Presentation 3: AI Deep Research - The Tools
(first given 27th June 2025)
All AI chat tools will develop towards the same endpoint of being able to handle and deliver text, image, video, sound and more. It is also likely that features called 'Deep Research mode' or 'Thinking mode' will be hidden behind a very universal front end, with the AI model deciding which mode is needed according to the query. For the moment though they are explicit in the interface.
[1a] Google's Gemini and NotebookLM etc
Google has sometimes been underestimated in the AI race but it has an increasing range of powerful tools. For this presentation the Gemini tool with its Deep Research function is the most relevant. NotebookLM also has powerful capabilities however - you could ask it for a Podcast on the research results and it would be very good.
Google's AI domains are thriving.
— Similarweb (@Similarweb) June 26, 2025
Month-over-month traffic growth in May:@GoogleDeepMind: 4,258,300 → 11,189,832 (+162.78%)@GoogleLabs: 9,796,459 → 19,792,943 (+102.04%)@GoogleAI: 1,349,027 → 2,042,165 (+51.38%)@NotebookLM: 48,340,200 → 64,906,295 (+34.27%)@GeminiApp:… pic.twitter.com/fNCFNi1JTS
[1b] Gemini asked to explain its Deep Research
This is using a free version at gemini.google.com with a question about its Deep Research capability, which it treats as deeply as any other question.
[1c] Snippet from thinking during Deep Research
Advanced AI Capabilities. A core differentiator lies in its advanced AI capabilities. Deep Research utilizes 'thinking models' (like Gemini 2.5 Pro and Flash) that break down complex problems into smaller, manageable steps, mimicking human-like thought processes to enhance reasoning and multi-step planning. It maintains continuity through a 1 million token context window complemented by a Retrieval-Augmented Generation (RAG) setup, allowing it to 'remember' everything learned during a session. This enables it to handle complex queries, analyze vast amounts of information, and even generate or explain code, demonstrating a significant leap in information synthesis and understanding.
[1d] The final Google Gemini report
Extract: Powered by the advanced Gemini 2.5 model, this capability transcends the reactive, text-centric conversational paradigms of earlier Google AI chat tools, such as LaMDA and Bard. Deep Research operates as a proactive, agentic, and multimodal research assistant, capable of autonomously executing complex research tasks from planning to comprehensive report generation. This advancement is underpinned by sophisticated architectural designs, including the Mixture-of-Experts (MoE) framework and vastly expanded context windows, coupled with enhanced reasoning capabilities that involve iterative self-critique.
Gemini took a few minutes to produce the full report comparing its Deep Research to its older versions. This is the
report in pdf form:
Gemini's pdf report about Deep Research
[2a] ChatGPT is the market leader measured by public use
Worldwide iPhone App Store downloads over the last 28 days:
— Similarweb (@Similarweb) June 24, 2025
ChatGPT: 29,551,174
TikTok + Facebook + Instagram + X: 32,859,208 pic.twitter.com/v4tIaC65ou
[2b] ChatGPT from OpenAI asked to explain its Deep Research
This is using the USD 20 per month Pro version at chatgpt.com with a question about its Deep Research capability compared to previous versions.
[2c] ChatGPT starts its Deep Research about itself
As with Gemini, the question is approached in the same way as a question about any other subject.
[2d] The final ChatGPT report
ChatGPT took just 6 minutes to produce a report comparing its Deep Research to its older versions. Below is a link to the
report in pdf form. Extract: The model reads and analyzes online texts (including articles, PDFs, and images) and pivots its search strategy based on what it finds, acting as an agent that “finds, analyzes, and synthesizes hundreds of online sources” into its answer. This allows it to handle current events and niche queries that go beyond its training data.
ChatGPT's pdf report about Deep Research
[3] Grok (from X)
Though Grok does have a DeepSearch mode it is not really competitive with Gemini or ChatGPT in this area yet. It does benefit from having access to the X (Twitter) feed which provides a vast amount of real time News. A new version of Grok is due shortly.
[4] Perplexity
Perplexity uses a range of AI models to power its system. It built its reputation on accuracy, with good references being provided for all listed information. It does a good job of providing multi-stage analysis and then reports on any question - and in many cases may be using OpenAI or Google provided tools behind its own user interface.
As seen below, this can include writing code to produce charts that illustrate the report.
[5] Other AI tools for Scientific and or Business research
These have not been used by author so not recommendations:
https://elicit.com/ Automate time-consuming research tasks like summarizing research papers, extracting data, and synthesizing your findings.
https://scite.ai/ Supercharge your research with AI and search powered by the world’s scholarly content - both Open Access and paywalled.
https://consensus.app/ Consensus is the AI-powered academic search engine. Search & analyze200M+ peerreviewed research papers.
https://kompas.ai/ Deep research business reports
Link to Back to main presentation.