Wordsmith is a legal platform built only for in house teams. It captures legal requests from the rest of the business, triages and routes them, resolves routine ones and records what happened. Requests arrive from Slack, Teams, email and other business systems, and plain language rules decide which become tasks and whether they start automatically or wait for approval. Agents then answer questions, review and redline contracts against the team's playbook, and run scheduled work.
Around them sit drafting, an assistant, repositories, Reports for extraction with clause and page citations, and Research across more than 130 jurisdictions in 60 countries. Usage & Value turns the work done into estimated hours and outside counsel cost. The product works in Word, Outlook, Gmail, Slack and Teams, and connects to iManage, SharePoint, Google Drive and other tools through native connectors and MCP. Models from OpenAI, Anthropic and Google run under zero data retention terms, and hosting is on AWS in the EU by default. Wordsmith AI Ltd is registered in England and Wales and has raised $100 million.
Side by side on the same 15 capability axes and 12 legal signals: where the two part company, and what neither one publishes.
Vendors in the same categories as Wordsmith whose graded profiles look most alike, with a line on what each product does.
Similar is not the same, so check that each one does the job you need. No vendor pays for placement.
Ordered by how closely the graded profiles match across the 15 capability axes, inside the same categories. See how each axis is graded in the methodology.
All 15 axes, graded from public sources on the date shown. Hover a grade to see what the letter means on that axis.
How much of the product is actually AI. Whether the machine learning is the mechanism the buyer is paying for or a feature layered onto conventional software, and whether the vendor is specific about which is which.
The product runs a four step loop of receive, route, resolve and record, and models carry the middle two. Triage classifies an incoming request by priority, jurisdiction and commercial exposure and sends it to the right lawyer, and agents answer or draft without one where the work is routine. Agents respond in Teams, Slack or email and also run on a schedule for matters that repeat.
Review checks inbound paper against the playbook with more than 100 checks, explains why a clause breaks policy using past negotiated positions, and on a returned draft reviews only what changed. Research applies jurisdiction aware answers that adapt to each entity's risk profile through Company Profiles, and Personal Memory folds a user's edits back into later answers.
Whether the vendor publishes measured accuracy on citations and assertions, grounds output to primary sources, and says plainly what its system does when it does not know. Legal has a documented public record of fabricated citations reaching filed briefs, so an untested claim of accuracy is not evidence.
Research draws on primary legislation, databases and guidance from curated government, regulatory and authoritative sources across more than 130 jurisdictions in 60 countries. The sources update continuously, and live web results fill in recent developments. Companies House, EDGAR and Open Corporates connect for live checks on entities and officers. The assistant returns cited summaries linked to the underlying legislation.
Reports links every extracted value to its exact clause, page number and surrounding context, and a reviewer clicks a cell to check it, then approves and locks it or flags it. How a conflict between a curated source and a web result is settled is not described, and no measured accuracy is given.
What the system decides on its own, what a lawyer must approve, and whether the vendor documents where the review point sits. A tool that drafts under review and a tool that files without one are different products and different risks.
Work reaches a lawyer only when it needs judgment. Intake rules, written in plain language, decide which requests become tasks and whether a task starts automatically or waits for approval, and each task can be assigned, tracked and moved through the legal team. Routine NDAs that arrive in Slack or email are run against the playbook and returned without a lawyer, while deviations are flagged for review.
Review surfaces are concrete. In Reports a reviewer checks each cell against the cited clause, then approves and locks it or flags it, and in Word each proposed redline can be accepted, rejected or improved. Where the line sits between routine work and work that needs a lawyer is set in the customer's rules, and what follows a wrong answer is not described.
Named, dated evidence that the product works in production at real firms or legal departments. Case studies with figures and identified customers count. Unattributed testimonials and launch announcements do not.
Trustpilot's story reports contract review time cut by 85%, with its chief trust officer, Anoop Joshi, quoted. Financial Times general counsel Dan Guildford and Belron's Janet McCarthy speak to how their teams use the product. Kantar's head of legal for APAC and MEA, Trevor Brannan, describes measuring value with Usage & Value.
The customers page carries stories from Canva, Trip.com, Bynder, Belron, Trustpilot, Selfridges, Starling, Sage, Multiverse, Typeform, Fugro, DataSnipper and others, and partnerships with Lawyers On Demand, Morae and Microsoft. Apart from Trustpilot's figure, the stories describe use rather than measured change.
How client confidences are handled: attorney client privilege and work product treatment, segregation of one client matter from another, whether client data trains any model, and what the vendor commits to in writing rather than in marketing.
Clause 3.2.1 of the terms of 31 October 2024 bars Wordsmith from training any large language model or other machine learning model on customer inputs. The security page adds that customer data is used for no purpose beyond service delivery. OpenAI, Anthropic and Google are each named with the contract under which zero data retention applies, and prompts and outputs are not stored, logged or used for training by any provider.
The security page says the product is designed to maintain attorney client privilege and that its architecture aligns with ABA Formal Opinion 512. Clause 11 binds each party to keep the other's confidential information. Data is kept separate from other customers, though how that separation works is not described.
Whether the vendor is clear that it supplies a tool rather than legal advice, who its audience is, and how it addresses unauthorized practice of law, competence and supervision duties, and jurisdiction limits. ABA Formal Opinion 512 is the reference point. Where the advice line is not the duty a product raises, the axis is read through the nearest professional duty it does raise: judicial conduct rules and the reviewing duty for products sold only to courts, and the duty to bill for time actually spent for products that draft time entries.
Clause 9.3 of the terms provides output for general information only, and clause 9.7 puts decisions made in reliance on the service, including legal advice, at the customer's own risk. The security page says the product supports professional judgment rather than replacing it and points teams to ABA Formal Opinion 512 and local bar guidance on disclosing AI use, noting that obligations vary by jurisdiction.
The product is built to resolve routine requests from sales, procurement, HR and security without a lawyer, under rules the legal team writes. The lawyer's own competence and supervision duties are not addressed, and no jurisdiction limit is set on what agents answer.
Published governance over model behavior: who owns it inside the vendor, what is tested before release, and what is disclosed about disparate output across matter types, parties, or populations.
No governance position for model behavior is set out. Nothing names a person or role accountable for what the models produce, describes testing before release, states responsible AI principles or reports uneven output across matter types or counterparties. The security material concerns information security, and the ISO 27001 work under way is a security standard.
Usage & Value reports what work runs through the product and its estimated value, which measures use rather than quality.
Retention, deletion, access control, and what happens to prompts and documents after they are processed. Whether the vendor states its subprocessors and its incident practice, or leaves the buyer to assume.
Data is encrypted with AES 256 at rest and TLS 1.2 or higher in transit, and internal access follows least privilege. Single sign on runs through Okta, Azure Active Directory and Google Workspace over OIDC and SAML 2.0, with multifactor authentication and role based access control. An outside firm tests the platform every year, and under a formal incident response plan a confirmed breach is reported to affected customers within 72 hours with details and remediation steps.
Customers control retention. Data can be deleted permanently at any time, all customer data is deleted when the contract ends, and custom retention policies can match internal governance. The subprocessor list, provider policies, data processing agreement and incident response plan sit in the Vanta trust portal, which is closed to automated reading.
What the vendor stands behind contractually when its output is wrong. Indemnities, caps, carve outs, insurance, and whether any of it is published or only reachable through a negotiated agreement.
Clause 9.1.2 of the terms caps Wordsmith's total liability at the higher of £10,000 and the fees paid in the 12 months before the claim. Clause 9.1.1 excludes listed categories of loss, and clause 9.2 keeps liability for death or personal injury from negligence, fraud and anything the law does not allow to be excluded.
The indemnity in clause 8 runs from the customer to Wordsmith, covering use of the service and breach of the terms. Clauses 9.3, 9.4 and 9.7 disclaim the accuracy of output and leave decisions made in reliance on it to the customer. No indemnity to the customer, warranty on output or insurance position is offered. English law and the courts of England and Wales govern.
How deeply the product reaches into the systems legal work already lives in: document management such as iManage and NetDocuments, Word and Outlook, contract lifecycle management, matter management, e-billing, and court filing systems.
Microsoft Word carries playbook reviews, redlines, precedent, translation and consistency checks inside the document. Slack and Microsoft Teams host intake and Ask Legal. In Outlook and Gmail the product summarizes threads, drafts replies from the playbook and starts reviews from attachments, and iManage, SharePoint and Google Drive connect as knowledge sources.
Native connectors cover Slack, Google Drive, SharePoint, Notion, Confluence, Attio, HubSpot, Jira and Ironclad, set up from Workspaces settings, and MCP and an API reach other systems such as a CRM. Zapier and n8n open more than 8,000 further integrations. What each connector syncs, and in which direction, is not described.
Where the software runs and where the data sits. Multi tenant cloud, single tenant, private deployment, on premises, and whether region of residence is a published option or an enterprise conversation.
Hosting is on AWS, with EU data residency by default and the United States available, on infrastructure certified to ISO 27001, SOC 1, 2 and 3 and PCI DSS Level 1. Data stays in the customer's chosen region unless the customer consents to a move. Each customer's data is kept separate from every other.
No single tenant, private or on premises option is offered. Prompts go to OpenAI, Anthropic and Google, and where that processing runs relative to the chosen region is not stated.
Independent attestation a buyer can pull without a sales call: SOC 2, ISO 27001, penetration test summaries, a trust center with current reports and named scope rather than a badge image.
SOC 2 Type II certification and GDPR compliance are stated, with annual independent penetration testing and ISO 27001 certification in progress. Full documentation sits in a Vanta trust portal linked from every page, which is closed to automated reading. No auditor, report period or scope appears on the open site.
Which models sit underneath, whose they are, where they run, and whether the vendor commits to telling customers when that changes. A legal buyer inherits every dependency it cannot see.
Three model providers are named, each with the contract that sets its terms. OpenAI runs under its API Data Usage Policy, Anthropic under its Commercial Terms and Google's Gemini under the paid API Gemini Terms, all through paid enterprise APIs with zero data retention. Research also pulls live web results.
No specific model or version is named, and no commitment covers telling customers when a provider or model changes.
Whether a buyer can learn what this costs without entering a sales process: published rates, the unit being charged, what sits behind an enterprise tier, and what implementation adds.
Every route in is a demo booking, and no price is listed. The terms set the shape. Fees are fixed in an order form and invoiced monthly, the subscription runs for an initial term and renews automatically, and either party can end it at the term's end with 30 days notice. Wordsmith gives 30 days notice of a fee change, and a material change to the service lets the customer cancel with a prorated refund.
An ROI calculator estimates savings, and Usage & Value prices the work done at the customer's own in house and outside counsel rates.
Who the product is actually built for. AmLaw, midlaw, small firm and solo, in house departments, government and courts, and which practice areas are supported rather than merely claimed.
Only in house legal teams are served, with workflows designed around general counsel, legal operations and in house lawyers rather than law firm billing. Solution pages cover general counsel and legal operations, the procurement, sales, security and HR teams that send requests, and company sizes from startups to enterprises.
Legal coverage is stated by jurisdiction, more than 130 across 60 countries, and the site runs locale versions in several languages. Practice coverage is expressed through the requesting team and contract type rather than as areas of law, and government and court use are not addressed.
The public pages on file for Wordsmith, with the recorded signals each one supports and the date it was last read. Open any of them and check the reading against the record.
Prompt and Output Retention, Ethical Walls and Matter Segregation, Bar Guidance Alignment and 1 more
Read Oct 12, 2026
Client Data in Training, Third Party Request and Subpoena Notice
Read Oct 12, 2026
Fabricated Citation Record
Read Oct 12, 2026
Billing and Fee Posture
Read Oct 12, 2026
Court Disclosure Support
Read Oct 12, 2026
Primary Law Corpus Provenance
Read Oct 12, 2026
Wordsmith joined the iManage Technology Partner Program. Its agentic workflows can now read content held in iManage through iManage's APIs and use it for drafting, review and answering routine legal questions. Wordsmith says returning finished work product to iManage is planned for future releases.
Wordsmith shipped enterprise controls aimed at extending AI assisted legal workflows beyond the legal team to other business units. The release adds the access and approval mechanisms needed to widen who can run legal work without widening who can approve it.
A signal records what public sources say on the date shown. It is not a grade and it is not a recommendation. Where a signal reads Not addressed, it means the index did not locate the material in public sources on that date, which is a statement about disclosure rather than about the product.
Can material a lawyer puts into this product be used to train a model?
The published terms prohibit training on customer content. Not a policy page, the agreement.
Clause 3.2.1 of the terms of 31 October 2024 commits Wordsmith not to train any large language model or other machine learning model on customer inputs. The security page adds that customer data is not used to improve the platform for other customers. OpenAI, Anthropic and Google process prompts under contractual zero data retention and do not train on them.
How long does the product keep what a lawyer typed, and can that be set to zero?
The customer controls the retention window, by product configuration or by contractual instruction, but zero retention is not stated as available.
Customers keep control of their data, can delete it permanently at any time and can set custom retention policies to match internal governance. All customer data is deleted when the contract ends. No default retention period is given for prompts or outputs, and the model providers keep nothing after processing.
Does retrieval respect the firm’s ethical walls, or can the model read across them?
Segregation is asserted in public materials with no published detail on how it is enforced.
Each customer's data is kept completely separate from every other customer's, with no mixing. Single sign on, multifactor authentication and role based access control sit on top, and playbooks and templates are shared across an organization only when a user chooses. How the separation between customers, or between teams inside one, is enforced is not described.
If someone subpoenas the vendor for a firm’s data, does the firm hear about it first?
Terms commit to notice where lawfully permitted. No transparency report located.
Clause 11.4 of the terms lets a party disclose confidential information when a law, regulator or court requires it, and where the law permits it gives the other party as much notice as possible. The disclosing party also takes account of the other party's reasonable requests about what is disclosed. The duty is mutual and survives termination. No transparency report is offered.
Where does the law in this product come from, and does the vendor have the right to use it?
Coverage is described by jurisdiction with no identification of the underlying corpus.
Research covers more than 130 jurisdictions across 60 countries, with primary legislation, databases and guidance from hundreds of curated government, regulatory and authoritative sources. Those sources update continuously, and live web results fill in recent developments. No individual source of legislation or case law is named, and no license basis is given. Companies House, EDGAR and Open Corporates are named, as registries for checking entities and officers.
Does the product tell you when the authority it just cited has been overruled?
No located public material addresses whether authority is checked for subsequent history.
Answers come back cited and linked to primary legislation, from sources that update continuously, with live web results for recent developments. Whether a cited provision or decision still stands is not flagged, and no citator or treatment signal is part of the product.
What does the product do when the answer is not in the corpus?
No located public material addresses what the product does when it cannot ground an answer.
Triage sends a request to a lawyer when it needs judgment, and intake rules can hold a task for approval before it starts. In Reports a reviewer can flag any extracted value for review. No behavior is described for a question the system cannot answer from its sources, and no confidence measure is shown.
Does a public court record exist addressing fabricated or hallucinated legal citations in output from this product?
No court order, opinion or disciplinary record addressing fabricated or hallucinated legal citations produced by this product has been located as of the date shown. This is a statement about the public record on that one subject, not a finding about the product, and this signal is not a litigation history.
The AI Hallucination Cases database maintained by Damien Charlotin, which records court decisions worldwide that address hallucinated AI content and the tool involved where known, has no entry naming Wordsmith or Wordsmith AI Ltd. Its products serve in house legal teams and the business functions they support.
Has the vendor engaged in public with the ethics opinions its buyers are bound by?
Public materials engage with at least one named ethics opinion.
The security page cites ABA Formal Opinion 512 twice, saying the product's architecture aligns with it and, in a FAQ on disclosing AI use, telling teams to review it with local bar guidance. That answer notes that some bodies recommend disclosure for significant AI use while others do not require it for internal tools. No opinion from another jurisdiction is named.
Does the vendor address what happens to the bill when the work takes an hour instead of six?
The product does not touch a fee between a lawyer and a client. It operates before an engagement exists, or it is bought by a team that bills no client for the work. Savings claims aimed at the buyer’s own cost are recorded in the summary and do not make the row a savings claim, because no client bill is in the loop.
Wordsmith is sold to in house legal teams, with solutions for general counsel and legal operations and for the procurement, sales, security and HR teams they support, from small businesses to enterprises. Named customers include Canva, Trip.com, Bynder, Belron, Trustpilot and Selfridges. Its savings claims are aimed at those teams. A customer story says Trustpilot cut contract review time by 85%, and the home page counts hours saved per lawyer each week and outside counsel spend. Wordsmith Usage & Value puts a number on what the product delivers to the legal department.
Can a firm get this vendor through a client’s AI clause without a bespoke negotiation?
A current subprocessor or model provider list is published.
The three providers that see customer content, OpenAI, Anthropic and Google, are named on the security page with the contract that sets each one's zero data retention terms, readable before any agreement. The full subprocessor list, data processing agreement and incident response plan sit in the Vanta trust portal. No client facing disclosure template is offered.
If a judge’s standing order requires an AI disclosure, can the product produce one?
Some elements of the record are available, short of a document level export.
Reports links every extracted value to its exact clause and page, and records whether a reviewer approved and locked it or flagged it. In Word, each proposed redline is recorded as accepted, rejected or improved. The model behind a passage is not shown, and no export for a court filing is offered.
Every grade and every signal on this index is drawn from public sources and dated. If a record is wrong, out of date, or missing an artifact the index did not locate, send the source and it will be reviewed and the record redated. Vendors are welcome to submit documentation. Nothing on this index is for sale, including a listing, a placement, or a grade.