How to do a literature review with AI: a step-by-step guide

By the end of this guide you will have a defensible literature review workflow: a clear research question, a set of relevant papers found with AI search, a map of how they connect, key data pulled into a table, and every citation checked against the original paper. The AI tools do the searching, summarising and organising. You do the reading that matters and the judging.

If you want to do a literature review with AI, the main decision is which job each tool does, because some find papers, some map a field and some check how a paper has been cited. This guide compares seven tools from our Research and search category, using only what each listing states.

Key takeaways

• Start from a narrow research question, use an AI search tool to find papers, a mapping tool to see how they relate, and an extraction or citation tool to organise and verify what you keep.
• The main decision is the task each tool is built for: search and synthesis (Elicit, Consensus, AnswerThis), field mapping (Connected Papers, Litmaps), discovery (Keenious) or citation checking (Scite).
• Free plans exist but are capped on usage, graphs or messages, so check limits against the size of your review.
• AI summaries can misstate a paper. Read the sources you cite, and ask your supervisor or institution what AI use is allowed in academic work.

What you need before you start

• A research question. One or two sentences, narrow enough to search.
• Inclusion rules. Years, fields, study types and languages you will accept or exclude.
• A reference manager. Zotero is supported by Elicit, Scite, Litmaps, Keenious and AnswerThis, among others.
• Time. An exploratory review can be set up in an afternoon. A formal review takes much longer, mostly for reading and screening.
• Difficulty. Moderate. The tools are easy to use, but the judging is the hard part.
• Rules. Check your course, journal or institution's policy on AI use before you rely on these tools in assessed work.

Which kind of AI tool fits this job

Match the tool to the stage you are at.

Search and synthesis tools. Elicit searches more than 138 million papers, answers questions with source links, and fills tables with data extracted from papers you find or upload. It also has a systematic review workflow. Consensus searches over 400 million scholarly articles and summarises only what it found, with each sentence linked to a paper. AnswerThis searches 250M+ papers and builds literature reviews from a single prompt, with systematic review features on paid plans.

Field-mapping tools. Connected Papers builds a graph of papers similar to one you choose, based on co-citation and bibliographic coupling, with Prior Works and Derivative Works views. Litmaps starts from one or more seed articles and uses citations and references to surface related work, with monitoring emails for new papers.

A discovery assistant. Keenious lets you describe a topic in plain language and shows results beside a topic map. Its companion explains methods and points to papers, but it is designed to decline writing drafts or finished reports.

A citation checker. Scite shows the sentence around each citation of a paper and classifies it as supporting, contrasting or mentioning, so you see how later work treated a finding.

Step 1: Write your research question and criteria

You end with a question and rules that make searching and screening consistent.

1. Write the question in one or two sentences.
2. List the key concepts and any alternative terms for them.
3. Set inclusion and exclusion criteria: date range, field, study type, language.
4. Decide what the review is for, such as a thesis chapter, a proposal or a background section.

Check it worked by asking a colleague whether the question could be answered by a set of papers rather than a single fact.

Step 2: Run a first AI search

The goal is a starting set of papers, not the final list.

1. Enter your question in a search tool. Elicit's Research Agent answers with sources you can open, and Consensus links each sentence in its answer to the paper behind it.
2. Try Keenious if you only have a vague topic. You can describe what you are curious about in plain language.
3. Open the top papers and read the abstracts to confirm they are on topic.
4. Save the good ones to a library or your reference manager.

Check it worked by confirming that most of the saved papers match your criteria.

Step 3: Map the field around your best papers

Mapping shows you what you may have missed.

1. Take two or three strong papers and use them as seeds.
2. In Connected Papers, enter one paper to see a graph of similar work. The free plan has a monthly graph limit.
3. In Litmaps, add seed articles and review recommended papers, then use More Like This to refine.
4. Look for the Prior Works (foundational) and Derivative Works (recent follow-ups) in Connected Papers.

Check it worked by finding at least a few relevant papers that your first search did not show.

Step 4: Screen and extract data

Turn a pile of papers into a structured comparison.

1. Decide the columns you need, such as sample, method, outcome and limitations.
2. Use a data extraction feature. Elicit fills tables with custom columns from papers you find or upload.
3. For a formal systematic review, AnswerThis lists agent-led screening with PRISMA diagrams on paid plans, and Elicit has a dedicated systematic review workflow.
4. Open the source for any extracted value you plan to use. Elicit explains its answers, but you should still check them.

Check it worked by spot-checking several rows against the full text of the paper.

Step 5: Check how papers have been cited

Before you rely on a paper, see how others treated it.

1. Look up key papers in Scite to see supporting, contrasting and mentioning citation statements.
2. Use Collections to track new citations and retraction alerts for papers you rely on.
3. Consensus flags retracted papers and leaves them out of answers, but also check the source journal.
4. Drop or caveat papers with serious contrasting evidence or retractions.

Check it worked by being able to say, for each key claim, whether later work agrees or disagrees.

Step 6: Synthesise and write

Now write the review yourself, using the AI output as notes.

1. Group papers by theme, method or argument, not by author.
2. Write each section in your own words, comparing findings and pointing out disagreements and gaps.
3. If you use an AI writing aid, treat its text as a draft that you check and rewrite. Your institution may restrict this.
4. Cite each paper from the original source and format the references with your manager.

Check it worked by reading each paragraph and confirming that every claim has a citation you have opened.

Step 7: Verify, update and keep a record

A literature review is only as good as its sources.

1. Open every cited paper and confirm the claim it supports.
2. Keep a log of the tools, queries and dates you used, which helps if a supervisor or reviewer asks.
3. Set up alerts so new papers do not slip past. Litmaps' Monitor sends emails when new papers match your topic.
4. Have a supervisor or subject expert read the review before submission.

Check it worked by having a list of queries, tools and dates that someone else could repeat.

Tools for this job compared

Tool	Good fit for	Pricing	Stated limits
Elicit	Searching papers, extracting data into tables, systematic reviews	Freemium, paid from $49/mo (checked 24 Oct 2026)	Basic has limited usage; monthly usage is one shared pool that does not roll over; figure extraction needs Scale or Enterprise
Consensus	Cited answers from scholarly articles	Freemium, paid from $12/mo (checked 24 Oct 2026)	Free plan: 10 Pro messages, 3 Deep reviews and 10 Study Snapshots a month; free API use limited to 30 calls a month
AnswerThis	Cited literature reviews and systematic review workflow	Freemium, paid from $35/mo (checked 24 Oct 2026)	Free plan limited to 5 saved projects; plan names and limits vary between pricing pages; Lab & Team needs 3 seats
Connected Papers	Exploring a field from one paper	Freemium	Free plan limited to 5 graphs per month; graphs show similarity, not a citation tree
Litmaps	Citation-based maps and new-paper alerts	Freemium	Free plan limits search inputs to 20 and maps to 100 articles; filters, Zotero sync and unlimited maps need Pro
Keenious	Discovery in plain language, in Word and Google Docs	Freemium, paid from $10/mo (checked 24 Oct 2026)	Free plan: 3 conversations a day; PDF uploads limited to 3 MB; companion refuses to write drafts
Scite	Checking how papers are cited	Freemium, paid from $20/mo (checked 24 Oct 2026)	Basic has no self-service REST API key; Scite MCP needs a paid plan, a trial or an enterprise license

Prices and plan limits change, so confirm the current terms on each vendor's site before you subscribe.

Common mistakes to avoid

• Citing a paper from an AI summary. Summaries can compress or misstate findings. Fix it by opening the paper and reading the relevant section before you cite it.
• Using one tool and one query. Each index and algorithm surfaces different papers. Fix it by combining a search tool with a mapping tool and at least one rephrased query.
• Skipping criteria. Without inclusion rules, screening becomes subjective and hard to defend. Fix it by writing criteria first and applying them to every paper.
• Ignoring retractions and contrasting evidence. A heavily cited paper may be disputed. Fix it by checking its citation context in Scite and looking for retraction flags.
• Breaking your institution's AI rules. Policies differ, and some limit AI in assessed work. Fix it by reading the policy and disclosing your use where required.

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 do a literature review with AI?

Yes, within limits. Elicit's Basic plan is free with limited usage, Consensus has a Free plan with unlimited paper searches, Connected Papers allows 5 graphs a month, and Litmaps and Keenious have free plans. Large reviews will probably hit those caps.

Can AI write my literature review for me?

It can draft one, but you should not submit that draft as your own work. Tools make mistakes, and many institutions limit AI in assessed writing. Use AI to find, organise and check sources, then write and cite the review yourself.

Can I trust the citations AI tools give me?

Trust but verify. Tools such as Consensus link each sentence to the paper behind it, which makes checking easier, but a link is not proof the paper says what the summary says. Open each source before you cite it.

What is the difference between Connected Papers and Litmaps?

Connected Papers builds a similarity graph from a single origin paper using co-citation and bibliographic coupling. Litmaps starts from one or more seed articles, uses citations and references to find related work, and can email you about new matches.

How long does a literature review take with AI?

The listings do not publish a standard duration. Reading, screening and writing take the most time. Elicit's systematic review workflow screens large sets of papers on Pro, but you still need to check its output.

Compare more options in our Research and search category. If you need sources for specific claims next, see how to find real citations and sources with AI.