August 2, 2025 • 19 mins
Discover how ChatGPT's Deep Research transforms everyday research for businesses and curious minds alike. Learn its strengths, pitfalls, and insider tips to supercharge your workflow without losing your critical edge.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
Hey everyone, and welcome back to the AI breakdown, your deep dive into how AI is shaping work life and everything in between.
(00:07):
Today's episode is all about digging under the hood of a genuinely powerful feature chat GT's deep research.
If you've ever felt lost trolling, endless browser tabs, or you wanna know if there's finally an AI that can actually do proper research, this one's for you.
And yes, I speak from experience.
I use deep research regularly for the show's preparation.
(00:29):
Product development, competitor analysis, and before buying expensive things like booking a holiday or my latest pair of running shoes, is it handy? Absolutely.
Does it get everything right? Well, not quite, but more on that later.
Here's what to expect in this episode.
We'll unpack what deep research is and why it's such a leap forward.
(00:50):
Dig into how it works.
Look at real business and personal use cases.
Compare it head to head with old school research Spotlight where it shines.
Give practical tips, and for the eye explorers out there, do a quick scan of alternatives.
So let's get stuck in and start with the basics.
(01:13):
What is this thing and how is it different from just Googling or flinging a question at your favorite chatbot? In a nutshell, deep research is chat GBTs way of acting less like a friendly robot that guesses answers and more like a virtual research analyst.
It's powered by an advanced O three large language model that actually browses the web reads articles, scans, PDFs, or even images, and then builds a plan to explore and answer your question in a serious step-by-step way.
(01:43):
So not just tell me about X, but I need you to troll dozens of sources, pull together the relevant bits, and give me a properly structured source cited report.
Deep research isn't for quick fun facts.
It's for genuinely big sticky questions.
Things like give me a competitive analysis of the smart speaker market citing up to date industry data, or what's the latest research on AI and education with recent case studies in the uk.
(02:11):
It's built for working professionals, think analysts, consultants, scientists, lawyers, students, but also anyone facing a question where the answer isn't just sitting at the top of Google's results.
Here's how it works.
You select deep research mode inside chat, GPTs chat box, type in your question, and you can even upload files or spreadsheets for context, sometimes chat GPT might prompt you with questions for extra details like the timeframe you care about.
(02:39):
Or specifics you wanted to focus on.
That context helps it build its search plan.
Once you kick it off, you'll see a sidebar in chat, GPT, tracking what it's doing, which sites it's searching, what docs it's reading, and it's evolving thought process.
This isn't instant, we're talking five, 10, even up to 30 minutes, depending on how gnarly the question is.
(03:02):
But you can walk away and come back to a detailed sectioned report complete with clickable sources.
Its methodology is revolutionary.
Unlike standard search engines that scrape surface level information, deep research conducts a thorough investigation sometimes through hundreds of sources.
It employs a structured, methodical approach, planning the research strategy, casting a wide net initially, then progressively narrowing its focus based on findings.
(03:31):
When initial results aren't comprehensive enough, it adapts its search parameters and dives deeper into specialized sources.
It can even execute Python code to analyze statistics or visualize data comparisons on the fly.
This strategic adaptive research process mirrors what you'd expect from a highly skilled human research assistant, but at machine speed, what you get back absolutely trumps a quick search.
(03:57):
The output is a full structured report.
We're talking thousands of words with easy to follow sections, bullet pointed, key facts, summary tables, and every claim or stat footnoted to the source, reviewers and professionals who have tried it from lawyers to engineers to analysts.
See, it's often better than what they'd get from a junior team member.
(04:19):
It's not unusual to get a report that would take a person a full day or longer to compile, where you'd usually end up lost in 20 browser tabs and four barely distinguishable PDFs.
Deep research hands you the whole thing neatly packaged With clickable citations, you shift from being the person doing the grunt work of find, read, summarize, repeat to being the supervisor.
(04:42):
Define the big question.
Give guidance and then use your time reviewing and judging instead of trolling.
It's great for market intel.
Imagine you're doing a market analysis for a new startup entering the e-bike market.
Rather than clicking 30 top 10 E-bike lists and making an Excel disaster, you can ask deep research.
Who are the current leaders in the e-bike market in the uk and what's their market share and what new technologies are emerging? What you get is a multi-source report that cross references, industry reports, news, and even niche cycling forums, all with links to chase up if you care for a deep dive in business.
(05:20):
Deep research also excels at competitor analysis if you need to understand the landscape.
Then instead of spending days building a briefing deck, you can ask for an UpToDate competitive briefing that scans news outlets, industry white papers and forums.
Delivering a comprehensive summary with links for each key point.
It's a serious tool for due diligence too.
(05:42):
In one experiment, an antitrust lawyer prompted deep research for an 8,000 word legal analysis and was genuinely impressed with the results.
He rated it comparable to work from an entry level attorney, something that would typically require 15 to 20 billable hours and said he'd use this model professionally.
In another real case, an architect asked deep research for a detailed building code checklist for a large educational building.
(06:09):
The system consulted 21 different sources, including state regulations, international code, PDFs, and accessibility standards, producing a comprehensive 15,000 word regulation site report.
In just 28 minutes, she estimated the same task would've consumed an entire working day, but it's not just for business folks.
(06:31):
For students and researchers, deep research transforms.
Literature reviews need a multi-sided review For a university essay, you might prompt it with something like Compare recent studies on climate adaptation strategies in coastal cities, highlighting contrasting methodologies and conclusions.
You'll get something comprehensive with sources to dig deeper though as you can imagine, educators are already grappling with how this affects academic integrity.
(06:58):
Hobbyists are finding creative applications too, from building the ultimate espresso setup to astronomy research.
Imagine asking, which telescope under 500 pounds would work best for viewing planets from a light polluted urban area.
Deep research can cross reference descriptions, user reviews, and specifications to solve the puzzle.
(07:19):
The common thread.
Whether you're A-C-E-O-A student or just want to geek out over coffee game, you can now offload most of the time consuming slog to the AI and focus your own attention where it actually matters.
So what about the strengths and limitations? Let's break down the superpowers and the slipups of deep research so you know where to lean in and where to ask.
(07:41):
Are you sure about that? Deep research is phenomenal when a question demands synthesizing information scattered across the web.
It'll wrangle perspectives from journals, case studies and forums into one coherent source backed report, perfect for multidisciplinary tasks, or when the answer isn't neatly contained on a single website.
(08:02):
Its thoroughness is impressive.
It typically delivers more than you ask for helping you surface, not just direct answers, but also tangential factors or emerging risks you hadn't considered.
The inline citations and summary tables build credibility with anyone reviewing the work.
It's been benchmarked to outperform entry level professionals in quality across multiple fields.
(08:24):
Its ability to run calculations and revise its approach when evidence doesn't align.
Represents a significant advance in AI research capabilities, but there are definitely some gotchas to watch out for first.
It still hallucinates though less often than standard chat GPT.
Without guidance, it struggles to properly weigh source authority, sometimes quote in random blogs alongside government statistics.
(08:50):
If there's conflicting evidence, it may not highlight this prominently unless specifically prompted deep research is only as current as its sources.
As an example, I always use deep research as part of my workflow to prep for the show and hand on heart.
It's genuinely impressive and saves me hours.
But interestingly, when researching a recent episode, it actually pulled in some out of date information about chat GPTs free tier features.
(09:17):
It's ironic that this misinformation was delivered by chat GPT itself, so you should treat time sensitive information with healthy skepticism.
The Golden Rule, every deep research report should be treated as a first draft, not gospel.
It'll amplify your research capabilities, but only if you maintain your critical thinking skills.
(09:37):
And as I said earlier, wait times vary considerably.
Sometimes five minutes, sometimes half an hour.
On a slightly related note, for any devs listening, if you're planning to automate research through the open AI API prepare for a serious game of patients.
Trust me, I've been there code in my own little research assistant to automate this stuff only to impatiently.
(09:58):
Watch it, sit in a virtual waiting room for an hour or more.
Why does it take so long? Because deep research is in high demand, and it's a complex resource intensive process, but let's not lose perspective.
It's still lightning fast compared to the human equivalent in terms of access.
As of summer 2025, deep research has rolled out to most chat GPT users.
(10:21):
Free tier included.
They get up to five light research tasks a month, and then it's 10 to 125 runs per month on plus team enterprise and pro plans, depending on your account and when you run out, it'll switch to a lightweight version, which is a bit faster and less thorough time for some practical tips.
If you wanna make deep research sing.
(10:43):
And not just mumble a Wikipedia heavy summary.
Here's how you can get the best bang for your query.
First up, context is king.
Be specific in your queries.
Don't just say, tell me about EVs.
Instead, try something like, give me a comparison of the top five electric vehicle models by price and reliability in the UK using data from the past 18 months, and prioritize peer reviewed or reports.
(11:09):
The more details you provide about what you need, the better guide the sources too.
If you want high quality information, explicitly tell it to prioritize academic, regulatory or industry reports, or to avoid blogs and PR materials.
If you need breadth over depth, say so.
If you value authority over popularity, make that clear for balanced viewpoints.
(11:31):
Request analysis of where sources disagree or ask for a summary table of pros and cons.
Adding a prompt like.
Highlight any conflicting evidence is particularly useful for controversial topics to ensure current information specified date ranges.
For instance, use sources published after January, 2024, or focus on statistics as of July, 2025.
(11:56):
Without this guidance, you might get outdated information without warning.
Always double check citations if anything seems suspicious.
An unlikely statistic or sweeping claim.
Click the link or conduct a quick verification search.
The best users approach reports with a critical eye.
My go-to for fact checking is perplexity.
Leverage attachments.
(12:16):
As part of your research, you can upload files, spreadsheets, or background documents to give chat GPT additional context and receive more tailored personalized outputs for better organization.
Specify your preferred format, ask for bullet points, summary tables, or an executive summary at the beginning.
(12:37):
And remember that patience is essential.
Deep research isn't instant since it runs.
As a background job, plan your queries strategically.
Start it before a break, and you'll likely have a comprehensive report waiting when you return.
Bottom line.
Approach deep research like a research supervisor managing a talented, but sometimes over ego Intern.
(12:58):
Provide clear direction, ask for clarification when needed, and always review the work critically.
With practice, you'll hone your very own playbook to crafting effective research queries.
I've made deep research my secret weapon at work.
Whether I'm diving down RD rabbit holes, tackling tricky product development puzzles, or trying to figure out the best tool for the job, it keeps me grounded.
(13:22):
It helps me stick to proven best practices, so I'm not reinventing the wheel.
Or accidentally build in technical debt for the future.
And honestly, it's even changed how I shop before I drop serious cash on anything, be it electronics or a new mountain bike for my son.
I do my homework.
Deep research has saved me from more than a few buyers remorse moments.
(13:44):
Now, before we wrap up, let's zoom out a bit because chat GPTs deep research is not the only game in town.
First hope, there's perplexity ai, which has made quite an impression in the research space.
One of perplexity main strengths is transparency.
It delivers lightning fast, highly cited answers with every claim linked directly in the text, making it extremely easy to verify information.
(14:08):
What really sets it apart is its recency pulling from real-time web data with helpful filters for source types and academic content.
Perplexity delivers current information with minimal hallucinations, making it ideal for academic research.
Journalism or any fact verification scenarios.
It's combination of an intuitive app design synthesized results, plus its real-time approach made Google search look outdated for the first time in decades.
(14:37):
Perplexity free version includes three deep research queries daily.
You can unlock 500 per day by signing up to their pro plan for $20 per month.
The drawback perplexity can feel clinical.
Unlike chat GT's, narrative rich explorations, perplexity often stays at surface level unless specifically prompted to dig deeper.
(14:58):
Next up is Gemini, Google DeepMind's, AI Challenger.
That's deeply integrated into the Google ecosystem.
Its standout feature is the interactive research workflow.
You approve a multi-step plan before it pulls information, builds reports, and provides clickable links that seamlessly flow into Google's suite of apps.
(15:19):
And of course, Gemini benefits from access to current data through Google search, allowing it to pull real time information directly from the web.
This integration provides Gemini with up-to-date facts and figures that can be crucial for time-sensitive research topics.
Their approach gives you more control upfront, preventing research tangents before they happen.
(15:40):
While Gemini works relatively quickly, typically five to 15 minutes, and handles large documents, well, it trades depth for brevity when compared to chat GPTs.
Deep research.
Gemini's reports are more concise, but often lack nuance.
With less rigorous citation of individual claims.
Gemini Deep Research is available to all users for free, though limited to a few sessions per month.
(16:04):
For more queries, you'll need to upgrade to Google AI Pro at $19 and 99 cents per month.
This subscription also unlocks deep research on Google's more powerful, large language model.
Gemini 2.5
Pro, which includes several advanced capabilities that significantly beefs up the research experience.
(16:24):
Superior analytical reasoning, massive context, window audio overviews, and more.
For those already working in Google's ecosystem, Gemini offers unmatched integration for collaborative research.
However, if you need comprehensive depth or meticulous sourcing chat, GPT and perplexity still have the edge.
Now let's take a look at the open source landscape.
(16:46):
Open source tools like Open Assistant LAMA Index with Haystack or Hugging Chat give you something different.
Real control and transparency.
These tools let you build research systems that connect to your own data sources and customize exactly how information gets retrieved and presented.
The privacy benefits are substantial with your data remaining exactly where you want it.
(17:10):
While community development ensures these tools keep evolving, what's particularly interesting is how these tools, power retrieval, augmented generation systems.
Or RAG for short.
Unlike traditional AI that just draws from training data rag actively fetches information from connected sources as needed.
This makes them perfect for creating specialized knowledge assistance that work with your organization's specific information.
(17:35):
On the Plus side, you get full visibility into how things work, the ability to connect to internal systems, and no ongoing subscription fees.
These advantages make open source options particularly valuable for industries with strict regulations or teams working with sensitive information.
The trade off is complexity.
You'll need technical expertise to manage the models, servers, and data pipelines.
(18:00):
And while quality is improving rapidly, most open source options still haven't quite caught up to the capabilities of chat GPT or Gemini.
So open source is ideal if you value privacy.
Need highly customized workflows or want to build systems that integrate seamlessly with your organization's proprietary data.
(18:20):
It may also save costs when scaling, though, remember that you're trading subscription fees for setup and maintenance expenses of self-hosting, which doesn't always result in actual savings. 178 00:18:33,53.3679497 --> 00:18:35,273.3679497 So here's the big takeaway chat. 179 00:18:35,273.3679497 --> 00:18:37,853.3679497 GPTs Deep research is a real leap forward. 180 00:18:38,243.3679497 --> 00:18:40,463.3679497 It's put a research assistant in your pocket. 181 00:18:40,868.3679497 --> 00:18:45,428.3679497 A diligent if slightly overzealous aid, who will dig for hours if you ask. 182 00:18:45,788.3679497 --> 00:18:52,328.3679497 Delivering a report that's comprehensive referenced and easy to sift through, but don't be lulled into thinking it's perfect. 183 00:18:52,718.3679497 --> 00:19:01,358.3679497 It can stumble on stale data, sometimes mixes up or overstates facts and won't always flag on uncertainty or poor sources unless you ask. 184 00:19:01,898.3679497 --> 00:19:07,688.3679497 It's a first draft generator, not a silver bullet for anyone serious about saving time on big questions. 185 00:19:08,33.3679497 --> 00:19:12,773.3679497 Whether you're leading a business, learning something new or just want the best espresso machine. 186 00:19:13,103.3679497 --> 00:19:17,633.3679497 Combining deep researcher speed with your own brain's critical sense is the winning combo. 187 00:19:18,203.3679497 --> 00:19:24,413.3679497 Thanks for tuning in and remember, responsible AI use means never completely outsourcing your judgment. 188 00:19:24,773.3679497 --> 00:19:29,933.3679497 There's loads more to dig into on this front, so let me know your thoughts and stay tuned for more breakdowns coming soon. 189 00:19:31,298.3679497 --> 00:19:38,858.3679497 If you've got your own deep research story, good, bad, or downright weird, share it with a message to hello@theaibreakdown.com. 190 00:19:39,458.3679497 --> 00:19:43,358.3679497 If you found this useful, subscribe for more honest AI deep dives. 191 00:19:43,778.3679497 --> 00:19:48,458.3679497 Leave a review wherever you listen and send your must tackle research questions my way.
Hey everyone, and welcome back to the AI breakdown, your deep dive into how AI is shaping work life and everything in between.
(00:07):
Today's episode is all about digging under the hood of a genuinely powerful feature chat GT's deep research.
If you've ever felt lost trolling, endless browser tabs, or you wanna know if there's finally an AI that can actually do proper research, this one's for you.
And yes, I speak from experience.
I use deep research regularly for the show's preparation.
(00:29):
Product development, competitor analysis, and before buying expensive things like booking a holiday or my latest pair of running shoes, is it handy? Absolutely.
Does it get everything right? Well, not quite, but more on that later.
Here's what to expect in this episode.
We'll unpack what deep research is and why it's such a leap forward.
(00:50):
Dig into how it works.
Look at real business and personal use cases.
Compare it head to head with old school research Spotlight where it shines.
Give practical tips, and for the eye explorers out there, do a quick scan of alternatives.
So let's get stuck in and start with the basics.
(01:13):
What is this thing and how is it different from just Googling or flinging a question at your favorite chatbot? In a nutshell, deep research is chat GBTs way of acting less like a friendly robot that guesses answers and more like a virtual research analyst.
It's powered by an advanced O three large language model that actually browses the web reads articles, scans, PDFs, or even images, and then builds a plan to explore and answer your question in a serious step-by-step way.
(01:43):
So not just tell me about X, but I need you to troll dozens of sources, pull together the relevant bits, and give me a properly structured source cited report.
Deep research isn't for quick fun facts.
It's for genuinely big sticky questions.
Things like give me a competitive analysis of the smart speaker market citing up to date industry data, or what's the latest research on AI and education with recent case studies in the uk.
(02:11):
It's built for working professionals, think analysts, consultants, scientists, lawyers, students, but also anyone facing a question where the answer isn't just sitting at the top of Google's results.
Here's how it works.
You select deep research mode inside chat, GPTs chat box, type in your question, and you can even upload files or spreadsheets for context, sometimes chat GPT might prompt you with questions for extra details like the timeframe you care about.
(02:39):
Or specifics you wanted to focus on.
That context helps it build its search plan.
Once you kick it off, you'll see a sidebar in chat, GPT, tracking what it's doing, which sites it's searching, what docs it's reading, and it's evolving thought process.
This isn't instant, we're talking five, 10, even up to 30 minutes, depending on how gnarly the question is.
(03:02):
But you can walk away and come back to a detailed sectioned report complete with clickable sources.
Its methodology is revolutionary.
Unlike standard search engines that scrape surface level information, deep research conducts a thorough investigation sometimes through hundreds of sources.
It employs a structured, methodical approach, planning the research strategy, casting a wide net initially, then progressively narrowing its focus based on findings.
(03:31):
When initial results aren't comprehensive enough, it adapts its search parameters and dives deeper into specialized sources.
It can even execute Python code to analyze statistics or visualize data comparisons on the fly.
This strategic adaptive research process mirrors what you'd expect from a highly skilled human research assistant, but at machine speed, what you get back absolutely trumps a quick search.
(03:57):
The output is a full structured report.
We're talking thousands of words with easy to follow sections, bullet pointed, key facts, summary tables, and every claim or stat footnoted to the source, reviewers and professionals who have tried it from lawyers to engineers to analysts.
See, it's often better than what they'd get from a junior team member.
(04:19):
It's not unusual to get a report that would take a person a full day or longer to compile, where you'd usually end up lost in 20 browser tabs and four barely distinguishable PDFs.
Deep research hands you the whole thing neatly packaged With clickable citations, you shift from being the person doing the grunt work of find, read, summarize, repeat to being the supervisor.
(04:42):
Define the big question.
Give guidance and then use your time reviewing and judging instead of trolling.
It's great for market intel.
Imagine you're doing a market analysis for a new startup entering the e-bike market.
Rather than clicking 30 top 10 E-bike lists and making an Excel disaster, you can ask deep research.
Who are the current leaders in the e-bike market in the uk and what's their market share and what new technologies are emerging? What you get is a multi-source report that cross references, industry reports, news, and even niche cycling forums, all with links to chase up if you care for a deep dive in business.
(05:20):
Deep research also excels at competitor analysis if you need to understand the landscape.
Then instead of spending days building a briefing deck, you can ask for an UpToDate competitive briefing that scans news outlets, industry white papers and forums.
Delivering a comprehensive summary with links for each key point.
It's a serious tool for due diligence too.
(05:42):
In one experiment, an antitrust lawyer prompted deep research for an 8,000 word legal analysis and was genuinely impressed with the results.
He rated it comparable to work from an entry level attorney, something that would typically require 15 to 20 billable hours and said he'd use this model professionally.
In another real case, an architect asked deep research for a detailed building code checklist for a large educational building.
(06:09):
The system consulted 21 different sources, including state regulations, international code, PDFs, and accessibility standards, producing a comprehensive 15,000 word regulation site report.
In just 28 minutes, she estimated the same task would've consumed an entire working day, but it's not just for business folks.
(06:31):
For students and researchers, deep research transforms.
Literature reviews need a multi-sided review For a university essay, you might prompt it with something like Compare recent studies on climate adaptation strategies in coastal cities, highlighting contrasting methodologies and conclusions.
You'll get something comprehensive with sources to dig deeper though as you can imagine, educators are already grappling with how this affects academic integrity.
(06:58):
Hobbyists are finding creative applications too, from building the ultimate espresso setup to astronomy research.
Imagine asking, which telescope under 500 pounds would work best for viewing planets from a light polluted urban area.
Deep research can cross reference descriptions, user reviews, and specifications to solve the puzzle.
(07:19):
The common thread.
Whether you're A-C-E-O-A student or just want to geek out over coffee game, you can now offload most of the time consuming slog to the AI and focus your own attention where it actually matters.
So what about the strengths and limitations? Let's break down the superpowers and the slipups of deep research so you know where to lean in and where to ask.
(07:41):
Are you sure about that? Deep research is phenomenal when a question demands synthesizing information scattered across the web.
It'll wrangle perspectives from journals, case studies and forums into one coherent source backed report, perfect for multidisciplinary tasks, or when the answer isn't neatly contained on a single website.
(08:02):
Its thoroughness is impressive.
It typically delivers more than you ask for helping you surface, not just direct answers, but also tangential factors or emerging risks you hadn't considered.
The inline citations and summary tables build credibility with anyone reviewing the work.
It's been benchmarked to outperform entry level professionals in quality across multiple fields.
(08:24):
Its ability to run calculations and revise its approach when evidence doesn't align.
Represents a significant advance in AI research capabilities, but there are definitely some gotchas to watch out for first.
It still hallucinates though less often than standard chat GPT.
Without guidance, it struggles to properly weigh source authority, sometimes quote in random blogs alongside government statistics.
(08:50):
If there's conflicting evidence, it may not highlight this prominently unless specifically prompted deep research is only as current as its sources.
As an example, I always use deep research as part of my workflow to prep for the show and hand on heart.
It's genuinely impressive and saves me hours.
But interestingly, when researching a recent episode, it actually pulled in some out of date information about chat GPTs free tier features.
(09:17):
It's ironic that this misinformation was delivered by chat GPT itself, so you should treat time sensitive information with healthy skepticism.
The Golden Rule, every deep research report should be treated as a first draft, not gospel.
It'll amplify your research capabilities, but only if you maintain your critical thinking skills.
(09:37):
And as I said earlier, wait times vary considerably.
Sometimes five minutes, sometimes half an hour.
On a slightly related note, for any devs listening, if you're planning to automate research through the open AI API prepare for a serious game of patients.
Trust me, I've been there code in my own little research assistant to automate this stuff only to impatiently.
(09:58):
Watch it, sit in a virtual waiting room for an hour or more.
Why does it take so long? Because deep research is in high demand, and it's a complex resource intensive process, but let's not lose perspective.
It's still lightning fast compared to the human equivalent in terms of access.
As of summer 2025, deep research has rolled out to most chat GPT users.
(10:21):
Free tier included.
They get up to five light research tasks a month, and then it's 10 to 125 runs per month on plus team enterprise and pro plans, depending on your account and when you run out, it'll switch to a lightweight version, which is a bit faster and less thorough time for some practical tips.
If you wanna make deep research sing.
(10:43):
And not just mumble a Wikipedia heavy summary.
Here's how you can get the best bang for your query.
First up, context is king.
Be specific in your queries.
Don't just say, tell me about EVs.
Instead, try something like, give me a comparison of the top five electric vehicle models by price and reliability in the UK using data from the past 18 months, and prioritize peer reviewed or reports.
(11:09):
The more details you provide about what you need, the better guide the sources too.
If you want high quality information, explicitly tell it to prioritize academic, regulatory or industry reports, or to avoid blogs and PR materials.
If you need breadth over depth, say so.
If you value authority over popularity, make that clear for balanced viewpoints.
(11:31):
Request analysis of where sources disagree or ask for a summary table of pros and cons.
Adding a prompt like.
Highlight any conflicting evidence is particularly useful for controversial topics to ensure current information specified date ranges.
For instance, use sources published after January, 2024, or focus on statistics as of July, 2025.
(11:56):
Without this guidance, you might get outdated information without warning.
Always double check citations if anything seems suspicious.
An unlikely statistic or sweeping claim.
Click the link or conduct a quick verification search.
The best users approach reports with a critical eye.
My go-to for fact checking is perplexity.
Leverage attachments.
(12:16):
As part of your research, you can upload files, spreadsheets, or background documents to give chat GPT additional context and receive more tailored personalized outputs for better organization.
Specify your preferred format, ask for bullet points, summary tables, or an executive summary at the beginning.
(12:37):
And remember that patience is essential.
Deep research isn't instant since it runs.
As a background job, plan your queries strategically.
Start it before a break, and you'll likely have a comprehensive report waiting when you return.
Bottom line.
Approach deep research like a research supervisor managing a talented, but sometimes over ego Intern.
(12:58):
Provide clear direction, ask for clarification when needed, and always review the work critically.
With practice, you'll hone your very own playbook to crafting effective research queries.
I've made deep research my secret weapon at work.
Whether I'm diving down RD rabbit holes, tackling tricky product development puzzles, or trying to figure out the best tool for the job, it keeps me grounded.
(13:22):
It helps me stick to proven best practices, so I'm not reinventing the wheel.
Or accidentally build in technical debt for the future.
And honestly, it's even changed how I shop before I drop serious cash on anything, be it electronics or a new mountain bike for my son.
I do my homework.
Deep research has saved me from more than a few buyers remorse moments.
(13:44):
Now, before we wrap up, let's zoom out a bit because chat GPTs deep research is not the only game in town.
First hope, there's perplexity ai, which has made quite an impression in the research space.
One of perplexity main strengths is transparency.
It delivers lightning fast, highly cited answers with every claim linked directly in the text, making it extremely easy to verify information.
(14:08):
What really sets it apart is its recency pulling from real-time web data with helpful filters for source types and academic content.
Perplexity delivers current information with minimal hallucinations, making it ideal for academic research.
Journalism or any fact verification scenarios.
It's combination of an intuitive app design synthesized results, plus its real-time approach made Google search look outdated for the first time in decades.
(14:37):
Perplexity free version includes three deep research queries daily.
You can unlock 500 per day by signing up to their pro plan for $20 per month.
The drawback perplexity can feel clinical.
Unlike chat GT's, narrative rich explorations, perplexity often stays at surface level unless specifically prompted to dig deeper.
(14:58):
Next up is Gemini, Google DeepMind's, AI Challenger.
That's deeply integrated into the Google ecosystem.
Its standout feature is the interactive research workflow.
You approve a multi-step plan before it pulls information, builds reports, and provides clickable links that seamlessly flow into Google's suite of apps.
(15:19):
And of course, Gemini benefits from access to current data through Google search, allowing it to pull real time information directly from the web.
This integration provides Gemini with up-to-date facts and figures that can be crucial for time-sensitive research topics.
Their approach gives you more control upfront, preventing research tangents before they happen.
(15:40):
While Gemini works relatively quickly, typically five to 15 minutes, and handles large documents, well, it trades depth for brevity when compared to chat GPTs.
Deep research.
Gemini's reports are more concise, but often lack nuance.
With less rigorous citation of individual claims.
Gemini Deep Research is available to all users for free, though limited to a few sessions per month.
(16:04):
For more queries, you'll need to upgrade to Google AI Pro at $19 and 99 cents per month.
This subscription also unlocks deep research on Google's more powerful, large language model.
Gemini 2.5
Pro, which includes several advanced capabilities that significantly beefs up the research experience.
(16:24):
Superior analytical reasoning, massive context, window audio overviews, and more.
For those already working in Google's ecosystem, Gemini offers unmatched integration for collaborative research.
However, if you need comprehensive depth or meticulous sourcing chat, GPT and perplexity still have the edge.
Now let's take a look at the open source landscape.
(16:46):
Open source tools like Open Assistant LAMA Index with Haystack or Hugging Chat give you something different.
Real control and transparency.
These tools let you build research systems that connect to your own data sources and customize exactly how information gets retrieved and presented.
The privacy benefits are substantial with your data remaining exactly where you want it.
(17:10):
While community development ensures these tools keep evolving, what's particularly interesting is how these tools, power retrieval, augmented generation systems.
Or RAG for short.
Unlike traditional AI that just draws from training data rag actively fetches information from connected sources as needed.
This makes them perfect for creating specialized knowledge assistance that work with your organization's specific information.
(17:35):
On the Plus side, you get full visibility into how things work, the ability to connect to internal systems, and no ongoing subscription fees.
These advantages make open source options particularly valuable for industries with strict regulations or teams working with sensitive information.
The trade off is complexity.
You'll need technical expertise to manage the models, servers, and data pipelines.
(18:00):
And while quality is improving rapidly, most open source options still haven't quite caught up to the capabilities of chat GPT or Gemini.
So open source is ideal if you value privacy.
Need highly customized workflows or want to build systems that integrate seamlessly with your organization's proprietary data.
(18:20):
It may also save costs when scaling, though, remember that you're trading subscription fees for setup and maintenance expenses of self-hosting, which doesn't always result in actual savings. 178 00:18:33,53.3679497 --> 00:18:35,273.3679497 So here's the big takeaway chat. 179 00:18:35,273.3679497 --> 00:18:37,853.3679497 GPTs Deep research is a real leap forward. 180 00:18:38,243.3679497 --> 00:18:40,463.3679497 It's put a research assistant in your pocket. 181 00:18:40,868.3679497 --> 00:18:45,428.3679497 A diligent if slightly overzealous aid, who will dig for hours if you ask. 182 00:18:45,788.3679497 --> 00:18:52,328.3679497 Delivering a report that's comprehensive referenced and easy to sift through, but don't be lulled into thinking it's perfect. 183 00:18:52,718.3679497 --> 00:19:01,358.3679497 It can stumble on stale data, sometimes mixes up or overstates facts and won't always flag on uncertainty or poor sources unless you ask. 184 00:19:01,898.3679497 --> 00:19:07,688.3679497 It's a first draft generator, not a silver bullet for anyone serious about saving time on big questions. 185 00:19:08,33.3679497 --> 00:19:12,773.3679497 Whether you're leading a business, learning something new or just want the best espresso machine. 186 00:19:13,103.3679497 --> 00:19:17,633.3679497 Combining deep researcher speed with your own brain's critical sense is the winning combo. 187 00:19:18,203.3679497 --> 00:19:24,413.3679497 Thanks for tuning in and remember, responsible AI use means never completely outsourcing your judgment. 188 00:19:24,773.3679497 --> 00:19:29,933.3679497 There's loads more to dig into on this front, so let me know your thoughts and stay tuned for more breakdowns coming soon. 189 00:19:31,298.3679497 --> 00:19:38,858.3679497 If you've got your own deep research story, good, bad, or downright weird, share it with a message to hello@theaibreakdown.com. 190 00:19:39,458.3679497 --> 00:19:43,358.3679497 If you found this useful, subscribe for more honest AI deep dives. 191 00:19:43,778.3679497 --> 00:19:48,458.3679497 Leave a review wherever you listen and send your must tackle research questions my way.
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
The Joe Rogan Experience
The official podcast of comedian Joe Rogan.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Special Summer Offer: Exclusively on Apple Podcasts, try our Dateline Premium subscription completely free for one month! With Dateline Premium, you get every episode ad-free plus exclusive bonus content.