How Are Law Firms Using AI in Marketing? What to Systemise and What to Leave Alone

I will say this plainly before anything else. What AI can do inside a law firm right now is the most interesting thing I have seen in fifteen years of doing this, and almost nobody is using it for the thing that matters.

Most articles on this subject are a list of five tools. Tool lists are easy to write, and they are the reason so many firms have spent two years using AI enthusiastically with nothing to show for it. The firms getting real value use ordinary tools inside a system that decides what gets automated, what a human still has to do, and how anyone would know whether it worked, and then hand that system to their team.

This article is that system. It draws on the campaigns and content we run for law firms in Australia, on the published research where it exists, and on the places where the research contradicts what the industry keeps repeating. Where the evidence is thin or comes from a company selling AI, I have said so rather than dressing it up.

How are law firms actually using AI in marketing?

In three places. Most of the hype attaches to the one that matters least, so it is worth separating them.

The first is production cost. Research, drafting, editing, repurposing, translation, image work, first-pass analysis. Work that used to be rationed by how many hours your team had is now rationed by how much of it you can check.

The second is how people find you. Long conversational questions increasingly get answered inside the search results or inside an assistant, and the person never visits the ten blue links that used to feed your website.

The third is decision quality, and this is the one that has barely moved. AI is not better than you at deciding which practice area to invest in or whether an enquiry is worth chasing. Treating it as though it were is where firms lose money.

Worth knowing before you spend anything: adoption has already happened and the money mostly has not followed. Clio's 2025 survey put AI use among legal professionals at 79%, flat on the year before, while only about a third of solo and small firms reported any revenue lift from it. The gap is not the software. In the same data, 86% of solo firms had not changed their pricing since adopting AI, which under hourly billing turns every hour saved into revenue removed. Thomson Reuters found the same split from the other direction: where a named AI strategy existed, 66% said AI met or exceeded expectations, and where it did not, 22%.

So the framing I would use is this. AI changes the cost of producing marketing and the route by which clients arrive. It does not change what makes a firm worth hiring, and it does not do your thinking.

Which parts of your marketing should you systemise first?

The order matters more than the tooling. This is the sequence we work in, and it is deliberately unglamorous at the start.

Most firms want to start at step four because it is visible. Starting there produces a large amount of content pointed at nothing in particular, which is how you end up with a busy blog and the same number of matters. It is also how firms end up paying for junk leads and calling it a traffic problem.

Start at step one because everything above it depends on knowing what a signed client currently costs. We have written separately about what a law firm lead should cost, and the first surprise for most firms is not the number, it is that they could not produce it.

The second reason to work in this order is capacity. Do one thing until it runs every week without anyone thinking about it, then take the next one. A firm that tries to systemise six things at once systemises none of them, and that is true whether the work is done by your team, an agency, or an outsourced CMO.

What does it mean to systemise marketing down to a button?

This is the part I care about most, and it is the thing I am building for every firm I work with.

The aim is not that a person uses AI to write faster. The aim is that a task stops needing a decision. Take the work, do it properly once, write down what good looks like, build the system that produces it, and then hand your team a process they run by pushing a button. I call it getting marketing to muscle memory: the socials, the articles, the reporting, all of it just activates every week, and the founders do not have to think about it.

That order matters. You cannot systemise something you have not first done well manually, because you will automate your own bad version of it. So we take one priority, work it until it is running every week without anyone driving it, write it down, automate what can be automated, and only then start the next one.

That last step is the one people skip. When a client tells us not to say something, or that a piece missed the mark, that correction does not get applied to one article. It becomes a rule the system follows from then on. It is like a suit. Off the rack it is close. The value is in the tailoring, and the tailoring only happens if someone is feeding back.

What does an AI content system look like in practice?

Not a person typing prompts into a chat window. That is where everyone starts and it does not survive contact with volume.

A working system has five fixed parts. A research step that gathers real sources and the firm's own material. A brief that encodes what good looks like for this firm, including the voice, the reading level, and the things the firm will not say. A drafting step. A human review step that cannot be skipped. And a publishing step that handles the structural work nobody enjoys, which is internal linking, schema, images and the update of anything the new piece makes stale.

The part people underestimate is the brief. The difference between content that reads like every other firm's and content that sounds like yours is almost entirely in what you put in front of the model, not in which model you use. On our own client work, the substance comes from the practitioner, usually a recorded conversation, and the system handles structure and speed. We have set out that whole approach in our AI content strategy.

The part people overestimate is volume. There is a real difference between forty carefully reviewed pieces that each answer a question a client actually asks, and four hundred pieces generated because the system could.

Does AI content at volume actually work?

The industry position is that it does not. Our own accounts say otherwise, and I want to show the evidence rather than assert it.

For one litigation firm, we started loading the site with content in April. The content system produced around 70 articles in the strongest month and about 55 of them went live. Search Console went from roughly 30 clicks and 1,400 impressions a week when we came on, to weeks of 200 to 337 clicks and 13,000 to 17,000 impressions. Over the period below, that is 22,400 clicks and 1.58 million impressions.

One client account. The flat line is the year before we started. The climb is content, and there is no ad spend component to it.

The numbers on that chart matter less than three things behind them.

The first is that it is free traffic. If the firm stopped working with us today, that pattern continues for a long time, because nothing about it depends on a monthly media budget. Ads stop the day you stop paying.

The second is that volume is a choice in a way that most marketing inputs are not. Links are like your metabolism, hard to control. Content is like going to the gym. You can go once a week, or seven times a week, or twice a day, and it is up to you how much effort you put in. You never know which article is the one that takes off, which is an argument for more attempts rather than fewer.

The third is that it is not the writing that produces the result, it is the structure. We optimise the service page, then build topic clusters underneath it: every question somebody asks about that matter type, each article linking to the others and back up to the service page. Somewhere between 20 and 50 articles on a subject is where a site starts being treated as an authority on it, and that threshold is the whole reason volume matters.

The honest counterweight, which I give every client: this produces lead volume, and volume is not the same as quality. In that same account, some channels ran at strong qualification rates and others produced a month of enquiries with almost nothing viable. I am happy to pan for gold when the traffic costs nothing, but you have to be honest that you are panning.

How do you turn what already happens in your firm into content?

This is the use case I would start with for most professional services firms, because the raw material already exists and nobody is using it.

Every month your firm produces things: matters resolved, questions clients asked, an article somebody wrote, a video, a talk, a change in the law you had to explain six times. That is a month of content sitting in conversations, and the reason it never gets published is that turning it into posts has always been somebody's job that nobody had time for.

The method we use is a belief map rather than a topic list. You write down what your prospect currently believes, then work out which of those beliefs you need to shift and what evidence shifts it. If a business owner believes the tax office is generally reasonable and their risk is the same as it has always been, that is the belief the content argues with, not a generic post about compliance.

I think about it like changing any entrenched habit. You can tell a smoker it causes cancer and nothing happens, because information is not what changes them. Something has to move the belief itself. Marketing that only educates is telling people it causes cancer.

The dosage matters too. A prospect might arrive holding one or two beliefs about your area, or none at all. If your content shifts ten of them, they arrive at the call already agreeing with how you work.

One finding worth knowing before you commit to video. When we audited a law firm's LinkedIn, video earned higher engagement but roughly a third of the impressions of their text posts. Video does other jobs well, including hiring and referral credibility, but if reach is the goal, text is still doing the heavy lifting on that platform.

Can AI help you hire, not just market?

Yes, and it is the use case that surprises firms most, so it is worth describing concretely.

For one firm's graduate intake we built this: an ad drives applicants to a page on the firm's site, a form collects the resume, the university transcript, the grade average and the position they are applying for, and an optional video introduction. Everything lands in the firm's CRM, and an AI pass reads across all of it against rules the firm set, scores each candidate and writes a short summary.

The interesting part is not the scoring. It is that the AI reads the documents against each other. It will tell you that what a candidate wrote in their application does not match what the transcript says. When I asked the firm whether that was accurate, the answer was yes, and that some of what it picks up is brutal.

Two design decisions did more work than the model did. We made the video optional deliberately, because only a small share of applicants will record one, and the ones who do have told you something. And we had to solve grading, because Australian universities do not share a scale: some report a grade point average out of seven, some out of four, some a weighted average mark out of a hundred, and a few changed their system partway through the last decade. Without a dropdown forcing the applicant to declare which scale they are quoting, every comparison is nonsense.

The firm raised the right objection before we launched, which was that a good candidate might get filtered out because the grading rules are unfair across schools. So the system tags and summarises, and humans decide. If there are a hundred applications you read them. If there are a thousand you lean on the scoring harder and accept the trade. That judgment stays with the firm.

What changes when your website can be edited by prompt?

More than I expected, and this is the one I would rank above the clever stuff.

The firms we have moved onto AI-built sites can change them the day they decide to. A new team member, a new practice area, a landing page for a campaign, a qualification widget that screens enquiries before they reach intake. What used to be a development ticket and a fortnight is now a prompt and a deploy, and the page speed scores come out near perfect because the output is plain, fast HTML rather than a stack of plugins.

I converted a client's site over a weekend because I was annoyed at how slowly I could move on the old one. The competitive effect is not really about cost. It is that your website becomes a living thing you improve every week, while your competitor's site is whatever their developer had time for in 2023.

There is a caveat worth stating, because the early versions of this deserved their reputation. The first generation of AI-built sites had genuine indexation problems. That is solvable now, and we solved it, but I would ask anyone selling you one how they handle crawling and indexation before you sign.

Why can't AI answer questions about your firm yet?

Because the answers are not anywhere it can read them, and this is the gap almost every firm hits the moment they get excited about AI.

Your enquiries live in one place, your matters in another, your ad spend in the platforms, your outcomes in somebody's head and your intake notes in a spreadsheet that gets updated when things are quiet. An assistant cannot tell you what a signed client costs you in a practice area, because your firm has never put that number anywhere.

So the unglamorous prerequisite for the good part is a layer that pulls it together. For our clients we built PixelRush HQ for exactly this reason: every enquiry, its source, what intake did with it, whether it signed and what it was worth, in one place, updated by the people doing the work. The tool matters less than the discipline. What matters is that the data leaves people's heads and lands somewhere structured.

Once that exists, the thing everyone wants becomes possible. You can point an assistant at your own numbers and ask the questions you would normally wait a week for a report to answer. Which claim types are actually profitable this quarter. What happens to conversion when we respond in under ten minutes versus an hour. Which source produced the matters we regret taking. Where the enquiries that never got called are coming from.

That is the part I would describe as exceptional, and it is not really about marketing. It is a principal being able to interrogate their own business in a minute rather than in a meeting.

One production company we work with has taken this furthest. The founder connected his meeting transcripts, email, drive and calendars into one assistant, and then did the unglamorous thing: he talked into it. Walking, or while physically building equipment, he narrated his own processes for forty minutes at a time, and the system filed the transcripts and turned them into documentation. His words for it were that he had no idea how much was in his head until he started talking. A single session on one part of the kit produced specification, assembly, setup, troubleshooting, maintenance and sourcing, at a level of detail nobody writes down voluntarily.

What it replaced is the part professional services should pay attention to. He had been about to hire a junior assistant and put it off. Founder-led training became the system training people instead. Support knowledge that existed only in his head became something a new casual could ask. None of that is marketing, and all of it is why his marketing now runs without him.

The structure that made it work was three layers: context first, which is just getting everything into one place; then skills, which are the repeatable jobs you want done a particular way; then automation, which only makes sense once the first two exist. Firms that start at automation build something impressive that produces the wrong thing quickly.

There is also a quality loop worth copying. When somebody edits the system's output before sending it, the edited version goes back in, so the system learns the gap between what it produced and what the business actually used. That feedback ledger is the difference between a system that improves monthly and one that drifts into slop.

I will give you the honest side too, because I lived it. The early version buried the founder in notifications until it was told to send one daily summary. It is slow and unpredictable, somewhere between thirty seconds and ten minutes. It loses context and makes assumptions. It is never finished, because it needs feeding. And my own view is that the brain impressed me less than the website did. The thing that changed my week was being able to prompt changes into a site in minutes. Both are worth having. Only one of them changed my work immediately.

How do you measure whether any of this worked?

By the only number that survives contact with a partner meeting, which is what a signed matter costs you and what it is worth.

That means the chain has to be joined up: enquiry, source, qualification decision, signed or not, value. Most firms have the first two and none of the rest, which is why so many agency relationships end in an argument about lead quality. We built PixelRush HQ to close that loop for our own clients, and the honest reason is self-interested as much as anything, because without the feedback from intake we would be guessing like everyone else.

Our own reporting pulls financial data in alongside the marketing numbers, so recurring revenue and performance sit in one view rather than two. For a law firm that raises a fair question about where the data lives and who can see it, and the answer is usually a controlled machine in the office rather than somebody's laptop. Almost anything can be integrated. The work is deciding who has access and keeping it confidential, and that conversation should happen before the integration, not after.

Two warnings. Attribution from any single source is clunky, and when systems integrate they often lose the source information in transit, so capture it twice: what the system recorded, and what the client said when your intake asked them. And treat the CRM like weeds. Most firms get it under control, look away for a month, and come back to a lawn they cannot see over. You are not systemising to get one tidy month. You are doing it to stop the weeds coming back.

There is also a use for this that has nothing to do with marketing. Once you can see how enquiries convert by the person handling them, you can set a standard for the next person you hire, rather than hoping they turn out like the partner who is good at it.

For the AI-specific work, add three measures that most firms do not keep.

The branded search measure is the one I would add first if you are worried about AI search. You cannot see most of the assistant conversations that mention you, but you can see the people who go looking for you by name afterwards.

Can a small law firm run like a big business?

This is what I keep coming back to with the firms I work with, and it is the ambition I would set if I were running a practice today.

Think about how a global franchise operates. Every location does the same thing, to the same standard, because the operations are written down and the training is built around them. Nobody relies on the manager being brilliant. Head office can see the numbers for any store, this week, without asking anyone to prepare a report.

Almost no small law firm operates that way, and there is no reason they cannot. The parts are the same: documented operations, training built from those documents, and a small set of numbers that mean something. What changes with AI is the cost of producing all three. Writing the operations manual used to be a project nobody had time for. Now the first draft of every process, script and training module comes out of the work you already do, if somebody is recording it.

The practical version inside a firm looks like this. A written intake process, and training built from it, so a new hire is useful in a fortnight rather than a quarter. A written qualification standard, so viability is judged the same way by every person. Weekly numbers that arrive without anyone assembling them, which is what PixelRush HQ was built to produce. And the marketing system described above, running on a schedule.

None of that is exotic. It is just the unglamorous operations work that firms never get to, made cheap enough that they now can.

This is also where the size argument actually lands. A national firm with a big department and a broadcast budget will outspend you, outrank you on the terms everyone wants and appear in more places than you can, and no subscription changes that. We have set the full comparison out in how a small firm competes with the big legal brands.

What a lean team with working systems beats them at is specificity and speed. A brief built from one practitioner's actual expertise produces content a national content team cannot match, because that team is writing for eleven practice areas and cannot get its senior people on a call. A firm that decides something on Tuesday can publish it on Tuesday. That holds across family, commercial, litigation, criminal, property and immigration work, and in Sydney, Melbourne, Brisbane, Perth and Adelaide alike.

How should AI change your intake, and how far should it go?

This is where the money is, and it is also where firms overreach fastest.

The case for automating here is strong because the problem is a timing problem. Enquiries from advertising do not behave like referrals. They arrive cold, at night, from people contacting several firms, and the firm that answers first usually signs them. We have set out the evidence on response time elsewhere, and it is the single most reliable improvement available to most firms. The underlying difference between advertised and referred enquiries is covered in why law firms lose cold leads.

What AI does well here is everything that surrounds the conversation. Acknowledging an enquiry instantly at 11pm. Capturing the facts that decide viability before a human reads it. Scoring and routing against your written qualification standard. Drafting the follow-up sequence. Transcribing and summarising the first call. Flagging the enquiries nobody has touched in 24 hours.

What it should not do is decide whether a matter is viable, or talk to a prospective client in a way that could be mistaken for advice. The first is a judgment call, and judgment is the thing AI gets wrong in a confident voice. The second is the one part of this that regulators have actually addressed: Florida's bar requires a lawyer to tell people when they are dealing with an AI rather than a person, which I come back to below.

There is also a quieter risk. The firms that get the most out of automated intake are usually the ones with the worst manual intake, and automation can hide that rather than fix it. If nobody currently rings enquiries back three times, an AI system that sends three emails will produce a better-looking dashboard and the same number of signed matters. That is the same failure we describe in stop blaming your marketing agency.

Are your clients really finding you through ChatGPT?

Some are, far fewer than the headlines claim, and almost none of them show up as measurable traffic. Both of those are true at once, and the gap between them is the most misunderstood thing in legal marketing right now.

What decides it is query length, not topic. Pew logged nearly 69,000 real searches and found AI summaries on 8% of one and two word queries and 53% of queries of ten words or more, with clicks on a traditional result dropping from 15% to 8% when a summary appeared.

I checked this against live legal results while writing. On "personal injury lawyer melbourne" and "car accident lawyer near me" there was no AI overview at all, just a local pack then organic listings. On "do i need a lawyer for a car accident" and "how much does a divorce lawyer cost" there was one, and it drew on Reddit, YouTube and the blog posts of individual small firms.

So the damage is concentrated exactly where consumer firms built their acquisition model: the informational article that used to pull people in early and hold them until they were ready to call. Hire-intent search is, for now, untouched.

The traffic numbers look like nothing. Audits of law firm portfolios put AI referrals under half a percent of sessions. Yet surveys have roughly 40% of people saying they would use ChatGPT to research a lawyer. Both figures are vendor-published and directional, but the gap is the finding: people research in an assistant and then arrive by another route, usually by searching your name or typing your address straight in.

While we are here: the most repeated statistic in legal marketing this year, that 78% of legal queries trigger an AI overview, comes from a press release with no published method, and the vertical tracking that does exist puts healthcare, education and business technology above legal. Do not build a budget on it.

How do you actually get cited by AI assistants?

The answer departs sharply from the SEO playbook most firms are paying for. When Ahrefs correlated AI mentions across 75,000 brands, the strongest predictors were YouTube mentions at 0.74 and branded web mentions at 0.66. Domain rating came in around 0.27, and backlinks lower again. Separately, only 16.7% of AI overview citations came from pages ranking in the top ten. Being talked about beats being optimised, and ranking first is not the entry ticket firms assume.

Most practice areas in most cities are currently unclaimed here, and they will not stay that way. One of our clients already sees the effect at intake: enquiries that arrive via an assistant come in warmer and better informed than any other channel, and they also take longer to sign, because those people built themselves a shortlist before they called.

What do other firms actually say about this?

Three reactions, all from people I work with, and they are more useful than any survey.

The first is a principal at a boutique firm who codes. He has built two internal websites himself with AI tools, and when his site broke he had already diagnosed it with a chatbot before he emailed me. His words were that he is one of those clients who thinks they can do everything themselves, and that he genuinely enjoys it. His only reservation was design, not capability: he did not want the result to look generic. That is where a lot of capable principals now sit, and it changes what an agency is for.

The second is a marketing lead at a litigation firm who was pitched an AI content product and said no. Her reasoning was better than the pitch: the firm had invested years in content that resonates with its audience, and its actual bottleneck was distribution, not production. Making more was not the problem. I agreed with her, and I would give the same answer to most firms whose content is already good and unread.

The third is a family law principal who was cautious, bought the content system, and then reviewed the first batch with her partner. Her verdict afterwards was that there were not many changes at all. That is the bar this has to clear: a lawyer reading output about their own practice area and not needing to rewrite it.

The pattern across all three is that the technology is not what they are deciding about. They are deciding whether it fits how their firm actually works.

What are the best AI tools for law firm marketing?

This is the question everybody asks and the least useful one. Most of the work described above is done with general-purpose assistants rather than legal-specific products, and that is where the market has moved: Clio's 2025 data found 40% of legal professionals using legal-specific AI tools, down from 58% a year earlier, while overall use held steady.

The categories worth having are small. Something that drafts and researches. Something that handles transcription, so a practitioner's expertise can be captured by talking instead of writing. Something in your website or CRM that scores and routes enquiries. And whatever your ad platforms already give you, which is mostly automated bidding you are using whether you decided to or not.

It is also worth knowing which parts of the old playbook still apply, which we have covered in why law firms are shifting away from SEO and sustainable SEO for lawyers.

Every firm in your market has access to that same list, so none of it is an advantage. The brief you put in front of the model, the practitioner knowledge you feed it, and the review step you refuse to skip are the parts your competitors cannot copy.

What should you never hand to AI, and where is the real risk?

Two things, and neither is the one firms worry about.

The first is judgment. A field experiment with 758 consultants at Boston Consulting Group found AI users 25% faster and rated over 40% higher on creative work, and 19 percentage points less likely to reach the right answer on a problem that required reconciling data with interviews. Same people, same tool, same week; the task type was the only difference. In your marketing, the equivalents are deciding which matters you want more of, reading whether an enquiry is genuinely viable, setting positioning, and any claim about outcomes. Hand over the drafting, keep the deciding.

The second is scale. Almost none of the AI-specific guidance for lawyers touches marketing at all, and where it does, Florida's is the clearest: tell people when they are talking to an AI rather than a lawyer. Assume that applies to any intake chatbot you run. But the real exposure sits where it always did, in the ordinary advertising rules, and AI makes those far easier to breach because it lets you publish everywhere at once.

Australia gives the sharpest example. Advertising for personal injury work is restricted by statute in several states, with Queensland's among the tightest, and we have covered what Queensland firms can and cannot say separately. Earlier this year a Queensland tribunal dealt with a firm whose billboard, website and social content all used one common form of words. The conduct was upgraded to professional misconduct and the penalty rose from an agreed $2,000 to $30,000 with a public reprimand. One phrase, replicated across channels, over several years. Now imagine it sitting in a template a system uses two hundred times.

Google is not the risk here, incidentally. Its position is that method does not matter and quality does, and I could not find a documented case of a law firm site penalised for using AI. What exists are the review failures, CNET and Sports Illustrated, which is the thing you are actually managing.

Where should you start if you are doing none of this?

Pick the smallest thing that gives you a number, and be honest that this is a twelve month build rather than a weekend of prompting.

If you cannot currently say what a signed client costs you, start there, because every other decision in this article is unanswerable without it. Our budget guide shows how to turn that figure into a number you can spend to. If you can, start with intake speed, because it is the cheapest gain in most firms. Only then move to content, and when you do, cap it at what your review capacity can genuinely carry.

Then write the two pages nobody writes: a one page AI policy that says which tools your staff may use and what may never be put into them, and a content brief that encodes your voice, your practice areas and the advertising rules you work under, which in Australia means the state based restrictions covered in marketing a law firm in Australia and, in Queensland, the PIPA advertising rules. Those two documents are worth more than any tool you will buy this year, and they are the difference between the 66% who say AI meets expectations and the 22% who do not.

If I were running a firm today, the order would be this. Get every enquiry recorded through to a signed matter and a dollar figure, because it makes everything else answerable. Fix how fast enquiries are answered, because it is free. Then build one marketing system properly, to the point where somebody junior runs it weekly by pushing a button. Then the next one. Then point an assistant at your own data and start asking it things you have never been able to ask.

The pattern across every credible piece of evidence here is the same. The technology is not the variable. The system around it is, and so is the honesty of the measurement underneath it. That is unglamorous, and it is also why so few firms have done it, which is the opportunity.

If you want to see how we run this for firms rather than read about it, the growth system page sets out the mechanics and our case studies show what it produced.