Using AI research tools to understand complex issues, fast
Table of contents
What is Research mode?
When to use Research mode
Potential use cases
How to use Research mode
Activating it
Writing a research prompt
Our example prompt
Clarifying your request
What GPT research tools are available?
Our verdict
Research tool usage limits
How Research mode works
Other ways to research with AI
Next steps and AI workflows
Appendix: six more examples
The rundown: for deeper answers to complex questions, most AI chatbots now offer a specialist research tool. We tested five leading options, and we think Google Gemini is currently the best – and it’s free to try!
What is ‘Research’ mode?
Switching your AI chatbot into ‘Research’ mode changes things up. Instead of replying instantly with an intelligent guess based on its training data, the AI scours the web for relevant, up-to-date information on your topic of choice.
Unlike a regular AI chat, this research process takes 5–30 minutes to complete. Your AI creates specialized 'subagents' to research different angles of your topic, before assembling a detailed response - which can run up to 8,000 words.
Research is currently available in ChatGPT (‘Deep Research’), Claude (‘Research’), Google Gemini (‘Deep Research’), Grok(‘DeepSearch’) and Perplexity(‘Research’) to name a few.
When to use ‘Research’ mode
By querying the web rather than relying on its training data, AI can deliver an answer with up-to-date knowledge, real citations and links, which makes it easier to verify.
It’s great for:
Learning more about a complex concept
Research you will ultimately need to verify against footnotes/sources – your research report will contain citations of third-party content, whereas traditional AI answers often contain confident assertions that are not backed up and difficult to fact-check
Getting to grips with multiple sides of an issue, or understanding different stakeholders
Research around relatively obscure topics that the agent is unlikely to be that familiar with, such as particular person or niche
Research with a cutting-edge vibe (such as recent developments in science) or that needs to be accurate as of today (for instance, comparing the current prices of things)
Exploratory, opening research into possibilities, opportunities and options
Topics where a ‘good enough’ answer will do fine
It’s particularly useful when you want to compare sources, synthesise insights, or gather perspectives from across the web. Things like option appraisals, investment research, medical treatment options, technology evaluation, legal precedent analysis, literature reviews, college selection, policy insight, real estate analysis, competitive intelligence and vendor evaluation are all areas where this might come in handy.
Potential use cases
Putting together a briefing on a company or public figure you’re about to meet
Getting a good overview of recent developments within an academic field
Understanding controversies or synthesising views of a range of stakeholders
Collecting information on possible purchases, comparing product information
How to use Research mode
Switching into Research mode is easy – in all five models we tested, Research can be activated using a control in the ‘tools’ area beneath the chat text area.
Here’s what that looks like in Grok and ChatGPT, for example:
Note, it doesn’t matter what ‘model’ you have otherwise selected. For example, if you’re working with Claude, it doesn’t matter if the conversation is with ‘Sonnet’ or ‘Opus’ – once you start the Research process, a custom model takes over.
Crafting your Deep Research prompt
As with all prompting, it’s worth taking the time to describe exactly what you’re looking for.
Introduce who you are and the ultimate purpose of your request
Pose the research question in straightforward language – what do you want to understand?
Describe the format you’d like the response to take – a formal written through report, a creative essay, a structured table?
Define your end goal: what content and analysis must the final product include? For instance, “The final report MUST provide a costed-out analysis of at least three options.” This helps the AI shape the research process and know when to stop.
Identify preferred sources: if you’d like the research AI to target academic sources, think tank reports, or genuine user reviews. Likewise, identify any types of source you’d rather the AI ignored.
If you want the research to take a particular ‘path’ – for instance, a broad ‘scan’ of literature and then a deep dive
Here’s the example prompt we used in our test:
“I'm writing my master's thesis on the cultural impact of social media on political movements in Latin America between 2015-2025.
Research how platforms like WhatsApp, Facebook, Instagram, and TikTok have influenced political organizing, information dissemination, and democratic participation across countries like Brazil, Mexico, Argentina, and Colombia.
I need:
academic sources
case studies of specific political campaigns or movements
analysis of misinformation campaigns
the role of algorithm-driven content
how government responses have evolved.
Please include:
recent scholarly articles
reports from organizations like Freedom House or Amnesty International
perspectives from Latin American digital rights advocates and political scientists.”
Clarifying your deep research request
Three of the five tools we tried present an opportunity to hone your request before the research begins.
ChatGPT and Claude will ask a few additional questions
Gemini presents an editable research plan you can modify
You can take a look at the kind of clarifying questions the AIs asked us here:
-
[The following test is editable]
(1) Research the general cultural impact of social media on political movements in Latin America, focusing on the period between 2015 and 2025.
(2) Investigate how specific platforms, including WhatsApp, Facebook, Instagram, and TikTok, have influenced political organizing, information dissemination, and democratic participation in Brazil, Mexico, Argentina, and Colombia.
(3) Identify and analyze case studies of specific political campaigns or movements in these countries where social media played a significant role.
(4) Examine the prevalence and impact of misinformation campaigns and the role of algorithm-driven content in shaping political discourse and movements in Latin America.
(5) Research how governments in Brazil, Mexico, Argentina, and Colombia have responded to and regulated social media's involvement in political movements.
(6) Find recent scholarly articles and academic research papers on the cultural impact of social media on political movements in Latin America.
(7) Locate reports and analyses from organizations such as Freedom House and Amnesty International concerning digital rights, political movements, and social media in Latin America.
(8) Seek out perspectives and analyses from Latin American digital rights advocates and political scientists regarding the influence of social media on political landscapes in the region.
-
Do you want equal depth for all four countries (Brazil, Mexico, Argentina, Colombia), or is there a primary focus?
Are you interested in grassroots movements, electoral campaigns, or both?
Should the research prioritise academic peer-reviewed sources over think tank and NGO reports, or treat them equally?
Are you seeking content in English only, or should I include Spanish and Portuguese sources as well?
-
I'd like to clarify a few things to ensure I gather the most relevant and useful sources for your thesis:
Methodological focus: Are you primarily interested in qualitative case studies of specific movements/campaigns, quantitative analysis of social media metrics and engagement, or a mixed-methods approach combining both?
Movement scope: Should I focus on particular types of political movements (e.g., electoral campaigns, protest movements, civil society activism, populist movements) or cast a wide net across all forms of political organizing?
Thesis framework: Are you analyzing this through a particular theoretical lens (e.g., digital divide theory, networked society theory, democratization theory) that should guide the type of academic sources I prioritize?
Tactic: getting the AI to think about what it’s learned, mid-research
Much of the research the AI conducts will happen sequentially, following a research plan the AI has invented. With your prompt, you can steer this process in advance. For instance, after providing information about your vacation preferences, you might add:
1) Explore and research possible vacation destinations. Identify at least five candidates.
2) Reason and think about the most promising candidate.
3) Identify how I would travel to that destination from my location
4) Find quality hotels matching my criteria at that destination
5) Identify things to do in the neighbourhood of that hotel
Your final result MUST conclude with a concrete proposal for me to follow with a recommended destination, travel plan, hotel choice and nearby activities.
This nudges the AI toward specific outputs, instead of a meandering discussion of various possibilities.
What AI Research tools are available – and which one is best?
We put ChatGPT, Claude, Google Gemini, Grok and Perplexity to the test. To get a feel of what these tools are capable of, let’s read their reports!
Read the reports
Let’s take a look at how their outputs compared.
*Claude’s high source count may be inflated — it appears to count all URLs discovered via search, not just those read or cited in the final report.Our verdict? Gemini and ChatGPT produced far more substantial reports. Claude’s output was excellently written, but rather more concise. Grok and Perplexity instead offered rather simpler research summaries with plenty of bullet points.
Gemini clearly capitalises on its access to Google’s own in-house copy of the web, while ChatGPT called upon its superior reasoning abilities to offer an impressive response.
We even asked a panel of AI judges to assess their favourite, and it was a split decision with Gemini in the lead:
“Gemini’s response best meets the brief. It provides a comprehensive, well-researched, and insightful analysis, excelling in clarity, depth, and perceptiveness. ChatGPT’s is strong and clear, but lacks Gemini’s thoroughness and nuance. Claude’s, while informative, falls short in scholarly rigor, clarity, and depth. Thus, Gemini’s essay stands out as the strongest.”
“Gemini offered the most complete, rigorous, and well-structured response. Excellent on all platforms, with good theoretical framing and a wide sweep of cases. But ChatGPT’s offered the best narrative and case studies, while Claude offered the keenest critical insight.”
“Claude’s analysis is sharp and Gemma’s organization is superb, but ChatGPT’s work is not just a report on the facts; it is a holistic and compelling analysis that provides the reader with a genuine, textured understanding of the issue. It is this combination of rigorous research, narrative force, and profound insight that places their submission at the top.”
“Claude’s response best exemplifies the kind of sophisticated, theoretically-grounded analysis expected at the graduate level. It successfully balances academic rigor with practical relevance, offering both scholarly insights and policy implications that advance our understanding of this complex phenomenon. ChatGPT’s work was highly competent, with excellent narrative flow, but Gemini’s reads more like a literature review than original analysis and lacks the interpretive sophistication of the other two.”
*Ironically, ChatGPT as judge preferred Gemini, but Gemini-as-judge preferred ChatGPT. Claude preferred itself, while Grok preferred Gemini. (They weren't told which model had generated which response.)Deep Research usage limits
Use of these tools is currently somewhat limited, as they use a lot more computing power than a regular conversation.
Luckily, the best-performing model, Gemini, is also the platform with the most generous research limits. Another reason to recommend it!
Gemini (the most generous): 10/month for free, 20/day for paid users
ChatGPT: 5 ‘light’ reports/months for free users; 10 full reports a month for paid users, plus another 15 ‘light’ reports (note: we have not yet tested the ‘lightweight’ report mode)
Claude: Research is not available for free users. 1-2 reports every 5 hours for paid users.
Grok: Not specified
Perplexity: 3 per day at regular depth (a more advanced model is available for paid users)
How does Research mode actually work?
Each tool uses a slightly different architecture — Claude, for instance, has a “lead agent” that delegates to subagents. Their engineering article explains more.
Peeking behind the curtain
Depending on the platform you’re using, you can often take a look at what the AI is thinking. For instance, Claude reasons in-depth about your needs before it begins:
Click to enlarge.
In a different test, we can see Grok thinking about individual search queries, and reflecting on the facts it discovers.
Reviewing this internal monologue can reveal how the AI approaches problems — and where it sometimes misfires.
Other ways to research with AI
Asking directly: for common, encyclopaedic knowledge –Shakespeare, or set theory – regular chat is quicker and often just as accurate.
Invite AI to use web search: increasingly, AI chatbots like Claude and ChatGPT will search the web for information when they think it’s appropriate. When using these tools conversationally, you can explicitly prompt them to do so. Try saying something like this: Use your tools to research recent mentions of this topic at these websites: example.com, nytimes.com, bbc.co.uk etc
For example, when we shared a similar prompt to our test research assignment to ChatGPT’s O3 reasoning model, it deployed many of the same techniques before answering:
Providing your own sources: Got your own data ready to go? Upload academic papers, PDFs, spreadsheets, etc, and ask for help summarising or analysing them. Or bring these documents into Google’s AI tool NotebookLM to explore in-depth.
Specialist AI research tools: beyond your general-purpose AI tools, try research-focused products such as Elicit, Scite and Sci-Space.
After the report drops: next steps
Your report can be just the start.
Chat about it! Download the document, then upload it to a new chat conversation to discuss the topic at hand.
Turn your report into another asset: for instance, you could use…
Gamma to turn the report into a presentation
NotebookLM to turn the report into a podcast
Canva to pull out key facts and turn them into social posts
Here’s an example presentation from Gamma, for example:
While Gemini was able to turn the report into a full microsite:
And NotebookLM generated a nearly hour-long podcast covering the topic in-depth:
Summary
If you’re looking for rich, up-to-date answers — with sources — Deep Research tools are a game-changer. They take longer, but can deliver far more. Gemini is the standout tool right now, with ChatGPT not far behind.
APPENDIX: 6 more examples of Research reports
-
“I need to choose a new electric vehicle for my family with a budget of $45,000-65,000. Please create a comprehensive comparison table of the top 10 EVs in this price range, including the Tesla Model Y, Ford Mustang Mach-E, Genesis GV60, BMW iX3, Volvo XC40 Recharge, and others you recommend.” Read Gemini’s report here.
-
Generate a deep and extensive research report into the potential of geothermal power as a solution to green energy generation over the next 30 years. Read Gemini’s report here.
-
Research the political landscape of the city of Omaha, Nebraska. - Focus on the present-day and recent events, except when examining trends over time. - How does the political system work, who holds power in practice? How competitive is it? - Who does this city elect and why? Who are the key individuals? Any long-term power brokers or up-and-coming political figures? I'll also want an overview of the long-term demography and economics of the area and how they're changing, and local perspectives on political issues such as health, crime, prosperity and so on. Also any nuances to neighbourhoods, areas of the city, and internal divisions in terms of their politics. Most importantly you should devote research to uncovering the unique and hot-button issues in the area, anything that's local and specific to the city itself – what are the issues that people are talking about? The briefing should be detailed, accurate and suitable for educating, say, a political journalist that is now adding this city to their beat for the first time. They will want to be able to cover the city's politics with sound knowledge of the stories that matter, the issues people care about, the policy challenges, political fault-lines and the current dramatis personae. At least one section early in the report should be a structured table of relevant public figures. Read ChatGPT’s report.
-
Research some cool, trending cities in Europe, but they should be small or mid-sized, not major capital cities like Paris. (NOTE: I am talking about 'trending' as in 'cool' or 'becoming popular for sophisticated tourists', NOT as in economically.)
Once you have found a few candidates, reason about the one you think is coolest.
Then focus on this city, and research the coolest neighbourhoods.
Then, look for the coolest hotel in that neighbourhood.
Then, look for great restaurants and things to do in that area.
Finally, research how to get to there from London, and how much it will all cost.
Your final report MUST consist of a specific recommendation of city, hotel, neighbourhood, restaurants, and price/itinerary/travel options for enjoying the trip.
Read Gemini’s report. -
“Identify as many businesses as you can that are suppliers of wedding-related services in the state of Rhode Island, and collate their contact details.” See Claude’s response here.
-
My 67-year-old father has been diagnosed with moderate aortic stenosis and his cardiologist is recommending either surgical valve replacement or transcatheter aortic valve implantation.
Research the latest clinical evidence comparing these two approaches for his age group, including success rates, recovery times, long-term outcomes, and potential complications.
We'd like to understand the criteria doctors use to choose between these options, what questions we should ask his medical team, and if there are any cutting-edge treatments or clinical trials he might be eligible for. Please include information about post-procedure lifestyle changes and prognosis. Read Gemini’s report.