Attorney-reviewed vs attorney-prepared: the real distinction

Two phrases appear on estate-planning vendor pages, and they describe different things. Attorney-reviewed means a lawyer examined a template before any household used it. Attorney-prepared means a lawyer took on a particular household and produced or revised its documents on its facts. One is a property of the library; the other is a relationship with a client. A firm that cannot say which of the two its clients receive cannot explain the service to them.

This article sets out what template review covers and what it leaves out, what individual representation adds, what to ask a vendor about the review record and about its use of generative AI, and what a diligence file on the question should contain.

What template review covers

The clearest statement of what template review means comes from a state that wrote it into its statute. North Carolina takes a document-generating website outside the practice of law only if, among other conditions, “an attorney licensed to practice law in the State of North Carolina has reviewed each blank template offered to North Carolina consumers, including each and every potential part thereof that may appear in the completed document,” and the provider keeps each reviewing attorney’s name and address on file and provides them to a consumer on written request (N.C. Gen. Stat. § 84-2.2(a)(2)). The same condition appeared in the 2015 consent judgment between LegalZoom and the State Bar, a settlement rather than a ruling, that preceded the statute.

That statute binds one state, but it names the parts of a review claim that can be checked anywhere: who reviewed, licensed where, which templates, which parts of them, and whether a record exists that a consumer can ask for. A review claim that answers those questions is a fact about the library at the date of the review.

Template review happens before any household answers a question. The reviewing attorney sees the clauses and, if the review was done well, the questionnaire logic that decides which clauses appear, but never a particular household’s facts. The attorney’s client is the platform, under an engagement whose scope the two of them set. That does not make the attorney the household’s lawyer, and a household that reads “attorney-reviewed” as “a lawyer looked at my plan” has misunderstood the service.

What individual representation adds

An attorney a household retains considers that household’s facts, advises on them, and prepares or revises documents within the engagement. The engagement can be broad or narrow. ABA Model Rule 1.2(c) allows a lawyer to “limit the scope of the representation if the limitation is reasonable under the circumstances and the client gives informed consent,” so review of one flagged document can be a legitimate engagement, and so can a full plan. The questions to ask are who the attorney represents, what the engagement covers, what it costs, and whether follow-up advice is included.

What “reviewed” cannot tell you

Review of the templates does not settle whether a workflow is lawful for an advisor in a given state. The definition of the practice of law “is established by law and varies from one jurisdiction to another” (ABA Model Rule 5.5, Comment [2]), and the UPL gray zone for financial advisors covers what the authorities on document-preparation software actually require. North Carolina requires review by a locally licensed attorney; Texas conditions its exclusion on a not-a-substitute statement instead.

Review also does not reach the answers. A correct template with a wrong answer produces a wrong document, which is why the questionnaire and its review rules matter as much as the clause text, and a review that covered only the clauses did not cover the logic that selects them. And review has a date. Templates change when statutes change and when the vendor improves them, so the useful question is what triggers another review and how the vendor knows a template has changed since the last one. An integrity check shows whether a source changed after it was locked; the review record shows whether a lawyer looked at it. A diligence file keeps both.

The AI question

On July 29, 2024, the ABA’s Standing Committee on Ethics and Professional Responsibility issued Formal Opinion 512 on lawyers’ use of generative AI. It is ethics guidance for lawyers; it does not evaluate products or address advisors. Its duties still describe what a buyer should ask of the attorneys behind a review claim and of the vendor itself.

On competence, the opinion says the tools “cannot replace the judgment and experience necessary for lawyers to competently advise clients about their legal matters or to craft the legal documents or arguments required to carry out representations,” and that the amount of independent verification a lawyer owes depends on the tool and the task. On confidentiality, it finds that for self-learning tools “a client’s informed consent is required prior to inputting information relating to the representation,” and that “merely adding general, boiler-plate provisions to engagement letters” does not obtain it. On supervision, it applies Model Rule 5.3(b), which requires a lawyer with direct supervisory authority over a nonlawyer to make “reasonable efforts to ensure that” the nonlawyer’s conduct conforms with the lawyer’s obligations, and it extends earlier opinions on outsourcing to AI providers, including reference checks and vendor credentials, the vendor’s security policies, confidentiality agreements, and a forum for relief if the vendor breaches. On fees, it says that where a tool lets a lawyer finish much faster, it “may be unreasonable under Rule 1.5 for the lawyer to charge the same flat fee when using the GAI tool as when not using it.”

For a firm evaluating a platform, those duties translate into the AI questions in the diligence file below: where the platform uses generative AI, whether documents are assembled from fixed templates or written by a model, what client information reaches a model, and what checks a response before a household sees it. A vendor that answers those in writing has given the firm something it can put in the file.

The diligence file

Ask each vendor, in writing, for the review record on the documents the firm’s clients will use: the reviewing attorney’s name, firm, and bar jurisdiction for those templates; the scope of the review, and whether it covered the questionnaire logic and sample outputs; how a change to a template is detected after the review and what record it leaves; what triggers another review; what a household receives if its own document needs individual review, who the attorney represents, and what that costs; where generative AI is used in the product, what information reaches any model, and what checks a response before a household sees it. Ask the same questions of every vendor, keep the answers with the contract, and revisit them at renewal. The reconstruction test explains why those answers belong in the firm’s own record.

How Bancroft handles it

Documents are assembled from fixed templates by code. When a document is generated, no language model writes or edits its text; the questionnaire’s answers select clauses from the library for the household’s state. Template changes follow a controlled release process, and each generated document’s audit record identifies the template version that produced it. The review record for the documents a firm needs is available on request.

Individual review of a household’s document is a separate service. When an answer triggers a blocking review rule, generation stops until the document is released from attorney review, with a notes page listing each rule that fired, the question, the household’s answer, and any directions it wrote; a household can also elect review of its principal documents on its own. The attorney is engaged for the review of that document, and advice on the household’s wider situation is outside it. The review is $299 per document, paid directly to the attorney by the household, or by the firm with the household’s consent. Where the attorney comes in covers that handoff in detail.

Generative AI is used in two places, and neither is part of generating a document: an assistant inside the questionnaire that explains what the question being answered means, which is given that question, and an assistant on the dashboard that answers questions about the household’s own plan data, which is given that data. Both run on a hosted language model, both are instructed never to recommend, and every response passes a server-side screen that replaces any response matching recommendation, situation-analysis, or tax-advice patterns with a fixed reply before the household sees it. To see the review notes page on a sample household, book a demo.