A buyer’s guide, built on original research
Using The Credibility Stack, a framework by Ryan Vet, from a census of 423 bureau-listed AI speakers.
Most AI sessions that disappoint were not booked badly. They were booked for the wrong job. Here is how I would pick, and the tool I would use to do it.
45% of the AI speakers major bureaus list have no AI work or study a buyer can verify. Not that they have none. None you can find.
Census of all 423 people listed under the AI category at 6 major North American bureaus. The full study · how it was done
That number gets attention and it is the least useful thing on this page. Unverifiable is not the same as unqualified. Some of the best communicators I have shared a stage with are in that 45%. The reason AI sessions actually go wrong is duller and much easier to fix: the room needed one kind of speaker and somebody booked another.
Three questions, about thirty seconds
Answer as the event, not as yourself. Nothing leaves your browser and nobody asks for your email.
For that session, insist on
Ask them
The framework
Expertise in AI gets earned three different ways. They are not alternatives, they stack: each one you can verify is another reason to believe the next thing out of their mouth. Across every AI speaker the major bureaus list, here is how they hold up.
01
They have done it
Built, deployed, managed a deployment of, or professionally taught the hands-on use of AI, with a dated third-party source behind it.
What it buys youCredibility with a room that has to implement something. They have hit the problems your team is about to hit.
02
They have examined it
Academic research, peer-reviewed publication, analyst work, or a sustained body of dated, sourced analysis.
What it buys youRange and accuracy. They can tell you what is actually known versus what is being claimed this month.
03
They were here first
Verifiable AI work dated before ChatGPT went public on November 30, 2022, in any role.
What it buys youPerspective across more than one hype cycle. Not a quality bar, and plenty of excellent speakers started later.
How many of the three they can prove
A third of the people listed as AI speakers can prove none of the three in public. Fifteen percent can prove all three. That spread is the whole reason this page exists, and it is also why “AI speaker” on a bio tells you almost nothing on its own.
What changed my mind about this
Both ran twice, so I got into both. One was a professor. Properly academic, really sharp, and honestly more my speed. The other had run a consultancy for years and got onto AI in the last two or three, and her session was all practical: here is the workflow, here is the tool, here is what you do Monday.
I asked people afterward which one they got more out of. It was not close. Almost everyone said hers.
Now here is what complicates it. Her screenshots and the videos she played were of things that had been deprecated. Not out of date, gone. You could not go use them if you wanted to. His content was, by any measure I can defend, more accurate.
And the worst part was not even that. She had clearly used Claude to build her slide deck, but only some of it, so half the deck looked one way and half looked rough. I am still not sure which was worse: using AI to write her content, or not using AI to make the ugly half of it look like the good half.
The room was not wrong. Neither was he. They came for something they could use on Monday, and only one of those two sessions gave it to them.
That is the whole thing in one afternoon. He was a breakout, which was right, and he should never be anybody’s keynote. She should never be the person you book to tell a company where this is all heading. Both were right for the session they were in. Neither would have been right for the other one.
The thing nobody warns you about
The AI landscape changes between walking on stage and walking off. That is not a joke about the pace, it is the working condition. A deck built in March is describing a different product by September, and a speaker who built a great tactical session two years ago is showing your people a tool that has moved twice since.
This is not an argument against practical speakers. It is an argument for asking one question before you book one: what changed in your material in the last ninety days, and why. Someone who works in this every week has an immediate answer. Someone who does not will tell you about their framework.
It is also why the deeper material holds its value. What AI is doing to how people work, think, and trust each other does not get deprecated when a model ships.
The three session types, in full
Answer this before you look at a single bio. It eliminates more wrong bookings than any credential check.
Change how the room thinks before it changes what the room does.
Change what people do on Monday.
Go further into one question than a keynote can.
The territory
“AI speaker” is a category about the size of “technology speaker.” Nobody would book that without asking what kind. It is not just ChatGPT and large language models: it is machine learning, superintelligence, ethics, the workforce, autonomous systems. Most speakers are genuinely strong in one or two of these and general everywhere else.
Prompting, automation, agents, what to put in front of a team next quarter. The most requested, and the fastest to go stale.
Machine learning, what a model is doing, why it fails the way it does. The literacy layer under every other conversation.
Adoption, resistance, what AI does to roles, teams, hiring, and how leaders hold a culture together through it.
Bias, surveillance, authorship, superintelligence, what we owe each other. The conversation that outlasts any model release.
What AI means in healthcare, or law, or manufacturing. Someone fluent in the general case can still be wrong about yours.
Ask a shortlisted speaker which of these they would decline. The answer is more informative than anything in the bio.
Ten minutes, public sources only
Not their own site. Press, a company page, a patent, a course listing, a conference program. If the only source for "built AI systems" is the bio that says so, you have found the bio, not the work.
Find the earliest dated AI work you can. This is not a bar to clear and a 2023 start is not a mark against anyone, but it tells you whether you are buying a decade of pattern recognition or two years of fast study. Both are real. They are not the same purchase.
The most reliable signal on this entire list. People who know a field deeply will tell you where their edge is. In the audit, fewer than one in fifty did it in public.
The AI landscape moves between walking on stage and walking off. Ask what they changed in the last ninety days and why. A speaker with a real answer has a process; a speaker without one has a deck.
Same size, same seniority, same format. A speaker who kills in a 40-person workshop is not automatically right for your 2,000-seat general session, and the reverse is just as true.
This is the same rubric the study used, minus the spreadsheet. The full method is published, including the decision rules and a worked example.
Asked often
Look for dated, third-party evidence of the thing they claim: press coverage, a company or product page, patents, peer-reviewed work, a course listing. A bio is not evidence. In a census of 423 AI speakers listed by six major North American bureaus, 45% had no publicly verifiable hands-on AI work and no verifiable disciplined study. That does not mean they have none, it means a buyer cannot find it.
It depends on the session, not the speaker. A keynote needs someone who can change how a room thinks. A workshop needs someone hands-on who will open the tools. A panel needs someone with a real position on one question. Booking across those lines is the most common reason an AI session disappoints, and it is usually nobody’s fault.
Not on its own. Someone who built a consultancy on practical AI workflows since 2023 can be exactly right for a room that needs to rebuild how it works. What matters is matching the depth you are buying to the job you need done, and knowing which you are getting.
What have you built or deployed, and where can I read about it. What do you not cover. What changed in your material in the last ninety days. Can I speak to an organizer from an event the size and seniority of mine. The answers separate speakers faster than any bio.
Published fee guidance for AI keynotes disagrees wildly, and none of the guides discloses a sample or a method. In the audit, published fee bands did not track verifiable evidence in any consistent way. Price signals demand and representation, not depth.
Declaring the obvious
I speak about AI, so this page is not disinterested and it would be insulting to pretend otherwise. Here is the honest version, using my own framework.
I am a keynote. What it means to live and lead in an AI world as human beings: the generational shifts, what gets lost, what leaders owe their people while the tools change underneath them. I also take panels and longer sessions on superintelligence, ethics, and what this does to the workforce.
I am not your workshop. If what you need is somebody to open the tools and rebuild your team’s workflows on a Tuesday afternoon, book a practitioner who lives in them every week. That is a different job and they will do it better than I would.
On my own rubric I am a Practitioner. I pass the hands-on test on third-party evidence dated 2019, and I fail the study test, because a weekly essay series is analysis and not peer-reviewed research. That one is worked through in full on the methodology page, including the test I do not pass.
