Deep research AI is a research assistant that splits your question into parts, searches scholarly databases, reads the most relevant papers and writes an answer in which every claim cites a paper. AI for Research's Deep Research is built for students, researchers and analysts who need an answer they can check, not a summary of web pages.
What you can do with it
- Get a quick, cited answer to a question before a meeting or a supervision: a Summary reads about eight papers and costs 1 credit.
- Start a literature review with a Deep report: about 250 papers screened, 45 or more kept, up to 12 open full texts read, written up in 2,500 to 4,000 words with figures.
- Map the evidence on a contested question with a Detailed report: 90 or more papers, a study-by-study comparison table, and a plan you review before any credits are spent.
- Ask follow-up questions about the same sources for 1 credit each.
- Take the result further: open it in the editor with live citations, save the cited papers to your Library, or turn it into a literature review, a thesis outline, a gap analysis or a slide deck.
The three depths: what each reads, returns and costs
| Summary | Deep | Detailed | |
|---|---|---|---|
| What you get | A short cited answer, about 150 to 250 words | A 2,500 to 4,000-word report with figures | A 5,000 to 8,000-word report with figures and a study table |
| Papers screened | up to about 30 | about 250 | up to about 450 |
| Papers kept | about 8 | 45 or more | 90 or more |
| Open full texts read | none (abstracts) | up to 12, more in the gap round | up to 25, more in the gap round |
| References and citations of top papers followed | No | Yes | Yes |
| Second search where evidence is thin | No | Yes | Yes |
| You review the plan first | No | No | Yes |
| Credits | 1 | 15 | 45 |
| With High quality (Pro) | 3 | 45 | 135 |
| Plans | All | All (Free: 2 a day) | Plus, Pro, Team |
High quality uses a stronger model and writes longer answers. It is part of the Pro plan.
How it works, step by step
- Type your question in the Ask box on Home or in Deep Research. A focused question works best: "Does sleep after learning improve memory consolidation?" rather than "sleep".
- Choose a depth: Summary, Deep or Detailed. The cost shows next to the Ask button before you start.
- Choose your sources in the Sources menu: All papers, Medical, Preprints, Web and My Library. You can pick several.
- Set filters if you need them: a year range (or "Last 5 years", "Last 10 years"), study types (meta-analysis, systematic review, randomized controlled trial, cohort, case-control, qualitative and more), journal quality (from the journal's h-index in OpenAlex), a minimum number of citations, open access only, and peer-reviewed only. Retracted papers are left out unless you turn that off.
- Add your own material with Attach: PDFs, Word files, text, data files, LaTeX or reference files, or papers from your Library.
- Set your role (undergraduate, master's student, PhD candidate, clinician, industry and others) so the answer is pitched at your level, and choose whether you want headings.
- Review the plan (Detailed only): edit or remove the sub-questions, then press start. Credits are charged only at this point.
- Watch the steps run, then read the answer, check the sources and ask follow-ups.
The eight steps you can see
Every run shows its work, and a saved answer keeps the record under "How this answer was made":
- Plan: the question split into sub-questions and searches.
- Search: every search run against your sources, duplicates removed.
- Screen: titles and abstracts read; off-topic and retracted papers excluded, each with a reason.
- Rank: an evidence score from study design, relevance, citations a year, recency and journal. It is computed, not typed by a model.
- Read: open full texts fetched, then one finding, sample, limitation and stance per paper. An effect size is kept only when its numbers appear in the paper's text.
- Fill gaps (Deep and Detailed): a second round of searching where evidence is thin or conflicting.
- Write: an outline, then each section from its own sources, citing every claim.
- Verify: each cited sentence checked against its source, invented citation markers dropped, and each cited DOI checked at Crossref.
What a report contains
A Deep or Detailed report has these parts:
- Summary of the answer.
- How this review was done: a methods paragraph written from the run's real counts.
- Figures computed from what was counted and read: a flow diagram in the style of PRISMA (records found, screened, excluded with reasons, included), papers by year, papers by study design, how the studies bear on the question by design, papers per sub-question, and the effect sizes the studies report, where they report them.
- Thematic sections, three to five in Deep and five to eight in Detailed.
- Where studies disagree, limits of the evidence and open questions. Detailed adds implications and a comparison of studies table.
- A consensus line, when the question can be answered yes or no, such as "2 of 4 studies that test the question support it", counted from how each cited study bears on the question.
The source panel has three tabs: Cited, All read and Web. Each source shows its study type, journal tier, preprint or retraction status, and "DOI verified" when Crossref confirms it. Under the answer, "Notes from the citation check" lists any sentence that was changed or could not be supported.
Example: one question, start to finish
This is an example of how a run is set up and paid for, not a real result.
A master's student asks "Does sleep after learning improve memory consolidation in adults?" She picks All papers and Medical, filters to the last 10 years, and chooses Deep: 15 credits. The report arrives with a flow diagram, a chart of study designs and sections on the main effect, the mechanisms and the populations studied. She asks two follow-ups ("Which studies used naps rather than overnight sleep?" and "What sample sizes did the RCTs use?") for 1 credit each, saves the cited papers to her Library, and opens the report in the editor. Total: 17 credits, out of the 45 her free plan gives her each month.
What it costs
| Plan | Price | Credits a month | Deep or Detailed reports a day | Detailed | High quality |
|---|---|---|---|---|---|
| Free | $0 | 45 | 2, one at a time | No | No |
| Plus | $12 a month | 750 | 15, one at a time | Yes | No |
| Pro | $29 a month | 1,500 | 40, three at once | Yes | Yes |
| Team | Contact sales, from 3 seats | 2,000 a seat, pooled | 100, ten at once | Yes | Yes |
Summary answers do not count toward the daily report limit. Adding headings to an answer written without them costs 1 credit; switching them off is free. The Defence Pass ($49, paid once) gives 90 days of Pro. See pricing for yearly prices and top-ups.
What it does not do
- It cannot read paywalled papers it has no open copy of. Those are read from their abstracts. Upload the PDF to your Library and choose My Library as a source to include it in full.
- It does not search Google Scholar, Scopus or Web of Science. Its paper sources are OpenAlex, Semantic Scholar, PubMed, Europe PMC and the preprint servers.
- Screening and ranking are done by AI. A relevant paper can be missed or a weak one kept. The excluded papers and the reasons are listed so you can check them.
- The consensus line is a count, not a verdict. It does not weigh study quality the way a meta-analysis does.
- It is not a systematic review. For a review you will publish, use Systematic Review.
- Medical and clinical answers need a professional's review. A report is not treatment advice.
Who uses it
- Students writing essays and theses: see AI for undergraduates and AI for a master's thesis.
- PhD candidates starting a literature review: see AI tools for PhD students.
- Clinicians and health researchers using the Medical source and study-type filters: see AI for medical research.
- Academics and analysts who need a cited brief: see AI for professors and AI for market research.
After the report
The "Build on this" row sends the report and its sources to the next job: write a literature review or a paper, find gaps, draft a thesis outline, check citations, start a systematic review, or make slides. To follow the citations of the key papers, use the Citation Map. To keep the report with the rest of a thesis and share it with a supervisor, move it into a project.
Learn more
- What deep research AI is, and when you can trust it
- The best AI for deep research on scholarly questions
- ChatGPT deep research for academic work
- Study types and levels of evidence
Last updated 2026-11-12