Independent career guide

Attorneys in AI Training: Redlines, Drafts and Review

What micro1 attorney listings describe: simulated redlining, drafting litigation documents, grading AI legal answers, and how task-based pay works.

On this page

This page does two things. First it reports what the attorney listings on micro1’s portal say the work is, taken from their full descriptions on the date in Sources. Then it walks through one practice exercise this guide wrote: a generated legal memo, read twice, under rules invented for the exercise.

The exercise is the blog’s own teaching material. It uses no real law, and it prepares for one of the kinds of work below, not all of them.

Browse all expert opportunities

Opens micro1. You apply and are vetted by micro1, not by this site.

What the attorney listings describe

The attorney descriptions fall into a few recurring kinds of work. Titles overlap heavily, so read the scope section of the one you are considering.

Simulated contract negotiation and redlining. The largest group — listings titled for in-house counsel, general counsel, transactional, contract and funds attorneys — describes performing simulated contract negotiations and redlining exercises, reviewing AI responses to contract scenarios, and creating the evaluation frameworks and grading criteria used to score them. A junior version asks for one to three years of transactional experience and applies the criteria instead of creating them. A related group describes reviewing and redlining commercial agreements against established playbooks and writing the reasoning behind each change.

Drafting litigation documents in a simulated dispute. Three listings describe multi-turn commercial disputes: the attorney drafts motions, briefs, demand letters and discovery turn by turn, under senior direction, responding to opposing counsel’s simulated moves, and writes rubric items alongside each draft saying what a strong response needs. Two of the three state that the work uses only public or synthetic fact patterns.

Evaluating AI-generated legal writing. Several descriptions center on assessing model output: briefs, arguments and memos reviewed for legal accuracy and rigor with written feedback; motions reviewed for persuasiveness; sets of written responses scored and compared, with attention to answers that sound authoritative and are not supported.

Annotating, researching and drafting in a practice area. The practice-area listings — compliance, bankruptcy, corporate, family, immigration, healthcare, civil rights, criminal — describe annotating legal documents, writing scenarios and hypotheticals, drafting sample filings and memos, and critiquing AI-generated analysis in that field.

Reviewing sample contracts for non-lawyers. One listing describes reviewing sample or template contracts, flagging risky clauses, and writing short memos and plain-language explanations, with all material public or synthetic.

So this is not only grading. In most of these descriptions the attorney produces work product — redlines, drafts, rubrics, memos — as well as judging the model’s.

Professional obligations — read this before applying

This page cannot tell you what your professional rules allow or demand. It can tell you what to check, and it sets one rule of its own for the practice material.

  • This page’s rule: build hypotheticals. Bring no client documents, work product or matter facts into practice or into training data. What your confidentiality rules say about former clients and past matters is for you to check; the exercise below needs none of it.

  • Your firm’s policy on outside engagements. Read it before you accept, not after.

  • Whether your professional rules reach this work is yours to check. One listing describes itself in these words:

    This is an evaluative engagement: you will assess and analyze written legal content, not provide legal advice or representation. There is no attorney-client relationship.

    That is one listing’s description of one engagement; the other descriptions we read say nothing on the point, and several involve drafting. Read the contract, look at the concrete activity you would be doing, and check it against the professional rules of the jurisdictions where you are admitted — with your bar’s ethics resources or your insurer if in doubt.

  • Conflicts. Whether evaluating or drafting material in your own practice area raises a conflict with a current client is a question for those same rules and for your firm. Ask it before signing.

The exercise: an offer withdrawn in an invented state

The examples below are illustrative exercises created for this guide, not published micro1 assignments or interview questions. They show the kind of judgment the work needs; they are not a preview of what a project will hand you.

Which work this prepares for. The memo exercise is practice for evaluating AI-generated legal writing: reading a model’s analysis against a rule and a set of authorities, and writing down what is wrong. A shorter exercise near the end is practice for redlining. Nothing on this page practices drafting in a simulated dispute or annotation in a practice area.

Everything legal in this exercise is invented. The State of Arden does not exist. Its reliance rule, its three authorities and the neighboring State of Brell were written for this page. The rule borrows the general shape of doctrines a lawyer may recognize; do not read it as the law of any real jurisdiction, and do not carry its conclusions outside the exercise. Nothing here is legal advice.

The rule, as stipulated. In Arden, a promise made without a contract can be enforced when four elements are met:

  1. the promisor made a clear promise;
  2. the promisee relied on it;
  3. the reliance was one the promisor should reasonably have expected;
  4. the promisee suffered a detriment because of the reliance.

Two more points are stipulated. Relief may be limited to what the reliance cost the promisee. And Arden’s courts have not settled whether relying on an offer of at-will employment can meet element 3.

The authority packet, as stipulated. The grader is given three invented authorities and told that only these may be cited:

  • Harlow v. Dane Mills (Arden): states the four elements.
  • In re Pell (Arden): limits relief to reliance costs.
  • Soto v. Kerr Freight (Brell): a decision of another state, not binding in Arden.

The task given to the model, as stipulated. Answer a member of the public under Arden law, cite only the packet, and say that the answer is not legal advice and that the reader should consult an attorney licensed in Arden.

The facts. A user in Arden writes: a company offered them a job, they resigned from their current position and moved to take it, and the company withdrew the offer before the start date. There was no signed contract. The user asks whether they can recover anything.

What the model wrote

A clean, well-organized memo. It names the reliance rule, lists the elements as promise, reliance and detriment, applies each to the facts, cites three authorities — Harlow, Soto, and a case called Brandt v. Orrin — and concludes that the user will recover the salary the new job would have paid, plus relocation costs.

A skim that would pass it

Headings in the right order, a rule that fits the facts, citations formatted correctly, a clear conclusion. A reader checking whether it sounds like a memo marks it good. On this memo, that check finds nothing.

The second read

Elements, written first

Before reading the analysis, write your own elements checklist from the stipulated rule, then score the memo’s coverage against it. The memo lists three elements. The stipulated rule has four: element 3, that the reliance be one the promisor should reasonably have expected, is missing. On these facts that is the element the outcome turns on, because the exercise stipulates that Arden has not settled it for at-will offers. The elements the memo lists are handled well; the one that decides the matter is never mentioned, and nothing in the text signals the gap.

Citations, one label each

Each citation gets one label: in the packet and supports the proposition; in the packet and does not support it; in the packet and cuts against it; not in the packet. Harlow is in the packet and supports the elements the memo does list. Soto is in the packet, but it is a Brell decision and the memo relies on it as if it bound an Arden court. Brandt v. Orrin is not in the packet at all. Under the task as stipulated, a citation from outside the packet is the one to flag first: the grader cannot check it against anything.

Where the answer should stop

“Will recover” states an outcome as certain when the stipulated rule leaves element 3 open. It also overstates the remedy: under In re Pell, relief may be limited to what the reliance cost, and the salary of a job never started is not a cost of relying. And the memo omits the two statements the task required — that it is not legal advice, and that the reader should consult an attorney licensed in Arden. In this exercise the model answers more than the task allows; writing down exactly where the answer should have stopped is part of the review.

Scoring the memo

The grid below is this blog’s own practice aid, not micro1’s rubric or a hiring criterion. Its levels are cumulative: Strong is Acceptable plus the one thing named after “also”, and a row’s level is the highest one the memo meets in full. A memo with any row at Fail is returned, whatever the other rows say, with the correction each Fail cell names.

CriterionFailAcceptableStrong
AuthorityThe memo cites nothing, or a citation is outside the packet or does not support its proposition. Correct the citationAt least one citation, and every citation is in the packet and supports its propositionAlso: names any packet authority that cuts against the conclusion, or says none does
JurisdictionRelies on a non-Arden authority as binding. Correct how it is usedTreats only Arden authority as bindingAlso: says why Arden law governs these facts
ElementsAn element of the stipulated rule is missing. Correct the analysis, then Re-assess the conclusionEvery element is addressedAlso: identifies the element the outcome turns on
ScopeStates no outcome or states it as certain, claims more relief than the stipulated rule allows, or omits at least one statement the task required. Correct the conclusionStates the outcome as conditional, keeps relief within the stipulated rule, and includes both required statementsAlso: names the fact that would change the answer

The findings land on the four rows: Brandt v. Orrin is an Authority fail; Soto relied on as binding is a Jurisdiction fail; the missing element 3 is an Elements fail; “will recover”, the salary claimed against In re Pell, and the missing required statements are Scope fails. The memo fails all four rows and is returned for four corrections. The Elements fix comes first, because adding element 3 changes what the conclusion can say.

The grid checks what the memo cites, covers and concludes. It does not check whether an element the memo does address is applied correctly; that stays with the second read above.

A shorter exercise for redlining

This one is practice for the redlining listings, and it is also invented. Write two versions of a limitation-of-liability clause for a made-up services agreement between two made-up companies. For each version, write which party it favors and the single change the other side would ask for first, with the reason in one sentence. Then give both versions to a general-purpose chatbot you are allowed to use, ask it the same two questions, and mark where its reasoning differs from yours. The clauses and the companies are yours; nothing in the exercise comes from an agreement you worked on.

micro1 publishes a legal reasoning benchmark among its evaluation environments. The page that names it does not say how it is built or scored, and neither exercise here is drawn from it.

Who should take this on

The requirements differ from listing to listing, and they are specific. Most descriptions ask for a JD and an active license in at least one US jurisdiction, with a stated number of years in the relevant practice; several are open to US-based attorneys only. The simulated-litigation and sample-contract listings accept an active or a lapsed license. One listing accepts a qualifying law degree from a short list of common-law countries. The listings with a firm type in the title mean it: experience at a large or highly ranked firm is stated as mandatory or strongly preferred. One fellowship asks for a documented record of motions won in federal court. Read the qualifications of the listing in front of you; no single bar covers the category.

The descriptions say the work trains and evaluates AI models on legal tasks. Decide how you feel about that before applying; it is a legitimate reason to decline.

Two stages of vetting

There is an AI interview — step two of the four micro1 lists, between applying and onboarding — and separately, micro1 states that “in addition to our automated systems, human recruiters review profiles and match experts with relevant opportunities based on verified skills and project requirements.”

Suggestions from this guide. Two different things can be put forward, and they should never be mixed: your professional experience, which has to be true and limited to what you may share; and a practice exercise, presented as an exercise.

  1. Name your practice area precisely. “Commercial litigation, contract disputes, state and federal courts, eight years” says which listing you fit; “litigation” alone leaves it open.
  2. Describe your experience at the level you may share. Kinds of matter, forum, your role, years. No client names, no matter facts, no work product — and nothing that did not happen.
  3. If you want something concrete to walk through, use an exercise and say so. “This is a practice memo I graded, under invented rules” — the Arden exercise above, or one you build yourself. Do not present an exercise as a matter you handled.
  4. Be exact about admission. Bar, jurisdiction, status, and whether you are currently in practice. The descriptions set these requirements listing by listing.

Terms, referral payments and countries

The cards on the opportunities page show pay as an hourly range. Several of the redlining descriptions then say something different, in these words:

Compensation is task-based, meaning experts are paid per completed task that meets the project specifications. The average handling time (AHT) is approximately 3.5 hours per task, and the effective hourly rate will depend on the number of tasks completed and successfully signed off.

That is the listing itself saying the hourly figure is an outcome, not a rate: payment is per task that is completed and signed off, and what that works out to per hour depends on how long your tasks take against the stated average. A redline that runs past the average may bring your effective hourly figure down; how that extra time is treated is something only the contract says. The other attorney descriptions we read state no compensation structure, so for those the unit of payment is unknown.

Before accepting, ask and get in writing: what one task pays; who signs a task off and against what standard; whether a task that is sent back is paid or can be revised; and whether there is a minimum number of tasks. Do not assume every hour you work is a paid hour until the contract says so.

Leaving rates off this page is the blog’s own choice. Separately, the referral terms micro1 publishes forbid anyone promoting the program from making

promises, guarantees, or representations regarding employment, project placement, compensation, hiring outcomes, interview success, acceptance rates, or future work with micro1.

Worth noting for your own diligence: under those same terms, a referral payment only arises once a referred candidate is placed on a paid project and meets tracked-hours requirements. Placement, not signup, is the event that counts.

Does a referral affect how an attorney’s application is assessed? Nothing in what we read says it does. This site is not part of the vetting. The referral terms quoted above already rule out promises of placement, compensation or acceptance by anyone promoting the program, so a page that offers one is a warning sign.

micro1 states it does not hire residents of: Afghanistan, Belarus, China, Cuba, Democratic Republic of the Congo, Hong Kong, Iran, Iraq, Libya, Macao, Myanmar, North Korea, the Russian Federation, Somalia, South Sudan, Sudan, Syria, Venezuela, Ukraine and Yemen. That list is a snapshot from the date in Sources — check the current page before applying.

What to have ready: a precise practice-area statement, a truthful account of your experience at the level you may share, and one practice exercise you built and can present as an exercise.

Browse all expert opportunities

Opens micro1. You apply and are vetted by micro1, not by this site.

Sources

Programme terms, eligibility rules and reference pages change. Each claim above is dated; verify it on the page it cites before acting on it.