Almost Timely News: |
|
|
| |||||||||||||||
Almost Timely News: 
Unconventional Deep Research Use Cases (2025-07-13) :: View in Browser
 |
The Big Plug
My new book, Almost Timeless: 48 Foundation Principles of Generative AI is now available!
Content Authenticity Statement
100% of this week's newsletter was generated by me, the human. You will see bountiful AI outputs in the video. Learn why this kind of disclosure is a good idea and might be required for anyone doing business in any capacity with the EU in the near future.
Watch This Newsletter On YouTube 
Click here for the video version of this newsletter on YouTube »
Click here for an MP3 audio 
only version »
What's On My Mind: Unconventional Deep Research Use Cases
In this week’s newsletter, let’s dig deep into Deep Research. Dig, delve, whatever the popular term is these days. Deep Research is probably the most under-rated AI tool we have access to, and at the cost of a premium membership ($20 a month per user) it’s a steal.
Let’s take a look at a few different use cases for Deep Research to see how you could be using it more effectively.
Part 1: What is Deep Research?
We’ll start with an understanding of what Deep Research is. In software like ChatGPT, Gemini, Claude, Perplexity, Grok… basically every major AI foundation provider, there’s an option to select Deep Research (or something similarly named). These are AI agents. Once you give them a prompt, they go off and do their best to execute the prompt as directed.
This is a critical thing to understand: Deep Research agents are just AI agents. They’re tuned to do research, to gather information, yes. But they’re also capable of more than just strictly research. Any text output you could want that involves gathering and synthesizing information from public sources, they are capable of doing to one extent or another, and that’s the secret to their power.
Most paid plans offer Deep Research, so it’s probably just a matter of looking for the appropriate buttons in your interface.
Most Deep Research plans also have limits for how many reports you can run. For example, on ChatGPT, the Plus plan offers 10 full size reports a month and 15 shorter, less thorough reports.
My personal preference right now is Google Gemini’s Deep Research because the limits are incredibly generous (250 a month) and the reports are quite thorough, but use whatever you’re already paying for. No one platform is so dramatically better than another that there's an imperative to buy into just one.
One really important thing to mention here: Deep Research tools are still AI. That means they still make mistakes, which in turn means you still need to fact check their outputs. Don't assume that just because a tool did its research means that it got things right.
Part 2: Deep Research Prompting
The most important part of Deep Research is the prompt, because you want graet results - and with a limited number of uses, you can't afford to be re-running the same reports over and over again. To create great research prompts, you want to use a framework, and the framework I recommend is the Trust Insights CASINO Deep Research Framework.
Here's how you use it. First, you'll open up the AI tool of your choice in regular mode - NOT deep research mode. Then draft out an overview of the Deep Research task you want to accomplish. Once you've done that, drop in the CASINO PDF and this prompt:
Using the Trust Insights CASINO Framework (included as a PDF), ask the user one question at a time and only one question at a time until you have all the information you require to build a complete CASINO prompt for Deep Research. If the user has not provided the CASINO framework as a PDF, ask them to provide it. The goal is a rich, deep, comprehensive research prompt in the CASINO format. Specify each part of the CASINO format in subheadings in your final output. Your final output should be Markdown format. Remember that you are outputting a CASINO formatted prompt, not executing the prompt. Your results should be the CASINO formatted prompt.
What will happen next is that the generative AI tool of your choice will ask you questions until it has enough information to build the prompt. I recommend using a reasoning model for this task, which means:
Google Gemini 2.5 Pro
Anthropic Claude Sonnet or Opus 4 with Extending Thinking
ChatGPT with the o3 model
Any other appropriate model that has reasoning mode (there are many, many)
You'll answer the questions, and at the end of the process, you'll get a prompt. Copy that prompt, then start a new chat with Deep Research mode turned on, paste in your prompt, and you're off to the races.
Now, let's talk use cases for Deep Research.
Part 3: Company Context Engineering / Knowledge Blocks
In my new book, Almost Timeless: 48 Foundation Principles of Generative AI, Principle 25: It's Easier To Build With Bricks than Mud is all about having knowledge blocks, chunks of pre-built information you can drop into prompts to make them far more effective. The AI nerd herd now calls this "context engineering", as though we needed an even more belabored, confusing piece of jargon, but here we are.
Deep Research tools make this a breeze. Consider a research prompt starter like this, substituting your business, of course:
Let's build a comprehensive profile of the business Trust Insights, found at TrustInsights.ai. I want to know everything about this company - what its strengths and weaknesses are, who its customers are, who its competitors are, analysis of it like from business school frameworks like SWOT, BCG Growth Matrix, Porter's Five Forces, and whatever else was being taught during my MBA program. I want to avoid gossip and unfounded information, so we need to focus on credible information and sources.
Then paste in the rest of the Trust Insights CASINO framework prompt, drop in the CASINO framework, and generate the research prompt. Then as before, copy the entire research prompt and let the Deep Research tool go do the heavy lifting on your company.
What you end up with is a comprehensive research report about your company from the outside in - a valuable perspective, especially if you're too close to things. It's also a great way to fact check what AI knows about you - I did this recently for a client and they found the AI was referencing stuff that should have gone away after a rebrand.
What do you do with this output? Any time you need to do some marketing or strategy work, you now have this pre-built document - after you fact check it, of course - that you can insert into a prompt for far better results. For example, maybe you're writing up an SEO report. Adding in this document would give a lot more context to the AI writing the report, helping it connect the dots from the tactical stuff in your analysis to the big picture.
Part 4: Competitive Analysis
It doesn't take a great leap of imagination to realize that if you can do this for your company, you can do this for your competitors. Perform the exact same process for your competitors.
And what you'll find, especially when you start to look at the report in depth, is that you might or might not even show up in a competitor's report. That's a signal, especially for who you deem as peer competitors, that you might not be a peer at all.
Part 5: Job Hunting
AI agents that can go search the web are powerful, and few things are more time consuming than going job hunting. Job hunting (I used to be a recruiter, many moons ago) is a full-time profession. It's a B2B sale with all the phases of B2B selling, and what's for sale is you. If you want to succeed at landing a job, you have to be a solid B2B sales exec.
That in turn means doing lots of job board hunting. Fortunately, this is where AI agents like Deep Research tools are incredibly powerful allies. Let's start with this prompt starter, which you should adapt to your needs:
I've attached my LinkedIn profile, which I will attach as a document in the final prompt as well. I want you to help me find 20 jobs that are incredible fits for me. I'm looking to make a salary of $250,000 a year. I'm looking for full time, in the Boston area or remote. Remote or hybrid with no more than 3 days a week in the office is a must. 20% travel or less is a must. Ideally I'd like to work at a tech company, but anything will do that meets the other requirements. Using job boards like LinkedIn, Indeed.com, and any other job boards you know about, help me find a new job. I need exact, working URLs for the job descriptions so I can go apply for the jobs. I need to know the company name, industry, how long the job position has been open if you can tell, and all the other requirements. Score each job 0-10 for fit of my requirements, and score each job 0-10 for fit of my background. Return your results as bullet point listings in Markdown, in descending order by best fit to my requirements and my background.
Then paste in the rest of the Trust Insights CASINO framework prompt, drop in the CASINO framework, and generate the research prompt. Then as before, copy the entire research prompt and let the Deep Research tool go do the heavy lifting on job boards.
What comes back, of course, is a list of exactly what we asked for. Now, depending on your market, your background, and the answers to the CASINO questions, you might not get 20 results back. The tool might not be able to find much, in which case you can ask followup questions about where you should be job hunting, or how to rearrange your professional profile to better fit other industries while still remaining truthful.
If you're actively job hunting and you've kept a list of jobs you've already applied for, you might want to add a negative prompt in your starter, something like:
Exclude the following job listings from your research as I have already applied to them or they're not relevant to me. {include list of URLs}
Part 6: Requirements Gathering
Another great use of Deep Research, especially if you're trying out vibe coding, is to use the tools to help you build requirements. Requirements gathering used to be a long, tedious process that many developers saw as taking time away from actually writing code, and thus skipped or skimped on it. Naturally, actions have consequences and you end up having to do requirements documentation at some point anyway.
What if you could shortcut that pain? Deep Research to the rescue! Let's say we wanted to vibe code an application that could ingest data from our social media monitoring software, like Agorapulse (a Trust Insights partner), process it with AI, and produce useful insights like what topics resonate best with our audience.
We might start with a thought starting prompt like this:
Let's build out the requirements for a piece of software that can take in a spreadsheet of data (attached as a CSV file), read through it, extract out the metrics and dimensions, talk to an AI tool like Google Gemini or ChatGPT, and autonomously help me understand what's resonating with our audience and what's not. We use Agorapulse, and I'm pretty sure they have APIs, but it's easier for me to just download a spreadsheet as a CSV and manually load it. I don't know much about programming but I've heard Python can be good for this sort of thing. I want to know stuff like sentiment by topic, engagement by topic, etc. so that I know what to do more of and what to do less of as a social media manager. I know requirements documents contain user stories, functional requirements, domain requirements, and non-functional requirements.
Then paste in the rest of the Trust Insights CASINO framework prompt, drop in the CASINO framework, and generate the research prompt. Then as before, copy the entire research prompt and let the Deep Research tool go do the heavy lifting on software development sites.
We answer the questions as best as we can, then let the agent go to work. What comes back is a document we review with a developer or coder (or AI) to enhance, then convert into a work plan, then send to a human or AI coding system to bring it to life.
Generally speaking, most developers - human and AI alike - won't be able to code directly from a requirements document, but it's trivial to have the AI of your choice convert the requirements document to a file-by-file work plan that anyone skilled at coding can implement.
Part 7: Media Pitching for PR
Let's go from social media to public relations. What if we wanted to use Deep Research for public relations? It's astonishingly useful for this context.
One of the most difficult things to get a hold of in PR is good media lists - lists of people and companies who cover what you want coverage for. In the age of AI, getting a media placement in a top tier publication is still nice, but getting lots of placements is better, especially in contextually relevant media outlets.
Let's say I want to get some media coverage for my book. I could just spam the media world about it, but that's a terrible idea that gets you blacklisted and banned from pretty much everywhere. No one loves those wildly off-target media pitches. Instead, let's use Deep Research!
Here's an example starter:
I'm launching my new book, Almost Timeless: 48 Foundation Principles of Generative AI, and I want to obtain some coverage from the media about it. I don't have a PR agency or a PR team, but I am more than happy to do interviews, send review copies, etc. to relevant media outlets. I've attached a copy of the manuscript so you can better understand what it's about and help me pitch it. What I need most are media lists. I need a list of 5 mainstream media publications and the specific reporters or journalists to pitch my book to that would find it worth their time to cover. I need a list of 5 LinkedIn influencers who would cover my book. I need a list of 5 podcasts that would have me as a guest to talk about my book. I need a list of 5 YouTubers I could record video with to cover my book. Develop your media lists from your research and score each media outlet, influencer, journalist, and creator 0-10 as to how good a fit they and their audience are for my book, then produce the four lists I asked for, ordered in descending order by goodness of fit. I need the individual names, their contact URLs or emails, the outlet or channel they work for where relevant, and an explanation of how to pitch them specifically. Return your results in Markdown format.
Then paste in the rest of the Trust Insights CASINO framework prompt, drop in the CASINO framework, and generate the research prompt. Then as before, copy the entire research prompt and let the Deep Research tool go do the heavy lifting in media circles.
The end result is usually an excellent document you can immediately use; however, you may need to run a second report if you weren't specific enough in the first one, or if there are media outlets that feel unrealistic for you to pitch. You can pitch them yourself or hand off the list to your PR team/firm to use.
Part 8: Content Gap Analysis
One of AI's most powerful latent skill sets is to know what's missing. Because AI models have been trained on the world's public information, and AI agents like Deep Research have access to search catalogs, they often can know about things and see the big picture in a way that's simply not possible for our mere human brains. Our brains can only hold so much information and recall it successfully, and AI can exceed that many, many times over.
Here's a question every creator and marketer should be asking: what aren't we doing that we should be? What aren't we covering that would make our audiences deliriously happy - or at the very least, willing to stay engaged? We can and should just ask them, but one of the challenges of asking people what they want that they're not getting is that they're often not even aware of what's possible. If all you've ever had was gruel, you'd never know you were missing steak.
So we use Deep Research to supplement - SUPPLEMENT - our first party data. You did see the emphasis on SUPPLEMENT, right? Okay. Just checking.
Here's a starter:
I need to perform a content gap analysis on my newsletter, the Almost Timely Newsletter, at
. You'll need to read past issues, infer who my likely audience is, and then understand what that audience's needs, pain points, goals, and motivations are. From that, research similar publications to mine and then identify what content strategies, tactics, and topics I'm not covering that I should be, what things I'm not doing that I should be, what things would make my audience more loyal and more engaged if I did. My goals for my newsletter are to educate and entertain, then to drive business for my company, Trust Insights, by encouraging my readers to buy my books, courses, and hire Trust Insights for bespoke AI consulting.
Then paste in the rest of the Trust Insights CASINO framework prompt, drop in the CASINO framework, and generate the research prompt. Then as before, copy the entire research prompt and let the Deep Research tool go do the heavy lifting in your publication and similar publications.
As with the other queries, you'll get back a report that should have everything you asked for, as long as the information is available. And from that, you should be able to take action on the results.
Part 9: Wrapping Up
I hope in this tour you've seen how flexible Deep Research agents can be. If you can think it, if you can describe it, you can use the Trust Insights CASINO prompt framework with your favorite AI tool to craft it.
A couple of final points as we wrap up. First, stay in the same ecosystem. If you're using ChatGPT for your Deep Research, do the CASINO prompting in ChatGPT as well, ideally using the o3 model. If you're using Gemini for your Deep Research, do the CASINO prompting in Gemini 2.5 Pro. The reason for this is simple: these models all share vocabulary and concepts within the same family.
That means using ChatGPT to refine the prompt and Gemini to execute it is probably not going to yield as good a set of results, because under the hood, Gemini Deep Research is powered by Gemini, and every model knows itself best.
Second, the amount you fact check should be proportional to risk. I will use Deep Research to uncover sources I didn't know about for high risk inquiries in fields like finance, health, and law, but then I'll go to those sources, download them directly, and use them in a tool like NotebookLM where I can see exact citations of where AI is getting its information. Deep Research tools don't do that as well, so if it's something really important or really risky, go old school and grab the sources yourself.
I would, for example, never use synthesized Deep Research for a critical health issue. The risk of hallucination is unacceptably high. I would use Deep Research to surface peer-reviewed studies I might not have found on my own.
I hope you found these examples useful, and that you give it a try yourself.
