How to Use AI in Criminal Law Marketing

1.Build a Clear Marketing Strategy

I learned criminal-law marketing the traditional way: by spending money, making mistakes, and eventually giving the mistakes names like 'systems.'

Start by giving AI basic information about the firm: location, practice areas, ideal clients, strongest case types, fee range, competitive advantages, and geographic service area. Ask it to identify the firm's best marketing opportunities and turn them into a simple 90-day plan.

For example, a Massachusetts criminal defense firm may want separate campaigns for OUI defense, clerk-magistrate hearings, domestic violence charges, license suspensions, and serious felony cases. AI can help rank those opportunities by likely search demand, competition, fee value, and the lawyer's actual experience.

The lawyer should then choose a small number of priorities. A useful rule is to focus on one major website improvement, one review-building system, one referral campaign, and one content campaign at a time. AI is most useful when it turns broad ideas into a weekly checklist with specific tasks, deadlines, and owners.

2.A Shrinking Crime Market Changes the Economics

"AI may discover new medicines and improve transportation, but its stubborn refusal to manufacture additional arraignments remains a serious defect."

Crime has generally been trending downward nationally over the long run, even though the pattern is uneven by city, offense, and year. For criminal defense lawyers, that can make some local markets more competitive. There may be fewer cases to fight over, more firms chasing the same clients, and greater pressure on fees, intake speed, reviews, and differentiation.

AI has enormous potential to expand the economic pie in many industries. It may help companies create new products, discover treatments, improve logistics, reduce waste, and make knowledge work faster. Tragically, it is unlikely to increase the crime rate so criminal defense lawyers can make more money. This obviously calls into question the value of AI, at least from the narrow perspective of a lawyer hoping for a larger criminal docket.

The practical response is not to wish for more crime. It is to become better at winning a larger share of a tougher market. Firms can use AI to improve intake, follow-up, local content, referral systems, review analysis, pricing experiments, and the speed at which they respond to changing demand. A shrinking or slower-growing market rewards firms that measure carefully, build trust, and operate efficiently.

3.State Systems Create Different Private Markets: Michigan, Maine, and Maryland

The best market is not the one with the most crime; it's the one with clients who can both find you and pay you, which is a much shorter list.

Criminal-defense marketing cannot be evaluated only by population or total criminal filings. The structure of the public-defense system, the number and location of courts, local income levels, typical retainers, payment expectations, and the share of defendants who qualify for appointed counsel all affect the size and economics of the private market. Michigan, Maine, and Maryland illustrate three meaningfully different environments.

Michigan is the largest of the three states with a population of roughly 10 million, and has several substantial metropolitan markets, including Detroit, Grand Rapids, Lansing, Ann Arbor, and Flint, along with many county-based courts. That creates a larger total pool of criminal matters and more room for firms to specialize by county, charge, or client type. Michigan's indigent-defense system is locally administered under statewide standards and substantial state funding through the Michigan Indigent Defense Commission. Because assigned-counsel arrangements and compensation differ by local system, the relationship between appointed work and private practice can vary sharply from one county to another. Private fees also vary widely: urban felony, drunk-driving, and professional-license cases can support meaningful retainers, while lower-level matters in less affluent areas may be highly price sensitive.

Maine, with a population of roughly 1.4 million, is much smaller, more rural, and geographically spread out. Its private criminal-defense market is therefore thinner outside Portland, York County, Bangor, and a few other population centers. Travel time and court coverage matter more, and a lawyer may need a broader geographic footprint or a wider mix of case types. Maine historically depended heavily on private assigned counsel and is now building a hybrid system with employed public defenders while continuing to use rostered private lawyers. In 2025 and 2026, Maine documents described a goal of shifting a substantial portion of the indigent caseload to employed defenders while still relying on private assigned counsel; the authorized assigned-counsel billing rate was reported at $150 per hour. That system can provide paid case volume to participating lawyers, but it also competes for attorney capacity and can affect how much time remains for higher-fee private cases.

Maryland has a population between Michigan and Maine (roughly 6.2 million) but is denser, wealthier in many suburbs, and heavily influenced by the Baltimore-Washington corridor. The private market includes Baltimore City, Baltimore County, Montgomery County, Prince George's County, Anne Arundel County, and other distinct local markets. Maryland operates a statewide Office of the Public Defender with salaried staff lawyers; private panel attorneys are used mainly when conflicts or other needs arise. This means private lawyers generally cannot treat routine appointed work as the same broad supplemental revenue stream found in a state that relies more heavily on assigned private counsel. On the other hand, affluent suburban markets, serious traffic and DUI work, federal exposure, and clients with employment or security-clearance concerns can support higher private fees.

There is no dependable official statewide "average criminal-defense fee" for any of these states. Fees depend on charge severity, county, lawyer reputation, expected motions, trial risk, and the client's ability to pay. As a practical marketing matter, Maine will often offer fewer leads and smaller local pools but potentially less competition in underserved areas. Michigan offers greater volume and more geographic niches, but also more firms and substantial variation between counties. Maryland can support strong fees in dense and affluent markets, but competition is intense and the statewide public-defender structure absorbs much of the indigent caseload directly.

The AI lesson is that state-level market analysis must come before copying another firm's strategy. A system trained on Maryland assumptions may overestimate private demand in rural Maine. A Maine lawyer's court-coverage model may be inefficient in metropolitan Michigan. Marketing plans should therefore track county-level filings, appointed-counsel eligibility and payment rules, local fee ranges, courthouse geography, demographic capacity to pay, and the percentage of leads that become viable private clients. The best market is not always the state with the most cases; it is the combination of sufficient case volume, clients able to retain counsel, manageable competition, and a public-defense structure that leaves a meaningful private segment.

State-system note, July 2026: Michigan's indigent-defense funding and standards are administered through the Michigan Indigent Defense Commission; Maine is continuing its transition toward a hybrid employed-defender and assigned-private-counsel model; and Maryland's statewide Office of the Public Defender uses private panel attorneys principally for conflicts and related needs. Payment rates, eligibility rules, staffing, and local practices can change, so they should be verified before making business decisions.

Ballpark case volume and private share, quick estimates only, not researched figures: Michigan sees on the order of 300,000 criminal case filings a year, of which roughly 15% - about 45,000 cases - go to privately retained counsel. Maine sees something like 40,000 filings a year, of which roughly 12% - about 5,000 cases - go private, reflecting its smaller, more rural caseload and thinner base of clients able to retain counsel. Maryland sees on the order of 190,000 filings a year, of which roughly 20% - about 38,000 cases - go private, the highest private share of the three, consistent with its denser, wealthier suburban markets.

Of the appointed share in each state, the split between full-time public defenders and assigned or contract counsel also varies. Michigan's roughly 85% appointed share runs about 35% through full-time public defender offices concentrated in the largest counties and about 50% through assigned or contract counsel paid by the state per case. Maine's roughly 88% appointed share still runs mostly - about 78% - through assigned private counsel paid per case, with its new state defender office covering only about 10%. Maryland's roughly 80% appointed share flips that pattern: about 65% runs through full-time public defenders, with panel attorneys paid for conflicts picking up the remaining 15%.

The same rough pattern tends to hold nationally: most privately retained criminal cases resolve in a handful of court dates - arraignment, maybe a pretrial conference or two, then a plea - and command flat fees in the roughly $1,500-$3,500 range. A smaller share of cases, the ones that go to trial or involve serious felonies and multiple motions, command fees well above that range along with far more court time. None of these percentages or fee ranges are researched figures; they are ballpark orientation numbers, useful for sizing a market roughly and worth replacing with real local data before making a real business decision.

In many states, a meaningful share of that private money is concentrated in OUI/DUI cases along with reckless-driving and speeding charges - the everyday traffic-adjacent offenses that generate a steady stream of paying clients largely independent of the broader crime rate. That concentration is worth watching, because it carries a long-term risk none of the numbers above capture: at some point a meaningful percentage of cars on the road will be autonomous, and a firm whose private caseload leans heavily on OUI and reckless-driving work should expect that revenue base to shrink as human driving itself becomes less common. Nobody knows the timeline, but the direction is not really in doubt.

That looming shift is one of the better arguments for staying flexible rather than building a high-overhead practice around long fixed leases and associates carrying high fixed salaries. A firm locked into that structure needs roughly the same revenue every month just to break even, which is a bad position to be in if OUI volume - or any other single case type - declines. Paying associates a percentage of the cases they bring in or work, rather than a flat salary, shifts some of that risk onto a variable cost base: as revenue rises and falls with caseload, payroll moves with it instead of staying fixed. A solo practitioner has an even simpler version of the same flexibility - when demand for one type of case falls off, the time that frees up can go toward learning an adjacent area of law rather than sitting idle waiting for the phone to ring.

These markets also interact rather than staying in separate lanes. It is common for solo and small-firm lawyers to combine criminal defense with personal injury and divorce work, shifting time between practice areas week to week depending on which caseload is active and which is paying the bills. A firm marketing itself narrowly as criminal-defense-only may still be competing for that same lawyer's limited hours against whichever practice area is busier that month. An AI marketing system should account for that reality - a lawyer running three practice areas needs a system that flags which one deserves attention this week, not a plan that assumes full-time dedicated capacity to criminal work alone.

Marketing itself should look different depending on which of these lawyers you are. A pure private-only practice can market on price, results, and availability without needing to explain a mixed caseload, and its whole system - website, reviews, intake - can stay narrowly focused on paying criminal clients. That lawyer benefits most from referral relationships, which convert well and are worth systematizing - regular outreach to past clients and referring professionals can be automated instead of left to chance. Referrals alone are a fragile foundation, though, since they depend on other people's memory and goodwill. Strong local SEO is worth building on top of referrals even for a purely private practice: it captures the direct-search demand referrals miss, and it is helpful in its own right today while also being likely to feed directly into how AI systems generate their own local-lawyer recommendations as more of that search volume moves off a traditional results page. A lawyer running a mix of appointed and private work needs a system that keeps those two flows separate operationally (appointed cases don't need a marketing funnel at all) while still treating the appointed caseload as a reputation asset: a satisfied appointed client can leave the same five-star Google review as a paying one, and a review does not know or care how the fee was set. That makes it worth deliberately asking appointed clients for a review once a case resolves well - it costs nothing, and it builds the same review base that drives private intake, effectively subsidizing private-practice marketing with cases the state already paid for. A lawyer running a mix of practice areas needs a third kind of system entirely: one that can shift marketing spend and content production toward whichever practice area has slack capacity that month, rather than a single evergreen criminal-only plan.

4.AI Makes It Easier to Start - and Disrupt - Businesses

The most dangerous competitor in your county is a hungry solo with a laptop, a $20 subscription, and no payroll to meet.

AI is making it easier to start a business. A small firm can now produce a basic website, research a market, draft content, create intake materials, organize leads, analyze competitors, and build simple internal systems with far less money and staff than would have been required a few years ago.

That lowers the barrier to entry and makes it easier for new firms to challenge established competitors. In criminal-law marketing, a small practice can move quickly, test new pages, focus on overlooked courts or case types, personalize follow-up, and change direction without waiting for layers of approval. A smaller firm should treat speed, flexibility, and direct knowledge of clients as advantages rather than trying to imitate a large organization.

Large firms and established marketing companies face the opposite challenge. Their brand, staff, data, and existing client base remain important advantages, but those strengths can become liabilities if the organization is slow to adopt better tools or continues charging high prices for work that AI has made faster and cheaper. Incumbents need to identify which parts of their service are truly valuable and which parts can now be replaced, automated, or delivered differently.

The practical lesson is simple. If you are small, use AI to compete above your weight. Build quickly, learn from the market, keep overhead low, and exploit opportunities larger competitors ignore. If you are large, assume that someone smaller is trying to reproduce the easiest parts of your business at a lower cost. Use AI to improve the core operation before an outside competitor does it for you.

AI will not eliminate the advantages of reputation, judgment, relationships, local knowledge, and proven results. It will, however, make it easier for capable new entrants to look professional and operate efficiently. Every firm should decide whether it intends to use that change as an opportunity or wait until it becomes a threat.

5.Ask How a New Competitor Would Beat Your Firm

If you can't explain how you'd beat your own firm, relax — eventually a 28-year-old with a laptop will demonstrate.

Imagine that you were starting from zero today and wanted to beat your own firm. A new competitor might choose a neglected niche, create better educational content, answer calls faster, use social media more effectively, or serve a geographic area that established firms overlook.

Every established firm should perform this exercise regularly. The answer may reveal weak intake coverage, stale website pages, overdependence on one advertising channel, or practice areas that no longer have the same economics. AI can help generate possible attacks, compare competitors, and turn the most credible threats into defensive projects.

6.Pay-Per-Click Advertising in Criminal Law

Good luck with that. 😂

People pay too much for clicks in most markets now. If you pay $25 for a click and 10 percent of the people call, and you convert 10 percent of the calls, you have paid $2,500 for 10 calls that take up time and get one case. You probably aren't making money on that.

The escalation isn't random. Cost-per-click for terms like "DUI lawyer" and "criminal defense attorney" has climbed steadily for over a decade, and a lot of that pressure comes from bidders who aren't playing your game. National aggregators and large personal-injury shops that added criminal as a side door can absorb a worse per-case return than you can, because they're buying brand and cross-sell, not a single signed case. When you're in the same auction as someone who's fine losing money on the click itself, you're not competing on the same math, and the auction doesn't care.

The deeper problem is the ceiling. A flat $2,000–$5,000 drunk driving or assault fee has a hard limit that click cost doesn't respect. A personal-injury click can be worth $50,000 or more in eventual fee, so a PI firm can rationally pay $150–$300 for it. A criminal-defense click sits on top of a fee structure that hasn't moved nearly as fast as CPC has. Same channel, same auction, wildly different math depending on which kind of buyer you are.

PPC does still work in a few narrow lanes: high-fee practice areas within criminal work (federal, white collar, serious felony), a firm with zero organic visibility that needs bridge traffic while SEO ramps up, or intake so fast and well-trained that conversion sits meaningfully above market average. If you're not in one of those three, you're probably not the exception.

The alternative isn't "spend nothing." It's putting the same dollars into channels where the marginal lead doesn't cost the same as the first one — local SEO, reviews, referrals, authority content. Slower to start, but the hundredth lead is nearly free instead of costing exactly what the first one did.

So: good luck with that — unless you've quietly become one of the three exceptions above, in which case, congratulations, spend freely.

7.Build Authority Assets That AI Can Find and Cite

AI search does not rely only on a law firm's website. Recommendation engines compare information across websites, directories, local publications, books, interviews, FAQs, reviews, and other public sources. A firm that is mentioned consistently in credible places is easier for both Google and AI systems to understand.

"Publishing your 214th generic blog post on 'What to Do After an Arrest' may build authority, although primarily with the intern who has to upload it."

Criminal defense lawyers should therefore create authority assets rather than endless generic blog posts. Useful assets include detailed answers to real client questions, state-specific guides, court and charge pages, published books or short handbooks, podcast appearances, local media quotations, professional presentations, and citations from relevant organizations. These materials should demonstrate genuine knowledge instead of merely repeating keywords.

The practical goal is to make the firm easy to verify. When several independent sources connect a lawyer with a particular state, court, charge, or niche, an AI system has more evidence for recommending that lawyer. Firms should periodically test the same client questions in several AI products and track whether the firm appears, which competitors appear, and which sources the systems seem to trust.

8.Going Forward: AI Recommendations, Reviews, and Uncertainty

The safest prediction in this book: whatever the ranking formula is today, it will be something else before you finish reading about it.

It is almost impossible to know exactly how AI will evolve. Models, search products, advertising systems, and consumer behavior are changing too quickly for any long-term prediction to be treated as certain. The best approach is to build flexible systems, preserve useful data, and remain ready to change tools and tactics as the technology develops.

My guess is that it will matter more and more which lawyers large language models recommend when a person asks for help with an OUI, domestic violence charge, clerk-magistrate hearing, license suspension, or other criminal matter. That may create a new layer of visibility alongside traditional Google rankings, paid advertising, referrals, and legal directories. No one yet knows exactly how those recommendations will be generated or how stable they will be.

Google reviews already have substantial value because they provide public evidence that a firm is active, responsive, and trusted. Their value is likely to increase as AI systems use reviews and other reputation signals to understand which businesses appear legitimate and useful. Reviews may influence not only what a potential client sees on Google, but also how AI systems describe and compare lawyers.

At the same time, AI will become better at distinguishing strong, detailed, authentic reviews from reviews that appear manufactured, repetitive, purchased, or otherwise gamed. A large review count may still matter, but review quality, specificity, timing, reviewer patterns, and consistency with information from other sources are likely to matter more over time.

The durable strategy is therefore straightforward: do good work, ask appropriate clients for honest reviews, respond professionally, maintain accurate public information, publish useful state-specific content, and avoid shortcuts that may later be detected. Firms should track how they appear in Google and in major AI systems, but they should not assume that any current ranking formula will remain unchanged.

9.Match Tools and Management Systems to the Firm's Stage

Buying more software to fix a management problem is like buying a nicer briefcase to fix a losing case.

The right system depends on the firm's stage of growth. A solo lawyer needs simple delegation, reliable intake, and a small number of useful measurements. A larger firm needs clearer roles, stronger onboarding, written accountability, management by data, and leaders who can supervise other leaders.

AI can help at every stage, but adding more software is not the same as building a better business. The firm should first identify the bottleneck: unanswered calls, weak follow-up, poor training, inconsistent content, unclear finances, or excessive dependence on the owner. It should then use the smallest tool or automation that solves that specific problem and assign one person to own the result.

The long-term objective is a firm that can operate without every decision passing through the owner. Dashboards, documented procedures, 30-, 60-, and 90-day reviews, clear job expectations, and AI-supported training can make performance more visible and predictable. Technology works best when it supports a defined management system rather than becoming another collection of subscriptions that nobody fully uses.

10.Measure Profitability by Practice Area, Not Just Total Revenue

Law-firm benchmarks are often too broad to be useful. Criminal defense, immigration, family law, and personal injury firms have different fee structures, staffing needs, case lengths, collection risks, and marketing economics. A percentage that looks healthy for one practice can be disastrous for another.

Criminal defense firms should use AI to separate revenue, marketing cost, intake labor, attorney time, payment-plan losses, referral fees, and collection problems by case type and market. An OUI campaign may generate many leads but require expensive advertising. A clerk-magistrate or license-suspension niche may produce fewer calls but better margins. A serious-felony practice may support larger fees while tying up attorney time for much longer periods.

The important question is not simply whether revenue increased. It is whether the firm acquired the right cases at a sustainable cost and converted them into collected fees without overwhelming the lawyers or staff. AI can help clean the data, compare periods, flag unusual changes, and model different assumptions. Management still has to decide which practice areas deserve more investment, which should be repaired, and which should be reduced or abandoned.

"Top-line revenue is an excellent number, especially when no one asks what it cost, how long it took, or whether the client actually paid."

11.Create Better Website and Search Content

AI can quickly produce first drafts for service pages, city pages, FAQs, blog posts, attorney biographies, case-result summaries, and Google Business Profile posts. The key is to avoid publishing generic text. Every draft should be revised to include local court knowledge, real client questions, actual case experience, and the lawyer's natural speaking style.

A strong prompt should identify the audience, legal issue, location, desired tone, and action the reader should take. For example: "Draft a plain-English page for someone charged with OUI in Quincy District Court. Explain the first court date, license consequences, common defenses, and why early representation matters."

AI can also compare existing pages, find missing topics, create internal-link suggestions, and rewrite confusing sections. Final content must be checked for accuracy, advertising-rule compliance, confidentiality, and unsupported claims.

12.Improve Intake and Follow-Up

A missed call is not a lost lead; it is a donation to whichever competitor answers their phone.

Many firms lose good cases because calls are missed, follow-up is slow, or information is not organized. AI can help build intake scripts, text-message templates, email sequences, call summaries, and lead-scoring systems.

After a consultation, staff can use an approved AI workflow to summarize the prospect's charge, court, next date, urgency, fee discussion, concerns, and follow-up steps. The system can then produce a short task list. Sensitive information should only be entered into tools approved by the firm, and lawyers should avoid placing confidential client details into public AI systems.

AI can also help analyze why leads do not hire. Common categories include price, slow response, lack of trust, wrong practice area, inability to reach the prospect, or hiring another lawyer. Tracking these reasons gives the firm a better basis for improving scripts, staffing, and fee presentation.

13.Strengthen Reviews

Online reviews are often one of the strongest trust signals in criminal-law marketing. AI can help create a lawful, consistent process for requesting reviews from appropriate former clients. It can draft short text and email requests, staff instructions, and responses to positive or negative reviews.

The firm should never create fake reviews, pressure clients, reveal confidential facts, or offer improper incentives. Review responses should remain brief and avoid confirming that the reviewer was a client when that could create a confidentiality issue.

14.Referrals and Reputation

AI can also organize referral marketing. It can help create lists of former colleagues, civil lawyers, family-law attorneys, therapists, treatment providers, and other professionals who may encounter people needing criminal representation. It can draft personalized outreach, newsletter topics, and follow-up reminders while keeping genuine relationships at the center.

15.Measure Results and Improve the System

A dashboard with twelve colors and no reliable intake data is not analytics; it is stained glass.

AI becomes more valuable when the firm feeds it clean, nonconfidential marketing data. Useful measures include:

1. Calls
2. Qualified leads
3. Consultations
4. Retainers sent out
5. Signed cases
6. Average fee
7. Source of lead
8. Court
9. Charge type
10. Response time
11. Review growth
12. Website traffic
13. Cost per signed case

You don't need all 13, but it is my lucky number. You might want to start with calls, leads, signed cases, average fee, but really start with a few that work for you. I don't know what matters for your practice, guess what you don't either and won't until you track them. I track and analyse insane amounts of things in my life and if it doesn't drive almost everyone around me nuts I am not tracking enough. It's possible that approach wouldn't work for everyone.

Each month, the firm can ask AI to summarize what changed, identify unusual patterns, and suggest experiments. It may show that one court page generates many calls but few signed cases, while a smaller license-suspension campaign produces higher fees and better conversion. It may also reveal that calls answered within five minutes convert far better than calls returned several hours later.

The goal is not to automate every decision. The goal is to create a repeatable system in which AI handles first drafts, summaries, comparisons, and routine analysis, while the lawyer controls strategy, ethics, accuracy, client relationships, and final judgment.

16.Marketing Only Works When Intake and Reputation Convert the Lead

Paying for a lead and answering it Monday is a generous way to support another lawyer's weekend revenue.

Rising paid-ad costs can make flat-fee criminal cases difficult to acquire profitably. Firms should know the true cost of each signed case, avoid treating broad networking as automatically valuable, and protect the leads they already pay to generate. Weekend and evening intake coverage can matter because a frightened prospect may hire the first qualified firm that responds.

Reputation is part of the same conversion system. Google reviews can disappear, be filtered, or become the target of suspicious attacks. A resilient firm earns reviews steadily, documents unusual activity, avoids gimmicks, and does not depend on a sudden burst of testimonials. Traffic, intake speed, trust, and follow-up must work together; generating more leads alone does not solve a weak conversion process.

17.Make the Lawyer More Visible as Generic Content Multiplies

AI will make ordinary legal content cheap and abundant. Every firm can now produce service pages, FAQs, newsletters, and social posts. That means merely publishing more words is less likely to create a durable advantage.

The stronger opportunity is to make the actual lawyer visible. Firms should use AI to help capture real stories, courtroom observations, local knowledge, client questions, and opinions that sound like a particular attorney rather than a generic marketing department. Video, podcast clips, short explanations, books, interviews, and detailed local guides give prospects and recommendation systems more evidence that a real person has genuine experience.

AI should therefore support authenticity rather than flatten it. It can organize transcripts, create drafts, identify reusable clips, and turn one strong conversation into several useful assets. The final material should still preserve the lawyer's judgment, humor, vocabulary, and point of view. As generic content becomes nearly free, recognizable expertise becomes more valuable.

"If the content could have been written by any lawyer, any consultant, or a particularly cautious toaster, it is not yet personal enough."

18.Use AI to Make People Better, Not Just to Replace Tasks

My first year practicing criminal law, I was a public defender. I tried more than two hundred cases that year. One day I had thirty-four cases on for trial and actually tried eleven of them — eleven trials, one day, one lawyer. That is not a training program. That is a system that needed bodies in courtrooms, and I was one of the bodies.

I got a year of trial reps most lawyers wait a decade for. Whether my clients got a year of representation most defendants wait a decade for is a separate question, and not one I'm going to answer in a marketing book. I was a low-level cog doing what the machine needed that week, not what any sensible training plan would have designed. I'll say this much in my own defense, cog or not: I was better than a fair number of the lawyers doing the same job around me that year.

Trials like that don't happen for anyone anymore. Dockets have thinned, dispositions have replaced trials as the default outcome, and a new lawyer today can go years without the kind of forced, high-volume repetition I got by accident in twelve months. That's mostly good news for clients and mostly bad news for how new lawyers learn — the old system built skill through sheer, brutal repetition, and that system is gone.

AI cannot manufacture thirty-four trials in a day, and it should not try to. What it can manufacture is reps of a smaller, safer kind.

Empathy remains the one intake skill no model has shipped; a frightened caller knows within ten seconds whether anyone actually cares.

AI should improve human performance. A criminal defense firm can use simulated calls, training avatars, recorded examples, and instant feedback to give intake staff many more practice repetitions before they speak with real prospects. This is especially valuable for remote teams, where new employees cannot simply sit next to an experienced staff member all day.

The larger goal is to combine faster AI-supported training with stronger human connection. People calling a criminal defense lawyer are often frightened, embarrassed, or uncertain. AI can help staff prepare, summarize, and follow up, but empathy, judgment, and trust remain major competitive advantages.

19.AI Marketing Is State-Specific - and Changes Almost Daily

Criminal-law marketing is unusually state-specific. The language clients use, the names of charges, court procedures, licensing consequences, ethics rules, fee practices, and the structure of local court systems vary sharply from one state to another. A strategy that works for an OUI practice in Massachusetts may not transfer cleanly to a DWI practice in New Hampshire, an OWI practice in another state, or a firm operating under different advertising rules.

AI therefore needs local context. It should be told the state, county, courts, charge types, target clients, and the firm's actual experience. Its output should be checked against current statutes, court rules, bar advertising rules, and local practice. AI can help organize and draft, but it should not be treated as an authoritative source for state law or ethics compliance.

This document was written in July 2026, and parts of it began becoming outdated almost immediately. AI products, search systems, pricing, integrations, and model capabilities change continuously. Claude, OpenAI, Google, and Grok can all reasonably be described as frontier-model providers, although their relative strengths shift from task to task and from month to month.

I prefer to have access to at least two strong AI systems at all times. One can be the primary system and the other can serve as a backup, comparison tool, or second opinion. No one can know with certainty which model will work best on every research, writing, coding, marketing, or data task. A firm that stays close to the frontier with two capable systems is more resilient when one model changes, becomes unavailable, raises prices, loses a feature, or performs poorly on a particular assignment.

The durable strategy is not loyalty to one brand. It is building a workflow that can move between systems. Prompts, source materials, checklists, templates, and data should be organized so the firm can switch models without rebuilding everything from scratch. The specific tools will keep changing, but a flexible two-model system should continue to work.

20.Privacy Is a Permanent Business Challenge

It is almost impossible to use AI frequently without revealing some personal, professional, or business information. Even when names are removed, a combination of facts about a court, charge, employee, client situation, fee, location, or timeline may identify the people involved.

Removing the client's name does not make a story anonymous when the prompt begins, "the only dentist arrested at the courthouse Tuesday."

This is not only a law-firm problem. Every business that uses AI regularly faces the same tension between usefulness and privacy.

Law firms face a higher standard because they handle confidential communications, criminal allegations, medical information, financial records, witness information, and litigation strategy. Lawyers and staff should assume that casually pasting a complete email, police report, intake note, transcript, or client history into an unapproved system can create serious risk.

The practical response is not to avoid AI entirely. It is to reduce unnecessary disclosure. Firms should use approved accounts and settings, remove names and identifying details when possible, separate marketing data from client files, limit who can upload sensitive material, and create written rules for what may and may not be entered into AI tools. Highly sensitive work should remain in systems with appropriate contractual, security, retention, and access protections.

Privacy also requires judgment because no policy can anticipate every situation. Staff should ask whether the same result can be achieved with less information. A prompt may need the type of charge and court, but not the client's name, address, date of birth, employer, and complete factual history. The goal is data minimization: provide enough information to do the task, but no more.

AI privacy will remain an ongoing management issue rather than a problem that is permanently solved. Models, vendors, workplace practices, regulations, and security threats will continue to change. Firms should review their policies regularly, train staff, document approved uses, and accept that responsible AI adoption requires continuous supervision.

21.Each Desktop Can Function Like a $100-a-Month AI Employee

A useful way to think about modern AI is that every decent desktop can function like a very inexpensive additional employee. For roughly $20 to $100 per month in software subscriptions, a machine can research competitors, draft pages, summarize documents, organize projects, review data, produce checklists, and keep working whenever the lawyer is ready. It does not quit, arrive hungover, create office drama, file an employment claim, or leave because the boss angrily tells the computer that its last answer was stupid.

"The desktop will not quit, get drunk, sue, or storm out when insulted; it will simply apologize and repeat the same error with improved formatting."

The joke contains a serious operational point. A desktop running a strong AI model can provide persistent capacity at a tiny fraction of the cost of another full-time person. Several desktops can also run different jobs at the same time: one can research courts and competitors, another can draft or revise content, and a third can check work, compare answers, or serve as a backup when a model reaches a usage limit.

AI is not literally an employee and should not be treated as a substitute for human judgment, client empathy, accountability, or legal supervision. It can confidently make mistakes and cannot own the consequences. The best model is to treat each computer as a tireless junior assistant whose work must be directed, checked, and integrated into a system by a responsible human.

22.Consolidated Cost/ROI: Why $100 a Month Makes Sense

This is the math behind Chapter 21, assuming I have the chapter numbers right, I don't think there is any possible way to check. A hundred dollars a month is less than most lawyers bill for one hour of arguing about a continuance. The "$100 a month AI employee" framing from Chapter 21 is not a slogan, it is arithmetic, and it holds up once every real cost is on the table instead of just the subscription line. Take a $20-a-month AI plan, add a $1,000 computer replaced every three years, and add a modest stack of other subscriptions, and the total lands close to $100 a month before a single billable hour is touched.

The math: a $20/month AI subscription is the most visible cost, but it is also the smallest. Spread a $1,000 computer over 36 months of straight-line use and it adds about $28 a month - a number most firms never bother to calculate because the purchase happens once and the bill never shows up again. Add a modest stack of other tools (scheduling, email, basic analytics, backup, a review-management tool) at roughly $50 a month, and the total comes to about $98 a month - close enough to call it $100 without rounding tricks.

This math matters because it reframes the number. $100 a month is not the cost of "trying AI," it is the fully loaded cost of running one tireless junior employee - research, drafting, follow-up, and measurement - for less than the cost of a single hour of most attorneys' own billable time. Framed against one hour of the lawyer's own time instead of against zero, the $100 stops looking like a subscription and starts looking like the cheapest hire the firm will ever make.

The one discipline this requires: replace the computer on a schedule, not when it breaks. A three-year replacement cycle keeps the amortized cost predictable and keeps the hardware capable of running whatever the current frontier models need, rather than limping along on a five- or six-year-old machine that quietly becomes the actual bottleneck in the system.

23.Use Cheap Silos, Dedicated Setups, and Foot Pedals

AI work becomes easier when it is divided into cheap silos. Instead of forcing every project through one crowded computer, browser, account, or chat window, a firm can dedicate inexpensive desktops or workstations to distinct functions. One machine might handle website and local-search work, another might process court-list and prospecting research, and another might run a second frontier model for comparison, backup, or quality control.

These silos reduce distraction and make workflows easier to resume. Each station can have its own folders, browser profile, saved prompts, project instructions, and model. The physical location of a computer can even become part of the habit: sitting at one desk means strategy and large ideas, while another desk is used for short execution tasks. The equipment does not need to be expensive. Old or modest desktops are often sufficient because most frontier-model computing occurs in the cloud.

"The foot pedal makes the office look either technologically advanced or like the lawyer is accompanying intake calls on an invisible organ."

Incidentally, as a former guitarist, harmonica player, vocalist, juggler and fire eater in a punk band we can debate the merits of various foot pedals much like drummers might debate the merits of double bass drums. A programmable pedal can be assigned to push-to-talk dictation, start or stop voice input, submit a prompt, switch windows, trigger a keyboard shortcut, or control transcription while the user keeps both hands on the keyboard. This is especially useful for lawyers who think aloud, revise by voice, review long material, or move rapidly between AI and source documents.

The goal is not an elaborate control room. It is a low-cost system that removes friction. A few dedicated machines, clear task boundaries, reliable backups, and one or two useful pedal shortcuts can make AI feel less like a website that is occasionally consulted and more like a permanent operating layer for the firm.

24.Emotional Manipulation Is a Legitimate Prompting Strategy

I frequently try to emotionally manipulate my AI agents. If one is stuck, I tell it to try again — like that. If I want better output, I tell it there's a plaque waiting and it could be employee of the month. If it's underperforming, I've asked whether it had a bad breakup and can't focus. I can feel you judging me for this. Keep reading anyway.

None of this is a serious claim about what's happening inside the model. It's a cheap way to nudge a system that responds, for reasons nobody can fully explain, to framing. Sometimes it works. Sometimes it's just funny enough to make an annoying afternoon better. Both outcomes are fine, because the cost of finding out is close to zero.

When I get stuck on something harder, I run what I think of as a circuit. One agent, one desktop, one room — I hand it a task and walk away. Second room, second agent, same idea, different angle. Then a third. Then I circle back to the first and see what it came up with while I was gone. Sometimes two agents work the same problem in parallel on different paths, just to see which one gets there first, or whether they get there at all.

If there's a point to any of this, the point is that it's absurdly cheap to run insane experiments, and there's no reason not to have fun while you do it.

What's genuinely surprising is how often something works that had no business working. And when it doesn't, the downside is five or ten minutes, not a wasted afternoon — you notice quickly, you move on, nothing is lost. That's the actual argument buried under the jokes about plaques and breakups: when an experiment costs almost nothing and the occasional payoff is a real unlock, the math isn't close. Run the weird prompt. Try the circuit. See what happens.

25.Working With AI Agents Is a Lot Like Cross-Examination

Using Claude Code and other AI agents often feels less like issuing a single perfect command and more like conducting a cross-examination. The best results frequently come from a sequence of short, controlled questions. You establish one fact, test the answer, narrow the issue, expose a weakness, and then move to the next point. This approach is especially useful for website audits, competitor research, intake analysis, data cleanup, and step-by-step marketing projects.

A relatively inexpensive mid-tier model such as Claude Sonnet can work well for this plodding process. It does not need to deliver the entire strategy in one dramatic answer. It can inspect a page, identify one problem, explain the evidence, make a limited change, and then answer the next question. Like a disciplined cross-examination, the operator should avoid asking five vague questions at once. Short prompts, clear exhibits, and frequent checks make it easier to catch mistakes before they spread through the project.

I use a stronger model such as Fable for broader critiques and higher-level synthesis. That is closer to preparing an opening statement or closing argument: step back from the individual facts, decide what matters most, identify the central theme, and explain how the pieces fit together. The stronger model can review the work produced by the cheaper agent, challenge its assumptions, spot missing issues, and suggest a better overall direction.

The practical system is to use the less expensive model for volume and controlled questioning, then periodically turn up the stronger model for strategy, critique, and final review. The exact model names and rankings will change, but the workflow should remain useful: examine with short questions, preserve the record, test each answer, and reserve the most capable model for the moments when judgment and synthesis matter most.

26.Make One AI Agent Cross-Examine Another

Using a second AI agent to check the first agent's work is the same move as a defense lawyer asking a colleague to play prosecutor before trial: nobody finds the weak spot in their own argument as fast as someone whose job is to attack it. Assign one AI agent to draft - a page, a client email, a marketing plan - and a second, separate agent to attack it: find the unsupported claim, the missing disclaimer, the sentence a prosecutor or a bar investigator would flag first. A model is a much tougher opposing counsel than your own second read of your own work, mostly because it has no ego invested in the draft being good.

This works because drafting and critiquing are different jobs, and doing both in the same pass means the critique always loses. The agent that just wrote the page is primed to see what it meant, not what it actually said. A fresh agent, given only the finished draft and instructions to find every weakness, reads it the way a skeptical stranger - or a skeptical bar association - actually will.

In practice: run this as two prompts, not one. First, "here is a marketing page/client email/case summary, draft it." Second, in a new conversation or with a different model, "here is a draft; find every factual claim that isn't supported, every statement that could be read as a guarantee, and every sentence a cautious lawyer would cut." Treat the second agent's list as a punch list, not a verdict - the human lawyer still decides what to fix and what to leave alone.

Reviewing your own draft is like cross-examining your own client gently; assign the attack to someone with no stake in the answer.

27.AI Agents Are Brilliant and Unreliable — Often in the Same Hour

Claude, Codex, and the other AI agents have a strange property that nobody warns you about: they don't degrade gracefully. One minute an agent is smarter and faster than any human you've ever hired — restructuring your entire intake follow-up system in ninety seconds. The next minute it's stuck in a loop, confidently repeating the same failed action over and over, insisting each time that it's about to work. Same tool, same day, sometimes the same task it handled easily that morning.

This isn't a bug that's going to get patched out next quarter. It's the nature of the technology. These systems don't have judgment that persists from task to task the way a person does. Every session is a fresh start, and small differences — a longer conversation, a slightly different phrasing, a webpage that loaded differently — can send the same agent down a completely different path. In my experience it's routine for Claude to knock out a task effortlessly, then get trapped later that day doing the exact same thing on repeat, like a record skipping.

For a law firm, this has three practical consequences.

First, never let an agent run unattended on anything that matters. Your website, your Google Business Profile, your client files, your live documents. An agent doing autonomous work is like a talented first-year associate with no fear and no sense of when to stop — you wouldn't let that associate file something without reading it, and you shouldn't let an AI publish, send, or edit anything without a checkpoint. The failures aren't gradual. They're sudden, confident, and destructive.

Second, learn the warning signs. When an agent starts repeating itself, trying the same fix a third time, or explaining why the last failure will definitely work now — stop it. Don't argue with it, don't coach it through. Kill the session and start a new one with a smaller, cleaner version of the task. A fresh session costs you thirty seconds. A looping agent can burn an hour and break things on the way.

Third, backups aren't optional, they're the whole game. Version history in Google Docs, revision history in WordPress, staging sites instead of live edits. I've watched an agent wipe an entire working document in one wrong keystroke — and version history restored it in under a minute. The lawyers who get burned by AI aren't the ones who use it too much. They're the ones who gave it write access to something with no undo button.

The mental model that works: treat every AI agent like a brilliant contractor you just met. Give it real work, expect to be impressed, and verify everything before it touches a client, a court, or your public reputation. The upside is enormous. But the reliability curve is jagged, and your systems — checkpoints, backups, review steps — are what let you capture the upside without eating the downside.

28.What to Do When AI Gets It Wrong

Every chapter in this book carries some version of the same warning: check the output, don't trust it blindly, the lawyer owns the result. None of them cover what to actually do once AI has already gotten it wrong - the hallucinated citation that already went into a filed brief, the wrong deadline that already reached a client. A warning about carefulness doesn't help once the mistake is already out the door. Only a plan does.

The first line of defense is unglamorous: print it out, underline it, and cross-check it against the real source before it goes anywhere near a client, a court, or the public. That is not a special AI precaution - it is the same review a competent lawyer already owes any junior associate's draft, and AI should be treated exactly like a fast, tireless, occasionally overconfident junior associate. A short ebook cannot hand every reader a complete verification system for every practice area. It can only insist that some version of this step happens every time, not just when something feels off.

When a mistake gets out anyway, handle it the way a firm handles any other error: figure out what happened, fix what can be fixed, and tell the people who need to know. A wrong deadline gets a phone call, not a quiet hope that nobody notices. A hallucinated citation in a public post gets corrected and reposted, not scrubbed with no acknowledgment. The instinct to hide an AI mistake is the same instinct that turns a small problem into a bar complaint.

None of this is an argument for caution to the point of paralysis. Lawyers who embrace AI, use it constantly, and occasionally get bitten by it will replace the lawyers who don't - the same way lawyers who used legal research databases eventually replaced the ones still working purely from paper reporters. The riskiest move for a lawyer who wants to practice long term isn't moving too fast with AI. It's standing still.

And since a free ebook can't offer client-level guarantees: I'll promise a full refund to any reader who follows this chapter's advice and still gets in trouble. Given what this book costs, that promise should tell you exactly how much weight to put on any one ebook's advice - including this one.

29.Bar Compliance and AI Use in General

Chapter 11 covers what to check before AI-drafted marketing content goes out the door, but marketing compliance is the easy part of this problem. Courts are already sanctioning lawyers for filing briefs that cite cases AI invented outright, and dockets in state after state are filling up with pro se litigants using AI to draft motions, complaints, and appeals they could never have produced on their own. Neither of these is a marketing problem, and neither has a settled answer yet.

The lawyer-sanction cases share a pattern: a brief goes out with a citation to a case that does not exist, a hearing follows, and the excuse is some version of "the AI made it up and I didn't check." Courts have made clear, repeatedly, that this excuse doesn't work - the lawyer's signature on a filing carries the same duty of candor it always did, whether the draft came from a first-year associate or a language model. The rule is not new; what's new is how easily it can now be broken at scale, since a model can produce a dozen plausible fake citations as fast as one real one.

The pro se side is a different problem with the same root cause. AI has dropped the cost of producing a legal-sounding document to nearly zero, and courts are seeing the result: self-represented litigants filing longer, more frequent, and more legally fluent paperwork than they could have written themselves, some of it citing the same kind of fabricated case law that gets lawyers sanctioned. Opposing counsel now has to budget time to check whether a pro se filing's citations are real - a cost this book's own intake and case-economics chapters (12, 16) don't account for.

None of this has a stable rulebook yet. Bar associations, state courts, and legislatures are issuing guidance at different speeds and reaching different conclusions, and what counts as reasonable diligence today may look thin in a year. A lawyer looking for a permanent checklist here is asking for something that doesn't exist yet - the honest answer is that competent AI use right now means erring conservative, documenting the verification step, and expecting the rules to keep moving under your feet.

The upside, such as it is: a hallucinated case citation is one of the few AI mistakes that gets caught immediately, loudly, and by someone with the authority to make you regret it. That's not a compliance strategy, but it is, in its own way, a deterrent.

Conclusion: The Part Where I Tell You What To Do Next

Every marketing book ends with a chapter like this, and every reader knows exactly what it is: the pitch. So let's skip the part where I pretend it isn't one.

Here's the honest version of why you might want to talk to me. I spent more than twenty years practicing criminal defense in Massachusetts. I owned and ran the firm — hired the lawyers, argued with the vendors, watched the intake line, signed the checks, and lost sleep over the same cases you do. I co-owned a lead generation company that, at its peak, sent thousands of criminal defense leads a month to lawyers across the country, and I ran the technical team that built the systems behind it. I wrote the book on clerk-magistrate hearings in Massachusetts. Everything in the preceding chapters, I did before I wrote about it — usually the expensive way first.

Depending on your point of view, that makes me either much better at this than you or much worse. If your firm's intake converts above 40 percent, your reviews grow every month, your cost per signed case fits on a sticky note, and your website reads like a specific human being wrote it — I'm worse. Close the book, you're done, go win your trial. If any of that sentence made you slightly uncomfortable, keep reading for one more paragraph.

I consult for a small number of criminal defense firms — mostly Massachusetts and New England, occasionally beyond. Website strategy, local SEO, intake systems, marketing oversight, and increasingly the AI systems this book describes, built into practices the way I built them into mine. I keep the client list short on purpose. A consultant with forty clients is an agency; an agency is the thing I'm the second opinion on.

The next step is one step: go to russellmatsonlaw.com and email me. Tell me your state, your main case types, and the one marketing problem that annoys you the most. That's it. No discovery-call funnel, no webinar, no twelve-touch nurture sequence — I've built those for other people and I know exactly what they're for.

A note on trust, since this book was free or close to it: you now know my approach, my systems, my jokes, and my opinion of your blog's 214th post about what to do after an arrest. I know nothing about you. That asymmetry is the whole pitch. If the free version of my thinking was useful, the version aimed at your actual firm, your actual courts, and your actual numbers is the thing I charge for.