By the end of this guide you will have a structured research report on a topic of your choice: a scoped question, a draft written by an AI research agent from many sources, every important claim checked against its citation, and a final version you have edited and can stand behind. The AI does the searching, reading and first draft. You do the scoping and the checking.
To produce a deep research report with AI, you choose an agent that plans sub-questions, searches in several rounds and writes with citations. This guide compares six tools from our Research and search category: hosted assistants, open source agents you run yourself, and a notebook for your own documents. It uses only what each listing states.
Key takeaways
- Scope the question tightly, run a deep research agent, then check every key claim against the cited source before you edit the draft into your own report.
- The main decision is hosted or self-run: Perplexity, ChatGPT, Elicit and Gemini Notebook are hosted, while GPT Researcher and Local Deep Research are open source and you run them yourself.
- Output quality depends on the model and sources. GPT Researcher's listing says users are responsible for fact-checking its reports.
- Treat the report as a draft. For legal, medical, financial or academic work, a qualified person must review it before it is used.
What you need before you start
- A research question. One or two sentences, with the audience and purpose of the report.
- An outline idea. The sections you expect, even if they change.
- Your own sources, if any. PDFs, spreadsheets or notes you want included.
- An account or setup. The hosted tools limit usage on their free plans. GPT Researcher needs Python 3.11 or newer plus LLM and search API keys, and Local Deep Research needs a model endpoint and SearXNG.
- Time. A first draft can arrive in minutes. Checking and editing take hours.
- Difficulty. Easy for hosted tools, harder for self-run agents.
Which kind of AI tool fits this job
Choose by where you want the work done and what you are researching.
Hosted assistants with deep research. Perplexity offers a deep research mode on paid plans with inline citations, and premium data sources such as PitchBook, Statista and S&P Capital IQ. ChatGPT lists deep research that compiles sourced reports on paid plans, with limited access on the Free plan.
Open source agents you run yourself. GPT Researcher splits a question into sub-questions, searches in parallel, reads more than 20 sources and writes a cited report, which exports to PDF, Word or Markdown. Local Deep Research searches the web, academic sources and your own documents, uses local or cloud models, and stores downloaded sources in an encrypted library.
Academic research. Elicit searches more than 138 million papers, extracts data into tables and writes Reports that draw on many data sources.
Reports from your own documents. Gemini Notebook, which Google renamed from NotebookLM, answers questions with citations back to the sources you upload, and can build Reports, Slides and other outputs from them. Its Deep Research can also find new sources.
Step 1: Scope the report
You end with a brief that tells the agent exactly what to produce.
- Write the central question and what the report will be used for.
- Name the audience, such as executives, students or a client.
- Set limits: time period, geography, industries, source types.
- List three to six sections you expect, and what you already know.
Check it worked by asking whether a stranger could tell, from your brief alone, what a good report would contain.
Step 2: Pick an agent and set it up
Match the tool to your subject, privacy needs and skills.
- For quick setup, use a hosted assistant such as Perplexity or ChatGPT.
- For academic topics, use Elicit.
- If you want control over models and data, set up GPT Researcher or Local Deep Research. GPT Researcher requires Python 3.11 or newer plus an LLM key and a search key, and Local Deep Research needs a model endpoint and SearXNG.
- Check the free plan or setup requirements against your timeline.
Check it worked by running one trivial question and getting a cited answer.
Step 3: Run the research
Hand the brief to the agent and let it work.
- Paste your question, audience, limits and expected sections.
- If the tool allows, upload your own files. GPT Researcher can research PDFs, Word files, spreadsheets and Markdown, alone or with web results.
- Choose the deep mode. GPT Researcher has a deep research mode that branches into follow-up questions, and Local Deep Research has a LangGraph agent strategy for autonomous research.
- Wait for completion. GPT Researcher's listing says a deep research run typically takes 30 to 60 seconds per query, and other tools differ.
Check it worked by receiving a report with a structure, a source list and inline citations.
Step 4: Check the citations
This is where the work is. Do not skip it.
- Open every source behind a key claim, number or recommendation.
- Confirm the source says what the report says. Fix or delete claims that it does not support.
- Look for primary sources behind secondary ones, and check dates.
- For papers, check retractions and how later work treated the finding.
Check it worked by being able to point to a passage in the source for every claim you keep.
Step 5: Fill gaps and test the reasoning
An agent can miss angles or overstate a case.
- Compare the report with your outline and list what is missing.
- Ask follow-up questions on the gaps, and ask for evidence against the main conclusion.
- For academic topics, use Elicit's table extraction to compare studies side by side, and open the papers for important data.
- Note the limits of the evidence, including sources the agent could not reach.
Check it worked by being able to state the strongest counterargument to your conclusion.
Step 6: Bring in your own material
Reports are stronger when they combine public sources with your own context.
- Upload documents, notes or data to a source-grounded tool such as Gemini Notebook. Free use allows up to 50 sources per notebook.
- Ask questions that only your documents can answer and check the citations.
- Keep confidential material out of tools whose data terms you have not read. Local Deep Research runs on your own hardware, but its listing says its egress controls are best-effort rather than an air gap.
- Combine the findings into your outline.
Check it worked by seeing which sections now rest on your own evidence.
Step 7: Edit, format and review
The final report should read as your work.
- Rewrite in your own voice and cut anything vague or repeated.
- Add an executive summary, a methods note describing the tools and dates, and a source list.
- Export to the format you need. GPT Researcher exports to PDF, Word or Markdown.
- Have a colleague or subject expert review it, and disclose AI use where your organization or publisher requires it.
Check it worked by reading it as your intended audience and finding no claim you cannot support.
Tools for this job compared
| Tool | Good fit for | Pricing | Stated limits |
|---|---|---|---|
| Perplexity | Hosted deep research with inline citations | Freemium, paid from $17/mo (checked 24 Oct 2026) | Free has limited weekly usage and basic models; plan details differ between consumer and enterprise pages |
| GPT Researcher | Cited long-form reports you host yourself | Open source | Needs Python 3.11 or newer, an LLM key and a search key; users must fact-check reports |
| ChatGPT | General assistant with deep research on paid plans | Freemium | Free plan has limited uploads, deep research and memory; check the official page for prices |
| Local Deep Research | Self-hosted research with encrypted storage | Open source | You run the app plus a model endpoint and SearXNG; egress controls are best-effort, not an air gap |
| Elicit | Academic papers, tables and reports | Freemium, paid from $49/mo (checked 24 Oct 2026) | Basic has limited usage; unused monthly usage does not roll over; figure extraction needs Scale or Enterprise |
| Gemini Notebook | Reports grounded in your own uploaded sources | Freemium | Free plan allows up to 50 sources per notebook; usage limits refresh on a cycle up to a weekly cap |
Prices and plan limits change, so confirm the current terms on each vendor's site before you subscribe.
Common mistakes to avoid
- Accepting the report as finished. Agents can misread sources, miss angles or state weak evidence confidently. Fix it by checking the citations and editing the draft yourself.
- Starting with a vague question. An unfocused brief produces a broad, shallow report. Fix it by setting the audience, limits and sections first.
- Trusting citation lists without opening them. A source list shows where text came from, not that it fits. Fix it by opening the sources behind every important claim.
- Uploading confidential documents without checking terms. Hosted tools handle data differently. Fix it by reading each vendor's data terms, or by running a self-hosted agent for sensitive work.
- Underestimating setup for open source tools. They are free to license but need hosting, a model endpoint or API keys, and some technical skill. Fix it by allowing time for setup and by testing with a short question first.
How we put this together
This is an evidence-based guide built from each tool's listing in our directory, which is compiled from the vendor's own site and pricing page. It is not a review based on using the tools. We don't rank tools, and nobody paid to be included. Some outbound links may be affiliate links; see our disclosure.
Frequently asked questions
Is there a free way to produce a deep research report with AI?
Yes, with limits. GPT Researcher and Local Deep Research are free open source software, though you pay for any model or search APIs you connect. Perplexity, ChatGPT, Elicit and Gemini Notebook have free plans with capped usage.
How long does a deep research report take?
The agent's run can be quick: GPT Researcher's listing says a deep research run typically takes 30 to 60 seconds per query, and Local Deep Research lists a Quick Summary mode for answers in 30 seconds to 3 minutes. Checking and editing is where most of your time goes.
Can AI write a research report I can submit or publish?
It can write a draft, but you remain responsible for accuracy. Check every citation, rewrite in your own voice, and follow your institution's or publisher's rules on AI use. Have a qualified person review legal, medical, financial or academic work.
Can I keep my research private?
Local Deep Research runs on your own hardware, and its README states there is no telemetry or analytics, though it describes its egress controls as best-effort. Hosted tools process data on their servers, so read each vendor's data terms before uploading sensitive files.
What is the difference between deep research and a normal AI search?
A normal search answer comes from a single pass. A deep research agent plans sub-questions, searches several times, reads many sources and writes a longer cited report. That makes the output longer, but it still needs checking.
Compare more options in our Research and search category. If you only need a quick sourced answer, see how to get AI search answers with cited sources.
