Midpage
Legal research platform built for litigators and law students doing in depth research, covering case law from United States federal and state courts plus tribal and military courts, statutes and regulations, and selected administrative decisions. Also generates briefs. Available as a standalone product and through a bespoke integration inside ChatGPT. Founded 2022. Participated in the 2025 Vals Legal AI Report legal research study, an independent third party benchmark, where it was evaluated on its standalone research product and scored 78 percent accuracy, 74 percent authoritativeness and 70 percent appropriateness for a 76 percent weighted aggregate, against a lawyer baseline the study measured at 69 percent aggregate and 71 percent accuracy.
Capability grades
All 15 axes, graded from public sources on the date shown. Hover a grade to see what the letter means on that axis.
AI Centrality
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 artificial intelligence is the product. Midpage is a legal research and brief generation platform where retrieval, synthesis and drafting are all model driven; there is no underlying workflow or document management system that would remain if the models were removed. The company was founded in 2022, after the generative wave rather than before it, and the product is distributed both standalone and as an integration inside ChatGPT, which is a distribution model only an AI native product can have.
Citation Accuracy and Hallucination Disclosure
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.
Independently measured accuracy with a published methodology and published per criterion figures, the second such record on the index. Midpage submitted its standalone research product to the 2025 Vals Legal AI Report legal research study, run by an outside evaluator. The evaluator published its rubric and weights in advance: accuracy at 50 percent, authoritativeness at 40 percent, appropriateness at 10 percent, across 200 United States legal research questions sourced from attorneys at named firms including Reed Smith, Fisher Phillips, McDermott Will and Emery, Ogletree Deakins, Paul Hastings, and Paul Weiss. Midpage scored 78 percent accuracy, 74 percent authoritativeness and 70 percent appropriateness, a 76 percent weighted aggregate, beating the lawyer baseline by 7 points on accuracy and 6 on authoritativeness. Those figures sit on the evaluator's own site and are checkable by an outsider without reference to any vendor claim. The evaluator also published the failures, which is why this is a strong A rather than a marketing one: it recorded three cases where Midpage returned no response at all due to technical issues, and eight cases where it explained it could not locate supporting documents. Also recorded from the same source: all evaluated systems, this one included, underperformed a generalist model when asked to survey all fifty states for a single statute, and struggled on multi jurisdictional questions generally.
Autonomy and Oversight Model
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.
Searched the vendor site, the product pages and third party coverage on 29 Aug 2026. No published position was located on what the system does unaided, what a lawyer must review, where a workflow stops, or what the vendor commits to when an output is wrong. The evaluator's report records that the product sometimes explained it could not locate supporting documents rather than answering, which is observed behaviour rather than a published oversight model, and it was not treated as one. Rebuttable with a single link to any published statement on human review.
Operational and Outcome Evidence
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.
Independent benchmark performance stands where deployment evidence would go. The strongest evidence attached to this vendor is measured test set performance from an outside evaluator, which is real and unusual but is not a production outcome: it shows how the system performs on 200 constructed questions, not what changed at a firm that adopted it. Searched the vendor site, the product pages and third party coverage on 29 Aug 2026 and located no named customer, no case study pairing an organisation with figures and a date, and no assessable deployment method. The evaluator does record that the study questions were sourced from attorneys at named firms, but those firms contributed test material rather than being disclosed as customers, and the two must not be conflated.
Privilege and Confidentiality Posture
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.
CORRECTED 29 Aug 2026 during the trust portal sweep. Previously graded D on the finding that nothing at all was located on how client confidences are handled. That was wrong: the vendor publishes a security page which was not reached in the original pass, and it answers two of the three limbs this axis tests more precisely than most records here. Training is addressed at both layers and in the negative: Midpage does not use customer data to train or fine tune AI models, and its agreements with AI model providers do not permit those vendors to train on Midpage customer data. Retention is published with actual periods rather than in general terms: web app data including user queries, uploaded materials and generated outputs is held while an account remains active and deleted within 60 days of account deletion or a valid deletion request; for plugins and integrations no submitted queries, uploads or outputs are stored at all, though those workflows may share queries with model providers who may retain them for up to 60 days. Disclosing the model provider retention window rather than only its own is unusually candid. Encryption is specified as AES-256 at rest and TLS 1.2 or higher in transit, and subprocessors are contractually required to maintain measures no less protective. Controls are stated to be independently validated through an annual SOC 2 Type II audit. Held at B rather than A because the third limb is unaddressed: no segregation model between customers, users or matters was located, and attorney client privilege and work product are not treated directly.
UPL and Professional Responsibility Posture
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.
The stated audience is litigators and law students, which is a lawyer and trainee population rather than a consumer one, and the vendor is explicit about it. That student audience is worth recording: it is the first record on the index whose published audience includes people who are not yet admitted, which is a professional responsibility question of a different shape rather than a UPL problem. Searched the vendor site, the product pages and third party coverage on 29 Aug 2026 and located no published position on the advice line, no treatment of competence or supervision duties, and no jurisdiction limits. Recorded at C because the audience is stated clearly while the position is not.
AI Governance and Bias Disclosure
Published governance over model behaviour: who owns it inside the vendor, what is tested before release, and what is disclosed about disparate output across matter types, parties, or populations.
Submitting to independent evaluation is the governance act on this record, and it is a real one. Midpage was one of only four systems that agreed to be measured and named in the Vals legal research study, in a field where the evaluator and trade press both recorded that major vendors declined to participate and at least one participated but withheld permission to publish its results. Agreeing to publication including your own failure cases is a meaningful accountability position. But it is a point in time evaluation rather than a management system. Searched the vendor site and third party coverage on 29 Aug 2026 and located no AI governance framework, no AI management certification, no named owner of model governance, no pre release testing regime, and nothing on uneven output across matter types, parties or populations.
AI Safety and Data Stewardship
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.
CORRECTED 29 Aug 2026 during the trust portal sweep. Previously graded D on the finding that no retention period, deletion control, access control detail, encryption statement, subprocessor list, hosting disclosure or incident practice was located. The vendor's security page supplies most of that. Published: AES-256 encryption at rest and TLS 1.2 or higher in transit using industry standard algorithms; a stated retention and deletion practice with a 60 day outer bound following account deletion or a valid deletion request; zero storage of queries, uploads and outputs for plugin and integration workflows; disclosure that model providers may retain submitted queries for up to 60 days; a contractual flow down requiring subprocessors to maintain security measures no less protective than the vendor's own; and independent validation of the whole control set through an annual SOC 2 Type II audit. Additional detail on service providers and data handling is available on request. Not located as of 29 Aug 2026, and why this is B rather than A: a named subprocessor list as distinct from the contractual flow down, a hosting provider or region disclosure, and an incident or breach notification practice.
AI Liability and Recourse
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.
Searched the vendor site navigation, the product pages and third party coverage on 29 Aug 2026. No published indemnity, liability cap, carve out, warranty on output or insurance position was located, and no customer terms of service was located as published on the property. Recorded as a pure absence on the surfaces reached. The exposure is worth naming: this is a litigation research product whose output is authority bound for filings, in a market where the published sanctions record for fabricated citations is growing, and nothing published addresses who bears the loss.
Practice Systems Integration Depth
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.
One integration is named and none is documented, and the one named is unusual. The product is accessible inside ChatGPT through what the evaluator describes as a bespoke integration, which is a distribution channel into a general purpose assistant rather than a connection into a firm's own systems. Searched the vendor site, the product pages and third party coverage on 29 Aug 2026 and located no legal document management connector such as iManage or NetDocuments, no practice management or court filing integration, no Microsoft Word add in, and no documentation of what the ChatGPT integration passes or what an administrator configures. For a product whose buyers are litigators inside firms, the absence of any firm system connector is the notable gap.
Deployment Model and Data Residency
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.
Searched the vendor site, the product pages and third party coverage on 29 Aug 2026, and ran targeted searches for security or infrastructure documentation without reaching any. Nothing was located on the deployment model at all: no tenancy statement, no hosting provider, no region, no residency option, and no statement of where processing happens. The product is evidently cloud delivered and reachable through a browser and through ChatGPT, but that is inference from how it is sold rather than a published position, and inference earns nothing on this axis.
Security Certifications and Trust Center
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.
CORRECTED 29 Aug 2026 during the trust portal sweep, superseding an earlier correction in the same session. The row was first graded D on the basis that no attestation was located anywhere, then raised to C when a customer of this vendor, GC AI, published that Midpage is SOC 2 Type II compliant. Both readings were working around the fact that the vendor's own security page had not been reached. It has now. The vendor states directly that its controls are independently validated through an annual SOC 2 Type II audit, which supplies the cadence as well as the standard, and states that on request it can provide SOC 2 documentation, questionnaire responses and additional detail on service providers and data handling. That is a published request route to substantive diligence material rather than a sales gate. The earlier observation that a buyer should not have to learn a vendor's security posture from a competitor's blog no longer applies and is withdrawn: the vendor publishes it itself. Short of an A because no coverage period, audit scope, report date or named auditing firm was located, and no trust portal exists, the route being a request to the vendor.
Model Supply Chain Disclosure
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.
Searched the vendor site, the product pages and third party coverage on 29 Aug 2026. No model, provider, hosting location or subprocessor was located, and no commitment to notify customers of supply chain changes. The evaluator's report distinguishes Midpage's standalone research product from its ChatGPT integration, which tells a reader the product runs independently of that channel, but nothing published identifies what powers the standalone product. A buyer cannot determine from published material which company processes their research queries.
Commercial Transparency
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.
Checked the vendor site navigation, the product pages and third party coverage on 29 Aug 2026. No pricing page was located, no rate is published, no unit of charge is stated and no tier structure appears. No free trial or self serve entry point was located on the vendor's own property, and no third party pricing figure was located either. Noted for a reader: the product being reachable through a ChatGPT integration may mean some access runs on that platform's commercial terms rather than the vendor's, which published material does not clarify.
Firm and Practice Coverage
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.
Practice focus and jurisdictional coverage are both stated with real substance. The practice position is narrow and explicit: purpose built for litigators, with the evaluator independently confirming the product is focused on supporting litigator workflows and that its responses are tailored to the research questions litigation practice actually produces. Jurisdictional coverage is enumerated by court system rather than claimed broadly, spanning United States federal and state courts, tribal courts and military courts, plus statutes, regulations and selected administrative decisions. Naming tribal and military courts specifically is a level of precision no other record on this index reaches. Short of an A because the segment is described by practice type without addressing firm size, and because no statement of what the product does not cover was located, with the independently measured weakness on fifty state surveys and multi jurisdictional questions coming from the evaluator rather than the vendor.
Legal Signals
What each signal meansA 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.
Client Data in Training
Can material a lawyer puts into this product be used to train a model?
A public policy or trust page states no training on customer content, with no matching term located in the published agreement.
CORRECTED 29 Aug 2026 during the trust portal sweep. Previously recorded as silent on the finding that no statement existed either way, with the note observing that the ChatGPT distribution channel added an unaddressed layer. The statement exists on the vendor's security page, which was not reached in the original pass, and it covers both layers. The vendor states it does not use customer data to train or fine tune AI models, and separately that its agreements with AI model providers do not permit those vendors to train on Midpage customer data. The second is a contractual constraint on third parties rather than a policy the vendor could revise alone. The distribution point raised earlier is also partly answered: the vendor distinguishes plugin and integration workflows, stating it does not store queries, uploads or outputs from them while acknowledging those workflows may still share submitted queries with model providers, who may retain them for up to 60 days. That is a candid account of what happens when the product is reached through another surface, and it is more than most vendors disclose about their own layer, let alone a partner's.
Prompt and Output Retention
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.
CORRECTED 29 Aug 2026 during the trust portal sweep, and this now reads as one of the two strongest retention disclosures on the index. Previously recorded as not addressed on the finding that no public material stated a period, a control or a deletion route. All three exist on the vendor's security page. Published: web app data, expressly including user queries, uploaded materials and generated outputs, is held while an account remains active and deleted within 60 days of account deletion or a valid deletion request, so the customer triggers deletion and a published outer bound applies to it; for plugins and integrations no submitted queries, uploads or outputs are stored at all; and model providers may retain submitted queries for up to 60 days, which is a disclosure about a third party's window that almost no vendor makes. Recorded at customer controlled on the same basis as GC AI, where deletion on demand was treated as control over how long data persists, and this record is stronger because a maximum is published alongside the right. Short of the top value because the zero retention path is a property of the plugin and integration surface rather than a setting a customer can select for the web app, which is the closest any record on this index has come to that value without reaching it.
Ethical Walls and Matter Segregation
Does retrieval respect the firm’s ethical walls, or can the model read across them?
No located public material addresses walls or matter level segregation.
Searched the vendor site, the product pages and third party coverage on 29 Aug 2026. No vendor material addresses segregation of any kind: not between customers, not between users, not between matters. No document management integration was located whose permissions the product could inherit at query time. Noted for context: a research tool that retrieves from published primary law rather than from a firm's own repository raises this question less sharply than a product indexing client documents, though brief generation from uploaded material would raise it.
Third Party Request and Subpoena Notice
If someone subpoenas the vendor for a firm’s data, does the firm hear about it first?
No located term or policy addresses third party requests for customer data.
Searched the vendor site, the product pages and third party coverage on 29 Aug 2026, and ran targeted searches for published terms or a privacy policy without reaching one. No clause committing to notify a customer of a government or law enforcement request for their data was located, and no transparency report was located.
Primary Law Corpus Provenance
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.
Coverage is described by jurisdiction and court system with unusual precision, and the sources behind it are not identified. Published coverage spans case law from United States federal and state courts, tribal courts and military courts, plus statutes and regulations and selected administrative decisions. Enumerating tribal and military courts is more specific than any other coverage statement on this index. What is missing is provenance: searched the vendor site, the product pages and third party coverage on 29 Aug 2026 and located no named source or publisher, no licence or public domain basis, no completeness statement per court system, and no update cadence or lag. The word selected in front of administrative decisions is itself an undefined boundary. Recorded at the jurisdictions only value because coverage is stated while its basis is not.
Good Law Verification
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.
Searched the vendor site, the product pages and third party coverage on 29 Aug 2026. No material was located addressing whether authority returned carries a treatment signal, whether subsequent history is checked, or whether any commercial citator is licensed. The independent evaluator measured authoritativeness, defined as whether cited sources are relevant and valid and support the statements made, and scored this product at 74 percent. Validity in that sense means the source exists and supports the proposition, which is not the same as the source still being good law, and the two were not conflated. Second research product on the index to record this absence where it sits inside the product's design rather than outside it.
Refusal and Uncertainty Behaviour
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.
Abstention behaviour is observed and quantified by an independent evaluator rather than documented by the vendor, which is a first on this signal. The Vals legal research study recorded that in eight cases Midpage acknowledged it was unable to locate the right documents and explained why the available sources did not support an answer, rather than fabricating one, and the evaluator awarded partial credit for the quality of those explanations. Separately and distinctly, it recorded three cases of pure technical failure where no response was returned at all, which is a different thing from principled abstention and is not counted as such. Recorded at the weakest value because the vendor itself publishes nothing on this: searched the vendor site and product pages on 29 Aug 2026 and located no described no answer path and no confidence signal. A reader should weigh the observed behaviour, which is favourable, against the absence of any commitment that it will persist.
Fabricated Citation Record
Does a public court record exist involving output from this product?
No court order, opinion or disciplinary record naming this product has been located as of the date shown. This is a statement about the public record, not a finding about the product.
No court order, opinion or disciplinary record naming this product has been located as of 29 Aug 2026. Instrument searched: the AI Hallucination Cases database maintained by Damien Charlotin, which tracks court decisions worldwide involving AI generated hallucinated content and records the AI tool implicated where it is known. Also checked published 2026 sanctions summaries and secondary sanctions trackers. The entries located name filers, and in some rows other products, rather than this one. This is a statement about the public record on the date shown and not a clearance. The exposure is direct for this vendor: a litigation research and brief generation product produces exactly the material that reaches filings as cited authority.
Bar Guidance Alignment
Has the vendor engaged in public with the ethics opinions its buyers are bound by?
No located public material engages with bar or ethics guidance.
Searched the vendor site, the product pages and third party coverage on 29 Aug 2026. No engagement with any named ethics opinion or professional guidance was located, including ABA Formal Opinion 512 and state bar guidance. Worth noting given the stated audience includes law students, where a published position on supervised use in an academic setting would be a natural place for such engagement and none was located.
Billing and Fee Posture
Does the vendor address what happens to the bill when the work takes an hour instead of six?
No located public material addresses billing, fee or disclosure treatment.
Searched the vendor site, the product pages and third party coverage on 29 Aug 2026. No per matter record of AI assisted work intended for fee purposes was located, and no guidance on billing, fee or client disclosure treatment was located. Distinct from most records on this index in that no time savings or efficiency claim was located either: the vendor's published positioning and the evidence attached to it are about research accuracy rather than speed, so there is no savings claim to weigh against the client's side of the equation. Recorded as not addressed rather than at the savings claims value for that reason.
Outside Counsel Guideline Readiness
Can a firm get this vendor through a client’s AI clause without a bespoke negotiation?
The material exists behind a sales conversation or an executed agreement.
CORRECTED 29 Aug 2026 during the trust portal sweep. Previously recorded as not addressed, with the note stating a firm could not assemble a response to a client AI clause from anything published. That is no longer accurate. The vendor states that on request it can provide SOC 2 documentation, questionnaire responses, and additional detail on service providers and data handling, which is an explicit offer of the three artifacts a firm most often needs. Published without any request at all: an express no training commitment binding both the vendor and its model providers, retention periods including the model providers' own 60 day window, encryption specifics, and a contractual flow down requiring subprocessors to maintain measures no less protective. A firm could answer most of a standard client AI clause from the public page and obtain the rest through the stated route. Recorded at on request rather than the subprocessor value because no subprocessor list is published and no model provider is named, only the terms binding them.
Court Disclosure Support
If a judge’s standing order requires an AI disclosure, can the product produce one?
No located public material addresses court disclosure or verification certification.
Searched the vendor site, the product pages and third party coverage on 29 Aug 2026. Output carries citations to primary law by design, so what was relied on is visible in the work product, but no per document export covering model used, sources retrieved and human verification together was located, and no model is identified anywhere in published material so the model used could not be stated. Recorded at not addressed rather than partial record because nothing beyond the citations in the output itself was located. Noted for a reader: of all products on this index, this one's output is among the most likely to reach a court, which makes the absence of a disclosure trail more consequential here than for the contract vendors.