Policy

Put the AI Expectation in the Role, Not the Offer Letter

An offer letter should say almost nothing about AI. Keep it to the terms that make an offer: title, pay, start date, reporting line, contingencies, and the at-will language. Put the AI expectation in the written role scope and the first performance plan, where a manager reviews against it and can change it without reopening a signed document. If legal wants a signature on something, use the employee AI policy.

The takeDetailed performance language in an offer letter is usually a symptom of something upstream. The posting was vague, the loop tested a standard the role scope never named, and the letter is the last document before the person starts, so the expectation gets stapled there. Fix the earlier document instead. An expectation you only feel safe stating in a contract is one you never wrote down properly. A binding sentence nobody reviews against buys less than a plain one a manager reads out loud in week one.

Where Olive fits

Open a role and see what the work shows

The same problem shows up one stage earlier, at the assessment: a judgment nobody can point at is hard to hold anyone to. Olive returns six findings on how a person worked with AI, each written or edited by a human reviewer and anchored to a timestamped excerpt, and every released report exports with its rubric, scorer and bank versions attached.

Assess your shortlist

What is the offer letter for, and what does an AI paragraph add?

An offer letter states the terms of the offer: title, salary, start date, reporting line, contingencies such as references and work authorization, and the sentence saying employment remains at will. Everything past that list is either duplication or exposure. A paragraph describing how the person is expected to work with a model is a performance standard sitting inside a contract document.

In 1985 the New Jersey Supreme Court decided Woolley v. Hoffmann-La Roche, a case about an employee manual. The court held that an implied promise in a manual, that an employee will be fired only for cause, may be enforceable against the employer even where the employment runs for an indefinite term 1.

The same opinion says what stops a manual from being read that way.

All the court asked for was a statement in a very prominent position: the employer promises nothing, remains free to change wages and working conditions without anyone's agreement, and keeps the absolute power to fire with or without good cause 1.

But a disclaimer that broad reads badly two paragraphs above a description of how the job is to be done. One sentence says nothing here is a promise. The next describes a standard.

Woolley is New Jersey law, and at-will doctrine varies by state, so ask counsel about your own before drafting either way.

As a drafting habit rather than a rule of law, assume a document handed to an employee is easier to read as a promise than a conversation nobody wrote down.

Where does the AI expectation belong?

In the written role scope first, and in the first performance plan after that. The role scope is the document that says what the job is, and it exists before the posting goes out. It is the document to hand over when somebody asks what the work requires.

Federal disability regulation already counts a written description as evidence.

Under 29 CFR 1630.2(n)(3), the evidence of whether a job function is essential includes the employer's judgment and 'written job descriptions prepared before advertising or interviewing applicants for the job' 2.

The regulation is about disability, not about AI. What it shows is one document the law weighs when it asks what a job requires. The list says it is not limited to its seven items, and the offer letter is not among them.

So the sentence goes where the requirement was first written.

If the posting already carries an AI line, writing an AI-skills requirement that is not legally vague is the same drafting problem one stage earlier.

Copy the wording across rather than paraphrasing it. The standard a person was hired against is the one they can fairly be held to, and a rewrite at each stage is how the two versions drift apart.

Write the expectation as tasks, not as fluency

'AI fluency required' is not a standard anyone can be measured against, and a signature underneath it does not make it one. Name the work instead. Which deliverables get drafted with a model here, what has to be verified before any of it ships, who reads it first, and what the person is still expected to do without a model open.

Three lines usually cover it:

  • Which tasks a model is expected to be used on, and which it is not.
  • What gets checked before the work leaves the desk, and against what source.
  • Who reads it before it ships, and when that review stops.

Each of those is answerable yes or no at a review. 'Fluent' is not.

Setting a defensible bar for good enough at AI in a specific role writes that standard for the requisition rather than for one hire, and the two should not disagree.

The third line has an end date, and the end date is the part that goes unwritten.

Retiring the second reader by decision type rather than by date is a call somebody has to make out loud, and a plan is where a call like that gets recorded. An offer letter cannot carry a review that is supposed to stop.

A letter has no revision history.

Who reopens the AI line after day one?

The manager does, in week one and again at the thirty-day mark, if anyone does. An offer letter is not reviewed after it is signed. It goes into a file and comes out again, if at all, in a dispute. A performance plan gets opened, argued with and revised, which is why it is the right home for an expectation that will move as the tooling moves.

Role clarity is one of three parts of newcomer adjustment in the standard meta-analysis of onboarding.

Talya Bauer and colleagues fitted a model across 70 unique samples of newcomers. They tested whether role clarity, self-efficacy and social acceptance are the path from onboarding to satisfaction, performance, intention to remain and turnover. The results generally supported the model 3.

The abstract, the only part of the paper that could be opened, reports a design and a direction. It gives no effect size.

Timing may matter too, though the evidence is thin.

In a meta-analysis of realistic job previews, previews delivered after hire were associated with a larger drop in voluntary turnover than previews given before hire 4.

The authors say the ranges around the two measurements overlap for voluntary turnover, and five studies stand behind the after-hire one.

Read it as the direction the evidence leans.

The AI line goes into the performance plan as a task, with a date and a reviewer on it, and it moves when the work moves. What actually goes in a 30-60-90 day plan, and who writes it is the longer answer to the same question.

Common questions

Can an offer letter create a binding obligation beyond at-will employment?

It can, depending on the state and the wording, though the leading case concerns a manual rather than a letter. New Jersey's Supreme Court held in 1985 that an implied promise in an employment manual, that an employee will be fired only for cause, may be enforceable even where the employment runs for an indefinite term. The court also said a prominent disclaimer is what defeats it. At-will doctrine and the treatment of employer documents vary by state, so the drafting question belongs with employment counsel. As a drafting habit: the more a document handed to an employee promises, the more work the disclaimer has to do.

Where should an AI expectation be written instead?

In the written role scope behind the requisition, and then in the first performance plan. The role scope is the document that says what the job requires. The ADA's implementing regulation treats a written job description prepared before advertising or interviewing as evidence of what a job's essential functions are. The performance plan is the document a manager actually reopens, which is what an expectation needs if it is going to change as the tooling does. Keep the wording identical to the posting and to what the interview loop tested.

Should the offer letter name the AI tools the company uses?

No. A tool list dates faster than anything else in the letter and binds nothing useful. Approved tools, data rules and disclosure requirements belong in the employee AI policy, which is written to be revised and reissued. If a candidate needs to know the stack before signing, tell them in the closing conversation and send the policy. A letter that names a product the company drops in March is a letter that has to be explained.

What if the candidate asks about AI expectations before they sign?

Send the role scope and answer plainly. The expectation should already be written down by then, in the same words the posting used and the interview loop tested, so the answer is a document rather than an improvisation. Say which work is drafted with a model, what has to be verified before it ships, and who reads it first. If the scope and the candidate's expectations differ, both sides learn it before the start date.

Do EU employers have to put AI literacy in the employment contract?

Article 4 does not ask for it. Article 4 of the EU AI Act has applied since 2 February 2025 and requires providers and deployers to take measures supporting the AI literacy of their staff and of anyone operating AI on their behalf. The European Commission's guidance says the duty carries no obligation to measure an employee's knowledge, no need for a certificate, and no mandatory training. It says organisations can keep an internal record of what they provided. The duty sits with the employer, so a clause in a contract does not discharge it.

References

  1. 1. Woolley v. Hoffmann-La Roche, Inc., Opinion of the Court ✓ Supreme Court of New Jersey, 99 N.J. 284 (1985), via Wikisource, 1985. en.wikisource.org Supports the claim that an employer document can be read as a promise: an implied promise in an employment manual that an employee will be fired only for cause may be enforceable even where the employment is for an indefinite term, and the opinion states that a prominent disclaimer of any promise is what defeats it.
  2. 2. 29 CFR 1630.2 - Definitions ✓ Code of Federal Regulations, via Cornell Legal Information Institute, 2011. law.cornell.edu Paragraph (n)(3) lists written job descriptions prepared before advertising or interviewing applicants among the evidence of whether a job function is essential, which supports the claim that the role description, not the offer letter, is the document the law reads.
  3. 3. Newcomer Adjustment During Organizational Socialization: A Meta-Analytic Review of Antecedents, Outcomes, and Methods ✓ PDXScholar, Portland State University (abstract page for the article in Journal of Applied Psychology, 92(3), 707-721; Bauer, Bodner, Erdogan, Truxillo and Tucker), 2007. pdxscholar.library.pdx.edu Supports the claim that role clarity is part of the newcomer adjustment carrying a hire through, stated as the design and direction of a model fitted across 70 unique samples, with no effect size attached.
  4. 4. Mechanisms Linking Realistic Job Previews with Turnover: A Meta-Analytic Path Analysis ✓ Personnel Psychology, 2011, 64, 865-897 (David R. Earnest, David G. Allen and Ronald S. Landis), 2011. stevenmbrownportfolio.weebly.com Supports the post-hire and pre-hire realistic job preview figures quoted in the article, together with the authors' own statement that the two confidence intervals overlap for voluntary turnover.
  5. 5. AI Literacy - Questions & Answers ✓ European Commission (Shaping Europe's digital future / AI Office), 2026. digital-strategy.ec.europa.eu Supports the date Article 4 began to apply (2 February 2025) and the claim that the duty carries no obligation to measure employees' knowledge of AI and needs no certificate, with an internal record of trainings advised.

5 sources, numbered by first appearance. Every one was opened and checked against the claim it carries. How Olive sources claims

General guidance, not legal advice. Hiring rules differ by state and country and change often; check anything here against your own counsel before you act on it.

Olive assesses how a person works with AI. It does not detect AI-written documents, and it never produces a score, a ranking, or a match percentage for a person. Candidates read the same report the employer reads.

Back to answers

Open your first role Ten attempts a month against a live item bank, with a human-reviewed report on every one.