Legal research tools
Legal research tools in 2026: what each kind is for.
There are three kinds of legal research tool: free public databases, paid subscription platforms, and AI research tools that answer a question rather than return a list. Most lawyers use at least two. This page sets out what each does well, where it fails, roughly what it costs, and the one test that tells you whether an AI tool is safe to rely on.
Side by side
The three kinds of legal research tool.
| Kind | Examples | Good for | Weak on | Typical cost |
|---|---|---|---|---|
| Free public databases | CourtListener, Google Scholar, BAILII, CanLII, EUR-Lex, SAFLII, Kenya Law, GhaLII | Reading a known judgment or statute; verifying a citation | Finding what you do not know to look for; citator and treatment; coverage gaps outside the US and UK | Free |
| Subscription platforms | Westlaw, Lexis, LawPavilion, Primsol, Juta | Comprehensive coverage in one jurisdiction; headnotes; citators; editorial commentary | Cost; one jurisdiction per subscription; Boolean search that punishes the wrong keyword | Hundreds to thousands per seat per year |
| AI research tools | ModulawAI, Lexis+ AI, Westlaw AI-assisted research, CoCounsel, Harvey, Paxton | Answering the question in plain language with cited authorities; summarising; cross-jurisdiction comparison | Only as reliable as its grounding; some tools cannot show their sources; coverage varies widely | From per-seat SaaS pricing to enterprise contracts |
Based on publicly available information at the time of writing. Vendor capabilities and prices change; verify with each vendor before deciding.
Choosing
Five questions that sort the tools quickly.
One
Does it show the source, or describe it?
An AI research tool that answers from model memory will produce a plausible case name that does not exist. A tool that retrieves the judgment first and links to it lets you check the proposition in one click. Run a question you have already researched and count how many citations you can open. That is the whole test.
Two
Which jurisdictions, actually?
Most tools are built for one country and will answer confidently about another. Ask the vendor to list the jurisdictions where they hold primary sources rather than summaries. ModulawAI holds eight: the United States, the United Kingdom, Canada, the EU, Nigeria, Ghana, Kenya and South Africa. No comparable AI tool covers the African four.
Three
Does it check whether the authority is still good?
A citator is what the subscription platforms charge for and what free databases lack. Some AI tools now surface subsequent treatment alongside the answer; ModulawAI flags overruled, distinguished and narrowed authorities when you cite them. If a tool cannot do this, budget for a second tool that can.
Four
Where does the research go afterwards?
A research tool that lives in a browser tab leaves the result to be copied into a memo, then into a draft, then into a time entry. A tool inside the practice management system attaches the authority to the matter, drops it into the Word draft with the citation formatted, and records the time. That is why ModulawAI is research inside a legal operating system rather than a standalone database.
Five
What is the real cost per seat?
Add up the research subscription, the AI add-on, the document tool and the practice management seat. That is the number to compare against a platform that includes research. ModulawAI is one line on the invoice, priced per seat with usage-based credits. See pricing.
Free tools worth knowing
The public databases, by jurisdiction.
- United States: CourtListener and the RECAP archive for federal dockets and opinions; Google Scholar for case law; state court websites for recent decisions.
- United Kingdom and Ireland: BAILII for judgments; legislation.gov.uk for statutes as amended; The National Archives’ Find Case Law for recent judgments.
- Canada: CanLII, which is unusually complete and includes a citator.
- European Union: EUR-Lex for legislation and CURIA for the Court of Justice.
- Nigeria: court websites and the Nigerian Law Reports for recent Supreme Court and Court of Appeal decisions; coverage is patchy, which is why the subscription platforms dominate. See Nigerian legal research.
- Kenya, South Africa, Ghana: Kenya Law, SAFLII and GhaLII respectively. Free, official or semi-official, and the first stop in each country.
ModulawAI’s research draws on these public sources and others, which is why every answer can link to a judgment you can open. The AI is the layer that finds and reads; the sources remain the sources.
The verification problem
Why the citation test is the whole test, in numbers.
The failure mode of an ungrounded research tool is not a wrong answer. It is a confident answer supported by an authority that does not exist, formatted correctly enough to survive a skim. Damien Charlotin’s AI Hallucination Cases database, which tracks court decisions worldwide involving fabricated AI-generated content, recorded 2,041 cases as at 14 September 2026 across more than 40 jurisdictions. The distribution follows litigation volume: 1,395 in the United States, 217 in Canada, 111 in Australia, 57 in Israel and 41 in Brazil.
Three decisions show the range of consequence. In Mata v Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023), the court sanctioned two attorneys and their firm $5,000 in June 2023 after a brief cited six decisions that did not exist, including Varghese v China Southern Airlines. In Ayinde v London Borough of Haringey and Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin), handed down 6 June 2025 under the Hamid jurisdiction, grounds for judicial review cited five non-existent authorities and misstated section 188(3) of the Housing Act 1996; wasted costs of £2,000 plus VAT were ordered against each of two lawyers and both were required to self-report to their regulators. More recently the US Court of Appeals for the Sixth Circuit sanctioned two attorneys over briefs containing more than two dozen fabricated citations, ordering full appellate fees, double costs and $15,000 each in punitive sanctions.
Regulators have converged on the same position. ABA Formal Opinion 512, issued 29 July 2024, requires competence in the tools used, confidentiality assessment before client information is entered, and billing that reflects time actually spent. The SRA’s warning notice of 17 August 2026 states that reliance on an AI output would not be a suitable defence and that solicitors remain accountable for all work delivered to clients whether or not AI was used. The SRA has reported receiving dozens of AI-related conduct reports in the year to July 2026, covering inaccurate citations, supervision and confidentiality.
A verification protocol worth writing into your AI policy
- Open every citation. Not a sample. If the tool links to sources, this costs seconds; if it does not, that is your answer about the tool.
- Check the proposition, not just the existence. A real case cited for something it does not say is the subtler and more common failure.
- Check subsequent treatment. A correct citation to an overruled authority is still a bad submission.
- Record who verified. Under both ABA 512 and the SRA notice, supervision has to be evidenced, not asserted.
- Never file from a general chatbot. Orientation is a legitimate use. Citation is not.
ModulawAI is built around the first three of those: retrieval before generation, a link on every citation, and treatment flagged when an authority is cited. The last two are firm policy, and no vendor can supply them for you.
Sources
- Damien Charlotin, AI Hallucination Cases database, 2,041 cases as at 14 September 2026.
- Mata v Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023).
- Ayinde v London Borough of Haringey; Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin).
- ABA Formal Opinion 512 (29 July 2024).
- SRA, Warning notice: misuse of AI (17 August 2026).
FAQ
Questions, addressed.
What are the best legal research tools?
It depends on the job. For reading a known authority, the free public databases are enough. For comprehensive single-jurisdiction coverage with a citator, the subscription platforms remain the standard. For answering a question in plain language with cited authorities, and for work across more than one jurisdiction, an AI research tool grounded in primary sources is faster. ModulawAI is built for the third job and covers eight jurisdictions.
Is AI legal research reliable?
Grounded AI research, where the tool retrieves the source before answering and links every citation, is reliable in the way a junior’s memo is reliable: you check the citations. Ungrounded AI, including general chatbots, is not, because it can invent authorities. The test is whether you can open every citation the tool gives you.
Can I use ChatGPT for legal research?
For orientation on an unfamiliar area, yes, with the understanding that it cannot show you where anything came from. For anything you will cite, no. Courts in several jurisdictions have sanctioned lawyers for filing invented citations produced this way.
Are there free AI legal research tools?
Some products offer limited free tiers and the public databases are free. ModulawAI is priced per seat with usage-based credits rather than offered free, because grounded research across eight jurisdictions has a real cost to run.
How should a firm evaluate a research tool?
Take three questions the firm has already researched, run them through the tool, and compare the answer and its citations with the memo the firm actually relied on. Do it before the sales call, not during it.
Related
Keep reading.
- Legal research software: the ModulawAI research product.
- AI legal research: how grounded research works and where it fails.
- AI legal assistant: research, drafting and analysis in one assistant.
- Research that crosses borders: one matter, many jurisdictions.
Run the test on ModulawAI.
Book a demo and bring a question you have already researched. Open every citation we give you.