AI Is Changing the Attorney-Client Relationship, but Not Always for the Better

Artificial intelligence has already changed the practice of law. Clients use it to interpret contracts, prepare negotiation points, research legal issues, and draft responses before they ever speak with an attorney. Lawyers use it to summarize records, organize information, test arguments, and improve efficiency. Opposing parties increasingly use it to generate letters, demands, and lengthy responses in seconds.

Some of these changes are genuinely valuable. AI can give people a better vocabulary for discussing a legal issue. It can help an unrepresented party identify questions that otherwise might never have occurred to them. It can make legal information more accessible and reduce the cost of organizing a problem before counsel becomes involved.

But access to legal-sounding language is not the same as access to legal judgment.

That distinction is becoming increasingly important. AI can help a person participate in a negotiation, but it can also create false confidence, obscure the real issues, and turn a productive exchange into a contest over machine-generated talking points. Used thoughtfully, AI can improve the attorney-client relationship. Used as a substitute for judgment, leadership, or direct communication, it can make that relationship, and the underlying legal matter, materially worse.

AI Gives Unrepresented Parties a Seat at the Negotiating Table

Historically, an unrepresented person negotiating against a company or an attorney often faced an immediate disadvantage. The person might understand what happened but not know the legal terminology, how to structure a demand, or which provisions of an agreement deserved attention. Even a relatively straightforward letter could require significant time, confidence, and research.

AI has narrowed that gap.

An unrepresented party can now upload a contract, ask for a summary, identify provisions that appear unfavorable, and generate a professional-looking response. That can be a positive development. Better-informed participants can ask better questions. They may spot an incorrect charge, an overlooked deadline, or an inconsistency that should be addressed. They may also feel more comfortable advocating for themselves rather than accepting the first position presented to them.

This is especially useful in lower-value disputes where retaining counsel may not be economically practical. AI can help a person organize dates, create a chronology, summarize correspondence, and articulate a proposed resolution. It can transform an emotional and disorganized account into something another party can evaluate.

The danger is that AI often makes the user sound more certain than the underlying analysis warrants. A polished letter may cite legal doctrines, characterize contract language as unambiguous, or threaten remedies without accurately accounting for the facts, governing law, procedural posture, or practical economics of the dispute. The result can look like legal analysis while functioning more like an elaborate first impression.

The technology has improved the ability to speak the language of negotiation. It has not necessarily improved the ability to decide what should actually be said, conceded, demanded, or left alone.

Legal Analysis Does Not Occur in a Vacuum

One of AI's most significant limitations is its weak understanding of market context.

A contract provision cannot always be evaluated simply by asking whether it favors one party. Many provisions are intentionally one-sided because of the nature of the transaction, the allocation of risk, industry custom, regulatory requirements, available insurance, bargaining leverage, or the economics of the relationship.

For example, AI may identify an indemnification clause as unusually broad without recognizing that comparable agreements in the same industry routinely allocate the relevant risk to that party. It may recommend deleting a limitation of liability that is standard for the service being provided. It may object to a franchisor's system standards without understanding that brand consistency is central to the franchise model. It may flag a personal guaranty as unfair without recognizing that the counterparty would not enter the transaction without one.

The reverse is also true. AI may describe language as “standard” even when an experienced attorney would know that the provision is negotiable in the current market, inconsistent with the client's objectives, or unusually aggressive for that particular transaction.

Market context includes more than what appears in published agreements. It includes knowing which provisions sophisticated parties actually negotiate, which risks are commonly insured, how deals in that industry are priced, what remedies parties realistically enforce, and when a contractual right is more valuable as leverage than as a claim to be litigated.

That knowledge is rarely captured by the four corners of a document. It comes from repeated exposure to comparable transactions, disputes, counterparties, and outcomes. AI can compare words. It is much less reliable at understanding why the words are there, how the provision operates in practice, and whether pushing the point will improve or damage the deal.

AI Can Miss the Point That Actually Matters

AI tends to be impressive at identifying many possible issues. That is not the same as identifying the important issue.

Legal work requires prioritization. A twenty-page contract may contain dozens of provisions that could theoretically be improved, but only two or three may have a realistic chance of affecting the client's business or changing the outcome of a dispute. An experienced attorney does not merely create a longer list. The attorney distinguishes between technical imperfections, acceptable risks, useful negotiating points, and issues that could materially alter the client's position.

AI often gives similar weight to all detected concerns. It may devote substantial attention to a stylistic ambiguity while overlooking a definition that changes the scope of payment obligations. It may focus on whether a notice provision is perfectly drafted while missing that a claim is approaching a limitations deadline. It may provide a technically plausible interpretation of one paragraph without tracing a cross-reference, exhibit, amendment, or incorporated document that changes the result.

It may also misread silence. In law, the absence of language can be as important as the presence of language. A missing remedy, an undefined approval standard, an omitted survival provision, or a failure to address ownership of a particular asset may create more risk than an obviously unfavorable sentence. AI is generally better at reacting to text than recognizing what a sophisticated agreement should have addressed but did not.

The problem is not simply that AI can be wrong. Lawyers can be wrong too. The more subtle problem is that AI can produce a confident and comprehensive answer that makes it difficult for the user to recognize what was missed. Volume and polish can create the appearance of completeness.

Legal Judgment Includes Consequences Beyond the Legal Answer

Even when AI identifies a legally supportable position, it may not appreciate whether asserting that position is wise.

A client may have the right to issue a default notice, terminate an agreement, refuse consent, demand an audit, or pursue emergency relief. Whether the client should exercise that right depends on more than the text of the agreement. It may depend on the client's long-term objective, the financial condition of the other party, the availability of evidence, reputational considerations, the cost of enforcement, and the likely response.

Sometimes the strongest legal argument should be made immediately. Sometimes it should be preserved but not emphasized. Sometimes a narrow business proposal will accomplish more than a detailed explanation of every available claim. Sometimes sending an aggressive letter creates evidence, hardens positions, alerts the other side to a weakness, or makes a practical resolution less likely.

AI does not bear the consequences of its recommendations. It does not sit with the client after a relationship collapses, explain the cost of discovery, assess the credibility of a witness, or revise strategy when the other party responds unpredictably. It can generate options, but it does not assume responsibility for choosing among them.

That responsibility remains at the heart of legal counseling.

When AI Slop Enters the Attorney-Client Relationship

The term “AI slop” is increasingly used to describe content that is polished, lengthy, and superficially responsive but lacks meaningful thought. In legal work, it often appears as a multi-page response that repeats general principles, raises every conceivable objection, and never directly answers the question presented.

This can make working with clients more difficult. A lawyer may ask a focused factual question: Who approved the change? Was the notice sent? Did the customer sign the revised agreement? Instead of an answer, the lawyer receives an AI-generated discussion of contract interpretation, possible defenses, equitable doctrines, and hypothetical risks.

The lawyer must then spend time separating actual facts from suggested arguments. That increases cost and can delay decisions. Worse, the generated response may unintentionally introduce inaccurate facts or legal conclusions into the client's narrative. Once repeated in correspondence, those inaccuracies can undermine credibility or complicate later testimony.

AI can also interfere with candid communication. Effective representation requires clients to explain what happened, what they want, what they are willing to accept, and where the business has made mistakes. A generated memo cannot replace leadership direction. The attorney still needs to know whether the client wants to preserve the relationship, exit it, establish a precedent, limit cost, or pursue the matter regardless of expense.

Without that direction, more words do not create a strategy.

When AI Slop Reaches the Other Side

The same problem is increasingly visible in communications with opposing parties and counsel.

A direct question receives a response that is several pages long but never says yes or no. A narrow contract dispute produces a catalogue of unrelated doctrines. A settlement proposal is met with a generic reservation of every conceivable right. Authorities are cited for broad propositions without explaining how they apply to the disputed facts.

This creates the appearance of engagement without the substance of engagement.

Negotiations move forward when parties identify their actual disagreement, exchange the information needed to evaluate it, and make decisions. AI-generated correspondence often does the opposite. It expands the number of words while avoiding commitment. It can cause recipients to spend time responding to arguments that no decision-maker has actually adopted. It can also make it difficult to determine whether the sender has authority, understands the position being asserted, or is willing to resolve the issue.

There is also a credibility cost. If a letter misstates the agreement, cites an inapplicable rule, or makes a threat that the sender plainly cannot carry out, the recipient may discount the rest of the communication. A strong point can be lost inside ten weak ones.

More fundamentally, negotiation requires human judgment. Someone must decide what matters, what can be conceded, and what outcome is acceptable. AI can draft a counterproposal, but it cannot supply the business decision behind it. When a party uses generated language to avoid making that decision, the technology becomes a barrier rather than a tool.

The Frustration of Feeling Like No One Is Actually Listening

There is also a human cost that is easy to underestimate: receiving an AI-generated response can be genuinely frustrating.

People usually reach out to an attorney, client, employee, business partner, or opposing party because they want someone to consider a specific question and make a decision. When the response reads like a generic summary of the topic, the recipient may reasonably conclude that no one actually reviewed the message, engaged with the concern, or exercised independent thought.

That reaction is not necessarily hostility toward AI. The frustration comes from the mismatch between the communication and the moment. A person may have spent hours explaining a problem, identifying the relevant provision, and asking two direct questions. Receiving five polished pages that restate the background but avoid those questions can feel dismissive. It shifts the burden back to the recipient to extract an answer from content that was supposed to provide one.

This can be particularly damaging in the attorney-client relationship. Clients are not only paying for written output. They are paying for attention, judgment, and a recommendation grounded in their objectives. If advice appears to have been generated without meaningful review, the client may begin to wonder whether counsel understands the business, has identified the controlling issue, or is willing to take ownership of the recommendation.

The same dynamic affects negotiations. Opposing counsel may be less willing to engage constructively if each direct proposal produces a generic catalogue of objections. Business leaders may lose patience when a request for approval results in an abstract discussion rather than a decision. Employees may feel ignored when a fact-specific concern receives language that could have been sent to anyone.

AI-assisted writing is not inherently impersonal. It becomes impersonal when the final communication contains no visible evidence of human judgment. A useful response should show that the sender understood the question, evaluated the particular facts, and decided what to say. Sometimes that means a carefully reasoned letter. Sometimes it means three direct sentences.

Before sending AI-assisted work, the author should be able to answer a simple question: What did I personally decide after reviewing this? If the answer is unclear, the communication probably is too.

Why AI Often Struggles With Redlined Documents

Redlined agreements present a separate and particularly consequential problem. AI tools often have difficulty distinguishing among the original language, deleted language, inserted language, comments, formatting changes, and the current operative text.

A redline is not simply a document containing more words. It is a visual and structural record of change. Deleted text may remain embedded in the file. Insertions may be stored as tracked revisions rather than ordinary text. Comments can contain proposed language that was never accepted. A comparison document may show changes between versions without clearly establishing which version is controlling. The meaning can also depend on whether revisions are displayed as final, original, simple markup, or all markup.

AI systems may flatten these layers when extracting the document. As a result, they can treat deleted language as though it remains in effect, read both the old and new versions of a sentence together, or mistake a comment for contract language. They may claim that a conflict remains even though the provision was revised to resolve it. They may identify a missing term that was added in tracked changes, or report that two sections are inconsistent because the model is comparing an obsolete version of one section with the revised version of another.

This problem is common when asking AI to confirm whether requested revisions were made. The tool may accurately recognize the subject matter but fail to determine the status of the change. It can say an issue “still appears” in the agreement because it sees the deleted text in the document data. Conversely, it may assume a proposed insertion was accepted even though it remains only a comment or unaccepted revision.

Cross-references make the problem harder. A defined term may have been revised in one location, moved to another, or changed differently in separate versions. A human reviewer can inspect the markup, determine the baseline and revised documents, accept or reject changes in a working copy, and then analyze the clean operative language. AI may instead compare an unstable mixture of text states and confidently report conflicts that do not exist.

That does not mean AI cannot assist with redlines. It can help summarize clearly identified changes, create a negotiation log, or compare clean versions after the documents have been normalized. But it should not be trusted blindly to determine whether every issue was resolved. A reliable review generally requires clearly identifying the original and revised versions, confirming how tracked changes were extracted, reviewing the clean post-revision text, and manually verifying material provisions and cross-references.

When the distinction between old language and operative language matters, document state is part of the legal analysis. If the tool cannot reliably determine that state, its conclusions may be polished but fundamentally misdirected.

The Lawyer's Role Is Changing, Not Disappearing

AI will continue to make basic legal information and drafting assistance more accessible. Lawyers should not treat that development as a threat. A client who has used AI to organize documents, create a chronology, or identify preliminary questions may arrive better prepared and use legal time more efficiently.

The lawyer's value, however, increasingly lies in the work AI performs least reliably:

• identifying which facts and legal issues actually matter;

• placing contract language within the relevant market and industry context;

• recognizing omissions, inconsistencies, and downstream consequences;

• assessing credibility, leverage, cost, timing, and enforcement risk;

• giving clear recommendations rather than a menu of abstract possibilities;

• helping the client make a decision and take responsibility for it; and

• communicating with enough precision and restraint to move the matter forward.

This also changes what clients should expect from counsel. The best legal advice may not be the longest answer. It may be a short explanation that identifies the controlling issue, describes the realistic options, and recommends a course of action. Efficiency should not be confused with superficiality, just as volume should not be confused with sophistication.

A Better Way to Use AI in Legal Matters

AI is most effective when used as an assistant to human judgment rather than a substitute for it.

Clients can use it to organize information, prepare timelines, summarize their own objectives, and generate a list of questions for counsel. Lawyers can use it to accelerate routine work, test alternative phrasing, and locate issues that deserve further investigation. Unrepresented parties can use it to make their communications clearer and more structured.

But several guardrails are essential.

First, distinguish facts from generated analysis. Tell your attorney what actually occurred before providing conclusions about what it means.

Second, answer direct questions directly. If counsel asks for a date, document, decision, or objective, provide it. An AI-generated memorandum is not a substitute.

Third, verify legal authorities, quotations, deadlines, and contractual references. AI can invent sources, conflate rules, and overlook jurisdiction-specific requirements.

Fourth, ask whether the proposed communication advances the objective. A response should not be sent merely because it sounds forceful or comprehensive.

Finally, preserve human accountability. The person or business involved must decide the desired outcome, acceptable risk, and settlement authority. Those decisions cannot responsibly be outsourced to a language model.

Better Tools Still Require Better Judgment

AI has given more people access to legal information and the ability to express their positions in professional language. That is meaningful progress. It can help unrepresented parties participate more effectively, help clients prepare more efficiently, and help attorneys devote more attention to higher-value analysis.

But AI does not understand a client's business merely because it can summarize the client's contract. It does not understand a market merely because it can describe common provisions. It does not exercise leadership merely because it can produce a decisive tone. And it does not resolve a dispute merely because it can generate a long response.

The attorney-client relationship has always depended on trust, candor, judgment, and clear direction. AI can support each of those things when used carefully. When used to avoid them, it produces more content but less communication.

The future of legal work is not a choice between attorneys and artificial intelligence. It is a choice between thoughtful use of powerful tools and the uncritical production of legal-sounding noise. The distinction will often determine whether AI makes a legal matter more efficient or simply makes everyone work harder to reach the real issue.