GenAI ain’t magic

‘GenAI ain’t magic’ is the presentation I did at Legal Geek Growth on 11 June 2026. I did a practice run to the International Practice Group (IPG) at their Spring Conference in Luxembourg on 9 May 2026.

The synopsis for the talk was:

A look at the reality, rather than the hype, surrounding GenAI. Covering hallucinations, the verification burden and neglected Legal Tech. Will things get any better anytime soon? Is it the end of lawyers or will AI actually generate a need for more? Oh… and there will be magic! ♠️ ♥️ ♣️ ♦️

Helen Burness summarised my talk, in a poem she wrote, in the verse:

Brian Inkster stepped up with a spicy first take:
Put down the AI, for all of our sakes,
Bring back the boring, the tried and the true,
Before the hype swallows everything you do.

I now give you the full ‘GenAI ain’t magic’ talk and slides as they appeared at Legal Geek (slightly amended following the practice run with the International Practice Group).

The talk when done live (without a script in front of me) may have departed a little from what follows and been condensed in places for time (it was an 8 minute TED Style talk) or lack of a photographic memory.

Thus, even if you attended the talk you might get something extra out of the written text. As I said in my talk the source material is always going to be better than a summary (especially a GenAI summary!). Furthermore, I have added in hyperlinks to more detailed posts on GenAI in legal practice that I have published previously.

Here goes – GenAI ain’t magic:

Good morning everyone. Today I am going to show you why GenAI ain’t magic… But this… is.

[I then produce a pack of cards as if from nowhere]

I will place the pack of cards down here and we shall come back to it later.

You’ve all seen the headlines about AI bringing about the death of lawyers. Richard Susskind, who you will be hearing from later this morning, has been suggesting that possibility since at least 2008, when ‘The End of Lawyers?” was published. But, don’t fret, it is all rather hyped up.

[N.B. Richard Susskind was giving a keynote at Legal Geek Growth on ‘AI and the Law in the 2030s’]

Especially when you realise that Generative AI is nothing more than “a random token predicting autocomplete on steroids toy”.

[N.B. Quote by Richard Self, Leadership and Keynote Speaker and member of the Data Science Research Centre at University of Derby]

The Copilot Terms and Conditions warn that it is for entertainment purposes only. It can make mistakes and may not work as intended. Harvey and Legora also warn in their terms and conditions of potential errors in their outputs.

It is a black box with a liar inside. A black box Pinocchio.

Those lies have become known as Hallucinations – An inbuilt feature of the system we are told!

An inbuilt feature that has resulted as at 28 May 2026 (when I had to submit my slides to Legal Geek) in 1,497 court decisions around the world where AI hallucinated content has featured. I checked this morning and that figure has gone up in the two weeks since the slides were prepared by 101 court decisions to a total now of 1,598 court decisions. Think about how many cases are going unreported every day.

[N.B. That is now 1,603 court decisions, three days later, as at the date of publishing this blog post. Almost two new cases per day. Check out the latest number collated by Damien Charlotin at www.damiencharlotin.com/hallucinations]

The verification burden of checking this erroneous output is very high. It was reported late in 2025 that an Australian law firm was having to hire more lawyers to deal with that burden.

Inkster’s Law (named after yours truly) says that it could take you twice as long to verify Generative AI output compared with just doing the work yourself from scratch.

You also must remember that it is not just about what it gets wrong but what it might miss out completely.

Legal Tech specific tools are not exempt from hallucinations. It is no coincidence that many associate Harvey with the imaginary rabbit from the James Stewart film rather than with Harvey Specter from Suits.

[N.B. Image credit © Stephen Allen, who created and first used the image in a LinkedIn post on ‘Harvey says AI will kill the billable hour‘. I did an April Fools joke about Harvey in 2024: Elwood launches claiming to be Real AI for Law]

Your best bet remains to ask a law librarian to assist you with legal research. They also have the benefit of access to many legal resources that have not been digitalised. And they won’t tell you to kill yourself.

[N.B. Photo is of the main Library at the Royal Faculty of Procurators in Glasgow (taken by Mark Sinclair for Inksters). Quote on overlaid image: “Instead of asking chatgpt, why not ask a librarian? You’ll get real answers and they won’t tell you to kill yourself.” by Spencer Koelle on Bluesky.]

So, don’t use it for legal research. We are told, however, that it is good for summaries. The result appears to be lawyers summarising documents they would never have considered summarising before to produce documents that they would never have produced before. Beware the danger of summaries missing key information that you would only ever ascertain from actually reading the source material.

We are told it is taking away mundane tasks. What are those? Not photocopying or scanning. Whenever I see someone post on social media how they have more time to do other things because of AI, I ask what it is doing for them. I, strangely, never get a reply. I wonder why?

Amongst all this hype lawyers are losing sight of existing Legal Tech that can make their lives more efficient with limited, if any, verification burden involved.

Check out your existing case management systems and the document automation functions that you have never implemented.

If Generative AI has a useful part to play within that case management system, your software vendor will no doubt find it and incorporate it. Let that happen rather than chasing shiny new toys meantime. However, alarm bells should ring if all your software vendor does is add in a Chatbot and say you are now AI enabled!

It has recently been announced that Microsoft has launched a Legal Agent in Word. That is where most lawyers live and may be all they really need. Interestingly, this doesn’t depend only on generative AI. It also uses a custom insertion method and a deterministic layer to apply edits more reliably. Not yet generally available outside the United States – but do watch that space.

We are continually told that we are at the worst point we will be with AI and it can only get better. Don’t bet on that happening as quickly as some may have you think.

Concorde first flew in 1969 and was in commercial service from 1976 to 2003. 23 years later and no sign of us seeing supersonic commercial flights again anytime soon.

Man first landed on the moon in 1969 and there were several more landings between then and 1972. We have not been back since.

We had expected a moon base by 1999. That is only now being mooted again following Artemis 2 travelling around the moon and back in April of this year. The plan is to next land on the moon in 2028. Some 56 years after we last did.

Whilst there was a leap in mobile phone technology in 2007, with the launch of the iPhone, has your mobile phone really got any better in the last 10 years or so? Was there perhaps a sweet spot somewhere in the middle of the overall evolution of the mobile phone that you would like to get back to?

[N.B. With hindsight, I should perhaps have mentioned in this section the Metaverse!]

The time gap is clear in AI hype. What was forecast to happen to self-driving cars and Radiologists is similar to the noise around AI and lawyers. Your jobs are safe for the foreseeable future.

[N.B. Meme shows images of (1) Elon Musk predicting in 2015 self-driving cars in 2 years; (2) Geoffrey Hinton predicting in 2016 that radiologists would be obsolete in 5 years; and (3) a motorway of cars in 2026 of radiologists in regular cars driving to work. I assume that the original creator (unknown) of this meme couldn’t spell obsolete or used AI to assist them!]

Indeed, I predict, that in the future, more lawyers will be required to clean up the mess being created by lawyers who are currently using AI badly.

Now let’s get back to the pack of cards from the beginning.

I then put on a top hat that popped up from nowhere.

I asked four different members of the audience to:

• Pick a colour (red or black) – Black was the choice
• Pick a suit, Clubs or Spades – Spades was the choice
• Do you want a number or a face? – Face was the choice
• Pick a face – Queen was the choice

So the audience had chosen the Queen of Spades.

I then picked up the pack of cards that had not been touched since the beginning and took out the cards from their box. I fanned the cards open so the card faces were facing the audience. Only one card in the pack had its back to the audience. I pulled that card out and it was the Queen of Spades!

When I did this magic trick at the end of my talk on ‘GenAI ain’t magic’ to the International Practice Group in Luxembourg the seven of Spades was chosen by the audience and then revealed by me from the pack

Now that’s magic!

[N.B. For an early post by me (from 2023) using the ‘magic’ comparison see: LLMs in Law: Hype v Magic]

Update on GenAI ain’t magic – The Great Big Legal Offsite (GBLO)

I presented ‘GenAI ain’t magic’, for a third time, at The Great Big Legal Offsite (GBLO) at Ribby Hall Village, near Preston on 3 July 2026. I added in a slide on the Metaverse and this time the 10 of Hearts was the card chosen and revealed

Credit for the Magic Training for ‘GenAI ain’t magic’

A big thank you to Scotland’s Magician, Scott Cuthbertson, for training me in the art of magic!

Reactions to GenAI ain’t magic

On LinkedIn the following comments have been made:-

Ralph Riddiough FRSA (Director, and Head of Corporate, at Holmes Mackillop Solicitors):

Great piece! Loved the meme about traffic jams. I’ve been playing around with CoCounsel recently which makes Practical Law far easier to use. I wouldn’t ever touch Chat GPT or CoPilot for assistance with the technical aspects of my job and I’m horrified to think solicitors would do so uncritically. However, I have been impressed with what CoPilot can do to help me with some of the management / admin aspects of my job.

Keep up your good work in this field, yours is an important voice of reason. Thank you.

Me:

Thanks Ralph. I think the problem began with GenAI does words, lawyers do words, GenAI must be able to do law. Then trying to make it do anything and everything a lawyer does without realising or appreciating its limitations. You clearly appreciate those limitations and your use reflects that. If only more would follow suit!

-+-+-+-+-+-

Paul Ryan (Data and Process Automation):

A good sensible analysis of where things are. I would only add that AI can make massive strides in getting more firms to implement true process automation. I say true automation because most have not because it got too hard and did a kind of enhanced matter management instead.

AI is perfect to get over the challenges of full process automation the main one being …it’s really really hard and time consuming. It also provides that layer of protection from hallucinations, inaccuracy and to some extent extreme cost variables in AI uptake. You’re using AI to help build and implement and maintain automation. It’s not actively involved in the legal process itself.

That feels a lot safer.

Me:

Thanks Paul. An interesting use case that I don’t think I’ve heard mention of before. Would be interested to discuss further. Will DM you.

-+-+-+-+-+-

Deborah do Carmo (Lawyer-Linguist. Raising the bar in legal translation since 2002.):

The magic framing works brilliantly here, Brian – props and all.

The bit about summaries immediately made me think of gist translation. People talk about ‘getting the gist’ with automated translation as if it’s risk-free, but the gist can be wrong too. And sometimes what gets left out is the whole point.

Me:

Thanks Deborah. Indeed. The general meaning is never going to wash where precision really matters.

-+-+-+-+-+-

Florence Brocklesby (Workplace and disputes lawyer. Founder of Bellevue Law, a certified B Corp):

Thank you for a really thought-provoking talk, Brian.

Me:

My pleasure Florence. Someone asked me over drinks at the end of the day: “How did they let you do that talk at this event?!”. The suggestion being that the vendor sponsors wouldn’t want any anti-hype messaging. It is indeed unusual to find a different voice at hashtag#LegalTech events. Thanks to LegalGeek for allowing my voice to be heard.

-+-+-+-+-+-

Helen Burness (Marketing strategist for the legal world’s doers, thinkers & change makers | Brand, strategy & LinkedIn | Visibility with purpose | Saltmarsh founder | HelenSquared co-founder | Strategic, creative and on your side.):

But I think the real slam dunk would have been sawing me in half.

Me:

I would have needed more than 8 minutes to include that particular trick. Maybe Jimmy Vestbirk will oblige another time!

Helen Burness:

Yeah you’re right it will have to be a whole workshop.

-+-+-+-+-+-

Kate Burt (Building businesses that challenge convention | Founder @ HiveRisk):

Looks amazing!! Too bad we’ve already booked our magician for GBLO this year! Sad to have missed this.

Me:

You could have had this first at GBLO 2025 following a prod by Michael Burne, after GBLO 2024, if I recall correctly. But if I also recall my Green Room requirements may have been unattainable. I mean a cage, carrots and security for my white rabbit did not seem unreasonable! GBLO 2027 perhaps if we start Green Room negotiations now?

Kate Burt:

Think you had a clash 2025 sadly otherwise it definitely would have happened. Still time to pull that rabbit out of a hat for 2026 – (all rider requests pre approved guaranteed)

Me:

GBLO 2026 is just a few weeks away!

Kate Burt:

That’s the beauty of magic, anything is possible.

Me:

When Kirsty Pappin asked me on Thursday if I was going to GBLO 2026, I said “No”. She will think it is magic if she sees me there. You better get your people speaking to my people about those riders in the morning (or maybe Tuesday – as we are having a special Scottish World Cup Public Holiday here tomorrow!)

Kate Burt:

My people will be in touch Tuesday – enjoy the celebrations ⚽️

Scott Cuthbertson (I help brands grow, scale and actually make money!):

Brian Inkster, Do you need an agent?

Me:

Might do!

-+-+-+-+-+-

John McCarthy (Helping Law Firm & Legal Tech Owners unlock £100K+ extra profit in 6 months):

Sounds like you had a lot of fun Brian Inkster I’m sure the audience did too.
What a great way to bang the message home that you can’t 100% rely on AI.
I use it a lot for marketing and copywriting but I have to treat it as an assistant not the guru.

Me:

Thanks John. Yes, it was a lot of fun. Indeed. Lawyers need to remember that they are the Gurus and GenAI is a very unreliable assistant that they need to hover over with a red pen to correct all its silly mistakes.

-+-+-+-+-+-

Richard Self (Leadership and Keynote Speaker and member of the Data Science Research Centre at University of Derby):

Excellent presentation.

Me:

Thanks Richard

-+-+-+-+-+-

Michael Burne (Founder, Carbon | Writing about Ordinary Pioneers | Curious about how people and organisations pioneer new ways of doing things.):

Looks like a cracking session to get folks thinking Brian.

Me:

Thanks Michael. It appeared to be well received from comments made after it.

-+-+-+-+-+-

John McKenzie (Chief Executive at the Royal Faculty of Procurators in Glasgow):

Love the photo of the Royal Faculty library. It may be from a few years ago, but it perfectly captures how the library remains a constant in a rapidly changing information landscape. And I love the message too: “Ask a librarian for a sensible answer.” Some things never go out of date.

Me:

Think taken circa 2007/2008. So almost 20 years ago!

-+-+-+-+-+-

Jude Copeland (AI Governance & Legal Tech Specialist | Legal Technology Group Lead | eDiscovery & Disclosure | Law Society NI Council Member):

I love this analogy Brian Inkster. As someone familiar with both the Paul Daniels Magic Kit and Generative AI, I see the similarities.

Me:

Thanks Jude. You do magic tricks too?

-+-+-+-+-+-

Alex Hutchinson (Independent Technology Director for Law Firms (Fractional / On-Demand) | Systems, Software & AI Change | Do It Once, Do It Right):

Look forward to reading this Brian Inkster

Me:

Will be interested in your thoughts Alex.

-+-+-+-+-+-

Joshua Blake (Helping Legal Leaders To Scale With Talent and Tech | Founder, JHB & Partners | Human-Led Search):

So gutted I missed this. Even belly laughed going through the slides on my own! 🤣

Me:

Thanks Joshua. It’s good to laugh. I had thought about appearing on stage with a Fez on rather than a top hat and with a trip in my step. But I guessed most of the audience would not know who Tommy Cooper was and therefore why I was wearing a Fez!

Joshua Blake:

haha even I know Tommy Cooper and he must’ve passed away a decade before I was born! Just be sure not to follow his health regime…

-+-+-+-+-+-

Allen Woods (How the hell did that happen?):

I have just read the piece. Excellent. Though domr will want to burn you as a heretic…

Me:

Thanks Allen. Some people did question how I got away with it!

-+-+-+-+-+-

Julie Saliba (Legal Technologist | Co-Founder and Chief Innovation Officer at echo.legal | Self-confessed geek and sometime lawyer whisperer):

eek – I hadn’t realised I’d “stolen” my Concorde idea from you!

I missed your talk unfortunately (so I think its more a case of “great minds” than actual theft!), but I totally agree that existing Legal Tech is often underused and find it baffling when users try to shoehorn AI into processes that older tech has been doing very accurately, efficiently and with not much more effort that setting up AI (and in some cases less!)

Me:

From our discussions at the event I’m sure it is “great minds”

-+-+-+-+-+-

Arpita Roy (Commercial Contracts | Legal Operations | Governance | 14+ years across in-house, global legal services and ALSP):

What I liked about your post is that it gets past the shiny demo and asks the less glamorous question: who checks the answer before anyone actually relies on it?

It brought to mind a contract-data process where reviewers validated extracted terms, mapped parent-child documents, identified duplicates and non-contract records, and manually captured commercial information. The records were corrected, but the learning often stopped with that record. The same issues could reappear in later batches and, depending on who reviewed them, may not always be corrected in the same way.

That led me to propose a wider feedback loop: capture corrections by error type, identify recurring patterns and differences in interpretation, feed them back to the product or technology team, and discuss those patterns with the client rather than treating review as a downstream clean-up exercise.

Human review can do more than catch mistakes. It can help the process become more accurate and consistent over time.

Otherwise, we risk building a very efficient system for repeatedly sending humans the same problems, sometimes with different answers.

-+-+-+-+-+-

Demi Markogiannaki (Founder at Executive Assistant Institute/ Founder at WeTeachMe / Board Member at HACCI / AFR 100 Women of Influence):

I like the focus on separating the reality from the hype. The most valuable conversations about AI are the ones that explore both its strengths and its limitations.

-+-+-+-+-+-

Fin Wycherley (Building trusted experts in the age of AI | Helping lawyers, consultants & innovation-led firms build authority, visibility & inbound demand | BBC Speaker | Meta & Google Trainer | Followed by Obama):

My favourite line here is the distinction between automation and judgment. One scales. The other remains stubbornly human.

-+-+-+-+-+-

Sarah Hassaan (Business Development Executive | Turning LinkedIn into a Lead Machine 🚀 | Social Media Strategy | B2B Outreach & Lead Generation | Helping Brands Grow Globally):

This is a refreshing take.

There’s so much hype around GenAI right now that the real conversations like hallucinations, verification burden, and actual implementation challenges often get overlooked.

The point about source material vs summaries is spot on as well.

Looking forward to reading the full piece.

-+-+-+-+-+-

Following my presentation of ‘GenAI ain’t magic’ at The Great Big Legal Offsite on 3 July 2026 the following comments were made on LinkedIn:

Matthew Letts (Strategic Tech Advisor, Founder, Entrepreneur, Solicitor, General Counsel and Legal Quant – Helping lawyers avoid burnout on busy work and law firms make more profit – Building big things in 2026 | Legal 500 – Key lawyer):

Just a shame that many of Brian’s views on AI isn’t borne out in fact, it’s just ideology.

Radiologists won’t be replaced by AI…erm, specialist AI is, objectively, more accurate at diagnosing specific types of diseases than the world’s leading radiologists.

It’s clear Brian believes in what he says. I believe history will demonstrate those beliefs are misplaced.

Me:

Facts cited are here!: https://www.damiencharlotin.com/hallucinations/ With 8 more hallucinated cases added since this morning.

… Just a shame that most commentators commit one or more of the Seven deadly sins of Legal Tech predictions: https://thetimeblawg.com/2018/10/01/travels-through-the-blawgosphere-2-artificial-intelligence-and-law-robots-replacing-lawyers/

Matthew Letts:

You operate on the premise that humans are infallible. Lazy use of Gen AI is problematic. Responsible use of specialist tools is not. You lump it all into one and overgeneralise.

Me:

If anyone is overgeneralising it appears to be you!

You also make assumptions which are simply not true.

I certainly do not operate on the premise that humans are infallible. In fact my talk demonstrates how fallible they can be especially when using technology badly or that is not fit for purpose.

Lazy use of GenAI is, of course, problematic. As of today’s date Damien Charlotin cites 1,696 cases identified so far in reported court decisions. As I said in my talk, imagine how many are going unreported.

Specialist tools are not perfect and those thinking they may be using them responsibly often will not be. Hence my James Stewart / Harvey slide.

Maybe you were hallucinating when watching my talk. All my points are backed with actual facts and there are links to greater detail involved in the blog post I wrote after the last presentation I did at Legal Geek Growth.

I note that you and Simon Marshall provide AI training to lawyers. If you and/or he are not highlighting the problems and risks that I do, and consider those to be ideological and not real, then you will be exacerbating the problem that currently exists with AI use within legal practice. On your heads be it.

Oh… and to come back to your comment regarding radiologists. My point was how far out timeline predictions are. Geoffrey Hinton predicted in 2016 that radiologists would be obsolete in 5 years. 10 years later and they are not.

I asked Copilot “Are jobs being lost to radiologists by AI”. It answered:

The short answer is no, radiologists are not currently losing large numbers of jobs because of AI. In fact, the evidence so far suggests that AI is changing how radiologists work more than it is replacing them.

It went onto say, amongst other things, that:

Could AI reduce the amount of work?

Possibly. A recent task-based analysis projected that AI could significantly reduce the number of hours radiologists spend on certain tasks, particularly report drafting and study triage. However, the same study concluded that job losses are unlikely in the foreseeable future because imaging demand continues to grow and there is a persistent shortage of radiologists. [medrxiv.org]

What is likely to happen?

The most likely scenario is that radiologists who effectively use AI become more productive, while the profession itself evolves. Routine image interpretation may become increasingly automated, but radiologists will spend more time on complex cases, multidisciplinary discussions, patient-facing activities, and oversight of AI systems. [medrxiv.org], [rcr.ac.uk]

Bottom line

AI is transforming radiology, but there is currently little evidence that it is causing widespread radiologist unemployment. The stronger concern today is not a shortage of jobs, but a shortage of radiologists relative to demand, with AI being deployed to help close that gap. [nature.com], [link.springer.com], [rcr.ac.uk]

As I said, in my talk, similar to what has actually turned out to be the case after similar predictions were made about AI and lawyers.

-+-+-+-+-+-

Kate Burt (Solicitor by profession. Entrepreneur by instinct. Connector by nature. Building businesses that challenge convention | Founder @ HiveRisk | Co-Founder @ The Lawyers Retreat, GBLO & Fear No Storm Podcast | NED):

There’s not one conversation that I’ve had with Brian Inkster where I’ve not been belly aching laughing and also been slightly terrified at the same time. Unpredictable conversation at its best. Loved having you with us for GBLO year 3 Brian. You’re not afraid to upset the apple cart. Robust debate needs voices of dissent. Voices that get people stirred up. Voices that demand the others to use their own voice to state what’s true for them.

Me:

Thanks Kate, for those kind words 🙂. I’ve noticed your changing facial reactions from apprehension to laughter when we chat 😉. I’m often told that I am brave to say what many others are thinking but afraid to say out loud. As you say, that generates robust debate. Happy to have participated in #GBLO this year and rocked the GenAI apple cart a little as a result.