Which research jobs can you hand to an AI agent today?

AI agents for research can do the gathering and first-draft part of a research job today: they plan a search, read many sources and return a structured report with citations. They cannot yet be trusted to get every citation right. A peer-reviewed audit of deep research systems found citation accuracy between 40% and 80% across the systems it tested. So the job you hand over is "find, read and draft with sources", and the job you keep is "check the sources and decide what the findings mean".

This page covers what research agents do according to their documentation, which research jobs you can hire one for, which roles change, and where to start. It describes the products and studies cited, as of November 2026.

What AI agents do in research today

The two deep research products read for this page document a similar workflow. You describe the question, the agent proposes a plan, you edit it, and the agent then searches and reads before writing a report.

• OpenAI's deep research in ChatGPT. According to OpenAI's help center, it creates a proposed research plan you can review and modify before research begins, lets you follow progress and interrupt it, and returns "a structured report with citations or source links so you can verify the information." It can search the public web or only sites you name, read files you upload, and use connected apps such as Google Drive or SharePoint where your plan and workspace allow it. OpenAI states it uses read actions from connected apps and "does not use app write actions as part of research." Finished reports include a sources-used section and an activity history.
• Deep Research in Gemini Apps. Google's help page says Gemini creates a research plan you can edit, uses Google Search by default, and can add Gmail, Drive, uploaded files or NotebookLM notebooks as sources. Google says a report "usually takes about 5-10 minutes," and longer for complex ones. Both OpenAI and Google document usage limits that depend on your plan.
• Elicit's Research Agent. Elicit, a research software vendor, says on its product page that the agent searches over 138 million academic papers plus clinical trials, patents and regulatory databases, and that "every claim links back to the exact supporting passage, table, or figure." That is Elicit's description of its own product.

The common thread: the agent does the reading volume, and the citations are how you check it. That makes the citations the part to test.

How reliable are the citations? DeepTRACE, a study published in the ICLR 2026 conference proceedings, audited generative search engines and deep research configurations from several public providers (the paper names GPT-4.5/5, You.com, Perplexity, Copilot/Bing and Gemini). Its authors report that deep research configurations "still exhibit large fractions of unsupported statements, with citation accuracy ranging from 40–80% across systems," and that they remain "highly one-sided on debate queries." Read that as: a cited sentence in an agent's report may not be supported by the source it points to, and the share varies by system. The study uses an LLM judge that its authors validated against human raters.

Research professions are setting rules for this. Cochrane, the Campbell Collaboration, JBI and the Collaboration for Environmental Evidence published a joint position statement in November 2025 on AI in evidence synthesis. According to Cochrane's summary, it "calls for human oversight, transparency, and justification when AI is used" and expects researchers to report AI use transparently and assume responsibility for their synthesis.

Jobs you can hire an agent for

Each job has its own page with a task breakdown, a brief template you can copy, risks and how to run a trial.

• JobSupervised
  AI agent for market research: what it does and needs

  An AI agent can moderate customer interviews at scale, draft short surveys and produce cited desk research. A researcher still designs the study, checks every source and quote, and decides what the findings mean.
• JobSupervised
  AI agent for competitive intelligence: what it does

  An AI agent can watch competitor sites, pull competitor mentions out of sales calls and win-loss notes, and draft battlecard updates with the source behind each change. A person reviews every change before it reaches sellers and owns positioning and strategy.
• JobSupervised
  AI agent for literature review: what it does and needs

  An AI agent can search databases, screen papers against your criteria and extract data with supporting quotes. A reviewer still verifies every citation, settles inclusion calls and owns the conclusions.
• JobSupervised
  AI deep research agent: what it can do for reports

  An AI research agent can turn a question into a plan, read many sources and draft a cited report. A person still sets the question, checks every claim that matters against its source and owns the conclusions.

Ready

0of 4

Supervised

4of 4

• AI agent for market research: what it does and needs
• AI agent for competitive intelligence: what it does
• AI agent for literature review: what it does and needs
• AI deep research agent: what it can do for reports

Assist only

0of 4

Not yet

0of 4

They differ in how easy the output is to check:

Job	What the agent does	What a person keeps
Deep research reports	Plans a search, reads sources, drafts a cited report	Checking key citations, deciding what to act on
Market research	Gathers public data on a market, customers and prices	Primary research such as interviews and surveys, sizing assumptions
Competitive intelligence	Tracks competitors' public pages, pricing and news	Interpreting moves and deciding the response
Literature review	Searches papers, extracts data, drafts a summary	Inclusion decisions, appraisal, and responsibility for the review

For finance-specific research (filings, earnings, company data), see financial research.

Roles they affect

This function covers two roles whose reading and drafting overlap with what research agents document. Each page goes task by task through what an agent can take and what stays with the person.

• Role
  Can AI replace market research analysts? Task by task

  AI tools documented here run customer interviews, draft short surveys, write cited desk research and turn survey data into report drafts, each with a person checking the output. Forecasting, measuring program effectiveness, directing interviewers and advising management stay with the analyst.
• Role
  Can AI replace research assistants? A task-by-task answer

  Documented AI tools search the literature, extract data from papers, run statistical analysis and draft cited reports, with a person checking the output. Work with study participants, consent, protocols and quality control has no cited agent evidence and stays human.

For scale: the U.S. Bureau of Labor Statistics puts the median annual wage for market research analysts at $78,760 (May 2025, SOC 13-1161, which groups market research analysts and marketing specialists). BLS projects employment in that group to grow 7% from 2025 to 2035. The BLS duty list for the role includes gathering data on consumers, competitors and market conditions and converting findings into reports, the parts closest to what research agents document. It also includes designing surveys and presenting results to clients, which the product documentation read for this page does not describe agents doing.

Where to start

Start with a question where a wrong citation is cheap to catch and cheap to be wrong about: a market overview for an internal discussion, a scan of competitors' public pricing pages, a first pass over the literature on a topic you know well. Avoid starting with anything that goes straight to a client, a board, a regulator or a publication.

For example (assume, for this example, a 20-person software company): you ask an agent for a report on how five named competitors price their entry plans, limited to their own pricing pages. The agent returns a table and a page of notes with a link for each price. A person opens every link, checks each price against the page on that day, and marks the rows that do not match. The number of mismatches is your measure of whether to trust the next report.

Before you hand over any research job, we recommend:

1. Write the question and the source rules down. Name the sources the agent may use and those it may not. OpenAI's deep research, for example, can be restricted to sites you list.
2. Ask for a citation on every claim. A claim without a link is a claim you cannot check.
3. Check a fixed sample of citations. Open them and confirm the source says what the report says. Given the 40–80% range DeepTRACE reports, decide in advance how many unsupported claims you will accept.
4. Check dates and access. Note when each source was published, and flag sources behind paywalls the agent may not have read in full.
5. Keep conclusions with a person. The agent drafts; someone who knows the subject decides what the findings mean and signs their name to them.
6. Know where your data goes. If the agent reads your Drive, email or internal files, check what the provider retains and who in your workspace can run it. OpenAI says deep research follows the same data handling settings as regular ChatGPT conversations.

If you want help scoping a research job, use the brief below. The guides on hiring an AI agent and writing an agent job description cover scope, trials and acceptance criteria in more detail.

Sources

1. 1
   Deep research in ChatGPT

   OpenAI Help CenterPrimaryAccessed Nov 14, 2026

2. 2
   Use Deep Research in Gemini Apps

   Google (Gemini Apps Help)PrimaryAccessed Nov 14, 2026

3. 3
   Research Agent

   ElicitVendorAccessed Nov 14, 2026

4. 4
   DeepTRACE: Auditing Deep Research AI Systems for Tracking Reliability Across Citations and Evidence (ICLR 2026)

   International Conference on Learning Representations (Venkit et al.)PrimaryAccessed Nov 14, 2026

5. 5
   Setting the standards for responsible AI use in evidence synthesis

   CochranePrimaryAccessed Nov 14, 2026

6. 6
   Market Research Analysts: Occupational Outlook Handbook

   U.S. Bureau of Labor StatisticsPrimaryAccessed Nov 14, 2026