in content, but in form and quality. Be explicit–give examples–ask for everything you want.

2. Review the results for quality. Do this step immediately, and change the prompt if need be to get what you really seek. Iterate!

***3. Ask for contrary points of view. *** Don’t give in to confirmation bias–proactively ask about other perspectives on the questions you’re researching.

***4. Double check everything. *** No surprise here, but be sure to leave enough time to do this. Don’t just copy/paste what you’ve found.

Keep searching!