ChatGPT Deep Research: Access, Tasks and Limits

ChatGPT deep research creates cited reports from complex questions. It can use web pages, files you upload and apps connected to your account. Your task counter shows what remains, and the answer can still be wrong.

Free has limited deep research. Go includes Free features. Plus has expanded deep research and Pro has maximum deep research. These are the labels on OpenAI's pricing page; your own counter shows the tasks left for your account.

This is useful when a quick answer leaves too much out. You can steer the research before and during a task, then read its account of what it found. But the links are a starting point for your judgement, not a promise that every sentence is right. Access also depends on your plan and country.

What does deep research actually do?

OpenAI recommends search for quick facts. Deep research is for a question that needs material gathered and weighed from several places. It reasons through the task, draws together what it finds, and gives you a documented answer. OpenAI describes the finished output as a structured report with citations or source links.

Consider a choice with several parts: rules, different versions and accounts that disagree. A short chat response may give you a useful starting point. The research output gives you room to trace the material behind its answer. That is the difference worth paying attention to, rather than the length of the final text.

The tool can use public web pages, files you upload, selected websites and eligible connected apps. Those are different kinds of source. A file you provide may supply your own context, while a public page may give a rule or a current announcement. You decide which material belongs in the question. OpenAI's product page also says you can add approved URLs and private files to the work.

Who gets deep research? Plans as listed on 26 October 2026

Free accounts have limited deep research access. Go is another personal plan and includes Free features. OpenAI's current personal-plan page labels Plus access as expanded and Pro access as maximum. Your in-product counter shows your remaining tasks, so that is the figure to consult before starting a long question.

Access to the feature also depends on your country or territory. In Enterprise and Edu accounts for organizations and schools, administrators control access by role. If the feature is missing from your account, your plan label alone does not settle the question.

OpenAI says usage varies by plan. For plans with a fixed monthly allowance, the counter resets every 30 days from the date of first use. Your first use starts that 30-day clock, rather than the first day of a calendar month. Look at the counter in the account you will actually use. It reflects your current allowance more directly than someone else's table.

The access label is only the beginning of the choice. Ask whether your question needs a long answer and whether you have time to read it. A person with limited access may be better served by saving a task for a problem that affects a real decision. A person with more access still needs a question worth investigating. Either way, the counter tells you what you can use now, while the plan name tells you only the broad tier.

What happens during a research task?

You begin by saying what decision or answer you need. In our view, name the topic, the place it applies to and the form of the result you want. Add any limit that would make the answer unusable. A loose request to research everything about a subject gives you less direction.

ChatGPT proposes a research plan before it starts. You can review and change that plan. This is the moment to remove a question that does not matter or add an important source. During the work you can follow progress and interrupt it to adjust the focus. In our view, these controls are most useful when you catch a poor direction early.

You can also decide how narrow the web search should be. OpenAI lets you limit research to specific websites. Or you can put those sites first while still allowing a wider search. The choice depends on the question. In our view, a regulator's rule deserves a tighter source set. An open comparison may need a wider view, followed by closer reading of the decisive pages.

OpenAI's release notes describe newer source controls and a larger set of connected apps. That helps explain why older guides to the feature may show a different workflow. For the current controls, use the help page linked below. The product page says a finished report can be shared or downloaded as a PDF or Word file.

Where are the real limits?

First, a task is subject to your allowance. OpenAI gives the current remaining count in the product and says use varies by plan. Do not treat a published number from another account as your allowance. Make your question specific before you start. If all you need is one simple fact, OpenAI recommends search instead.

Second, an app that appears in ChatGPT is not automatically a source for deep research. Availability varies by plan, region and work account. The provider account's own permissions still matter. Deep research can read available information in a connected app, but it does not change that app's data as part of the task.

Third, read the cited pages before you trust a conclusion. OpenAI has warned that deep research can make factual errors and draw wrong conclusions. Open the source behind any figure, rule or decision that matters to you. Read the relevant passage and its date. If two strong sources disagree, the answer should help you see the disagreement rather than hide it.

Start with the point you would repeat to a colleague. Open the link closest to that point and look for the original wording. If the page does not say what the answer claims, set that sentence aside. Then examine the figure or rule that would change your decision. A long document is easier to use when you separate its central evidence from its background detail. Give the decisive links your attention first.

Fourth, think about the material you put into a task. OpenAI says personal consumer content may be used to train its models, depending on account settings. Its Data controls page says you can turn off “Improve the model for everyone” for new conversations. The same page says Business, Enterprise and Edu workspace content is not used to train OpenAI models by default. Workspace restrictions also affect which connected apps you can use.

When should you choose search instead?

Choose search when you need a short answer with a few links: a date, a definition or the latest notice. For a comparison or a disputed point, consider deep research and allow time to read the result. OpenAI makes the same broad distinction: search for quick facts, deep research for depth and thoroughness.

For example, asking when an office closes is a quick lookup. Asking how several official notices affect a worker in a particular situation needs more care. In the second case, tell the tool which country and worker category matter. For a question tied to one official site, choose the source restriction. The result should make the answer easier to examine, not replace your own reading of the rule.

Frame the request around the decision. Ask for the relevant rules, a link for each one and a clear account of any disagreement. This asks for the shape of the answer and visible evidence. It is still your job to read the pages before you act on the result.

When you get the result, compare its opening answer with your original question. If it solves a different problem, change the question before spending time on the details. If it answers the right question, move to the source links and the most important caveat. Keep a short note of what you accepted and what remains uncertain. Keep that note for the next conversation about the decision.

Before you share the answer, scan its main conclusion, open the links that carry it and check that the documents are current. If the answer mixes your uploaded file with public material, be clear about which one supports each point. In our view, a clean source trail matters more than length. For a decision with financial, medical or legal consequences, take particular care with the original documents and qualified advice.

Frequently asked questions

Can I use deep research on a Free account?

Yes, OpenAI lists limited deep research on Free. Your ChatGPT task counter shows how much access remains for your account. Availability also depends on your country or territory.

How many deep research tasks do I get?

OpenAI says usage varies by plan and points you to the in-product counter for remaining tasks. For a plan with a fixed monthly allowance, it resets every 30 days from first use. Check the counter rather than relying on an old number.

Can deep research read my connected apps?

It can use apps connected to your account when those apps support deep research and your permissions allow it. An app visible elsewhere in ChatGPT may not support deep research. Deep research can read available information in a connected app, but it does not change that app's data as part of the task.

Does a report with citations mean the answer is correct?

No, OpenAI says the feature can make factual errors or draw wrong conclusions. Follow the links behind important claims and read the original passages before relying on them.

Sources

OpenAI deep research help; ChatGPT plans; Deep research product page; Introducing deep research; ChatGPT release notes; Connected apps help; Consumer data use; Data controls help.