Practitioner
17% of those listed
Verifiable hands-on AI work: built, deployed, managed the deployment of, or professionally taught the hands-on use of AI systems, with accountability for results.
A public-evidence study
The AI Speaker Market Audit
Where the expertise of bureau-listed AI speakers comes from, and how much of it a buyer can verify without picking up the phone. 423 speakers, 6 bureaus, no names.
Fall 2026 edition. Bureau listings captured October 1, 2026. Next edition: Spring 2027.
Why this exists
The most requested keynote topic is the least checkable
Of the 423 people these bureaus list as AI speakers, 45% showed no publicly verifiable hands-on AI work and no verifiable disciplined study. Not that they have none. None a buyer can find.
AI is the topic meeting professionals are asked for most, and the number of people marketed as AI speakers has grown faster than any buyer’s ability to check them. There is plenty of advice on how to choose an AI speaker. There is almost no data on what the market actually looks like.
Buyer guides sort AI speakers into archetypes, builder, futurist, academic, practitioner, but the labels are self-applied in the bios and none of the guides gives a buyer a way to check which one fits. This study measures the thing the labels stand in for: where the knowledge came from, and how much of it is publicly verifiable.
-
Buyer guides do sort AI speakers into archetypes, but not the same ones. One bureau guide uses builder, strategist, ethicist and translator; a speaker’s own guide uses futurist, academic, practitioner and vendor-adjacent. The verification each suggests is reels, reference events and a call before signing, which tests whether someone can hold a room. None of them offers a way to check the archetype claim itself.
Aurum Speakers Bureau, April 6, 2026 · Ben Farrell, June 2026
-
A Press Gazette investigation named more than 50 apparently fabricated expert personas, invented people rather than real ones, whose commentary featured more than 1,000 times in UK newspapers, magazines and online titles.
-
Published 2026 fee guidance for AI keynotes disagrees widely. One guide puts established AI practitioners at $15,000 to $30,000, another puts AI and innovation specialists with media presence, books and enterprise rosters at $30,000 to $75,000, and a third spans $10,000 to $50,000. None of the three discloses a sample or a method, and two are speakers quoting the market they sell into.
Susan Sly, July 30, 2026 · Shawn Kanungo, June 2026 · Joel Comm, 2026
-
Peer-reviewed work on generative engine optimization found that adding statistics, quotations and citations was among its most effective techniques, improving a source’s visibility in AI-generated answers by up to 40%: 30 to 40% on position-adjusted word count, 15 to 30% on subjective impression, and up to 37% on a live generative engine.
Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024
Every claim above was re-opened and checked against its own source on October 2, 2026. Two of the fee guides are speakers quoting the market they sell into, and one source is a speaker bureau; they are cited for what they published, which says nothing either way about whether that bureau is among the six in this study. The bureaus are not named.
The framework
Four kinds of AI speaker, and what each one is for
Not a ranking. A fit tool. The question is never which type is best, it is which type your room needs, and the answer changes by event. What a buyer cannot do today is tell which one they are looking at, because every bio uses the same words.
Researcher
22% of those listed
Verifiable disciplined study of AI without hands-on work: academic research, industry analysis, or sustained AI-focused journalism.
Practitioner-Researcher
16% of those listed
Meets both the hands-on test and the study test.
Secondhand Explainer
45% of those listed
Neither test is met in public. The AI knowledge that can be verified appears to come from others’ work, courses, or the tools themselves.
Starting after ChatGPT is not a mark against anyone. The study reports work dated before November 30, 2022 because continuity is a useful thing to know, not because it is a bar to clear. Someone who built a consultancy on practical AI workflows in 2024 can be exactly the right speaker for a room that needs to rebuild how it works. That is a different booking from a professor with a decade of published research, and both can be the right call at the same conference, for different sessions.
What this study adds is the part nobody could check before: how the listed market actually divides, and what evidence sits behind each label.
How it was done
The short version
Every individual listed under an AI category page on 6 major U.S. speaker bureau sites was captured on October 1, 2026, de-duplicated across bureaus to 423 unique people. Every one of them was then coded. This is a census, not a sample: there was no draw, so there is no sampling error and no margin to quote. The full listing text was re-read the following day.
All 423 were coded by hand against public evidence only. No speaker or bureau was contacted and nothing behind a login was used, because the study measures what a buyer can find on their own.
Every record was coded by one researcher against the published rubric, and every record was reviewed by the author. Where a call was close, the rule was written down and the affected records were recoded, which is why the decision log runs to twenty entries rather than none.
Exclusions: 0. No individual was excluded. Ryan Vet appears on none of the AI category pages captured, so no exclusion was needed on his account either.
Practitioner
Verifiable hands-on AI work: built, deployed, managed the deployment of, or professionally taught the hands-on use of AI systems, with accountability for results.
Product or company pages, press coverage of a deployment, patents, code repositories, dated role history, dated course listings.
Researcher
Verifiable disciplined study of AI without hands-on work: academic research, industry analysis, or sustained AI-focused journalism.
A faculty or lab appointment, peer-reviewed AI publications, analyst reports, or a dated body of sourced AI reporting.
Practitioner-Researcher
Meets both the hands-on test and the study test.
Evidence from both of the rows above.
Secondhand Explainer
Neither test is met in public. The AI knowledge that can be verified appears to come from others’ work, courses, or the tools themselves.
Only self-description, general commentary, or content with no verifiable hands-on or research history behind it.
Insufficient evidence
Too little public information to classify with confidence. Reported as its own category and never merged into another.
A one-line listing, no site, and nothing findable.
Continuity is not part of the classification. For every speaker, the earliest date of publicly verifiable AI-related work is recorded, along with whether it precedes Nov. 30, 2022, and it is reported alongside the classification rather than inside it.
The wording matters. Everywhere below, a category or a count means no publicly verifiable evidence of something, never no experience of it. Many strong practitioners work under confidentiality.
Findings
Six questions, six figures
Every chart carries its base, its capture window, and the edition, and every chart is followed by the numbers behind it. No subgroup smaller than ten speakers is reported.
Research question 1
Where the expertise comes from
Every speaker listed was classified by the source of their AI knowledge, using public evidence only. The labels describe evidence, not quality, and not whether someone is any good on a stage. A speaker in any category can be the right speaker for an event.
Of 423 AI speakers listed by 6 major North American speaker bureaus serving U.S. buyers, 17% (73 of 423) showed publicly verifiable hands-on AI work only, 22% (93 of 423) showed verifiable disciplined study only, 16% (66 of 423) showed both, and 45% (191 of 423) showed neither. 0% (0 of 423) had too little public information to classify. (Fall 2026 AI Speaker Market Audit, 423 speakers, captured October 1, 2026; listing text read October 2, 2026)
| Category | Speakers | Share of the group |
|---|---|---|
| Practitioner | 73 | 17% of 423 |
| Researcher | 93 | 22% of 423 |
| Practitioner-Researcher | 66 | 16% of 423 |
| Secondhand Explainer | 191 | 45% of 423 |
| Insufficient evidence Reported as its own category and never merged into another. Every record was resolved, so this category is empty. | 0 | 0% of 423 |
Secondhand Explainer is the largest group. It means no publicly verifiable hands-on AI work and no verifiable disciplined study were found, not that none exists.
Research question 2
Work that predates ChatGPT
Whether the speaker was working with AI before it became a category to speak about. ChatGPT went public on November 30, 2022; everything dated before that counts, whatever the role.
61% (258 of 423) of bureau-listed AI speakers had publicly verifiable AI-related work dated before ChatGPT's public release on November 30, 2022. (Fall 2026 AI Speaker Market Audit, 423 speakers, captured October 1, 2026; listing text read October 2, 2026)
| Category | Speakers | Share of the group |
|---|---|---|
| Verifiable AI work before Nov. 30, 2022 | 258 | 61% of 423 |
| No publicly verifiable work before that date | 160 | 38% of 423 |
| Unclear Evidence exists but could not be dated to either side of the line. | 5 | fewer than 10 of 423 |
| No web archive available for the speaker’s own site context Context, not a verdict. Reported so the archive gap is visible. | 0 | 0% of 423 |
Rows marked context are reported alongside the finding and are not part of it.
Research question 3
What a buyer can check without asking
Three things a meeting professional can verify from a listing and a website, before any call: whether the speaker says what they do not cover, whether client references are public, and whether they describe how the material is kept current.
Of 423 bureau-listed AI speakers, 74.7% published client references, 7.1% described how they keep material current, and 1.7% stated the limits of what they cover. 24.6% showed none of the three. (Fall 2026 AI Speaker Market Audit, 423 speakers, captured October 1, 2026.)
| Category | Speakers | Share of the group |
|---|---|---|
| States the limits of what they cover | 7 | fewer than 10 of 423 |
| Publishes client references | 316 | 75% of 423 |
| Describes how material is kept current | 30 | 7% of 423 |
| Showed none of the three context Context row. Not one of the three checks, but the share that showed no sign of any of them. | 104 | 25% of 423 |
| Showed all three context | 2 | fewer than 10 of 423 |
Rows marked context are reported alongside the finding and are not part of it.
The full-length-video measure was dropped before coding, which is why this is three items and not four. See the dropped-measures note in the methodology.
Research question 4
Superlatives against evidence
Unqualified expertise claims were counted in each listing: leading, top, foremost, world-renowned, pioneer, visionary, AI expert, AI authority, and the rest of the list in the methodology. Third-party awards cited with a source were not counted. The question is whether the strength of the language tracks the strength of the evidence.
Among listings with three or more unqualified expertise superlatives, 62% (48 of 77) showed publicly verifiable hands-on AI work or study. Among listings with none, 51% (78 of 152) did. The difference is not statistically significant (Fisher's exact p = 0.12). (Fall 2026 AI Speaker Market Audit, 423 speakers, captured October 1, 2026; listing text read October 2, 2026)
| Category | Speakers | Share of the group |
|---|---|---|
| No expertise superlatives | 78 | 51% of 152 |
| One or two | 105 | 54% of 193 |
| Three or more | 48 | 62% of 77 |
Each bar has its own base, because the three groups are different sizes. The differences between them are not statistically significant: Fisher’s exact p = 0.12 comparing three-or-more against none. Louder language did not predict weaker evidence, and it did not predict stronger evidence either.
Research question 5
Published fee bands
Whether a buyer can see a price before contacting anyone, and where the published prices sit.
38.6% of bureau-listed AI speakers published a fee band. (Fall 2026 AI Speaker Market Audit, 422 listings readable at the text capture, captured October 1, 2026.)
| Category | Speakers | Share of the group |
|---|---|---|
| Publishes a fee band | 163 | 39% of 422 |
| No fee band published | 259 | 61% of 422 |
Fee bands broken out by source type are reported in the data file only where the subgroup reaches ten speakers.
Research question 6
Disclosed commercial relationships
Whether a commercial relationship with an AI vendor is disclosed anywhere a buyer would see it, on the bureau listing or the speaker’s own site. This measures disclosure, not conflict: an undisclosed relationship may not exist at all.
32% (134 of 423) of bureau-listed AI speakers disclosed a commercial relationship with an AI vendor, on a bureau listing or on their own site. (Fall 2026 AI Speaker Market Audit, 423 speakers, captured October 1, 2026; listing text read October 2, 2026)
| Category | Speakers | Share of the group |
|---|---|---|
| A relationship is disclosed | 134 | 32% of 423 |
| No disclosure found | 289 | 68% of 423 |
No disclosure found is not evidence of an undisclosed relationship. It means nothing was published where a buyer would look.
For meeting professionals
What to do with this
None of this says who to book. It says what to ask, and which answers cost a speaker nothing to give.
-
Ask where the knowledge comes from, then ask for the dated proof.
The four classifications here are not grades. They describe how someone learned what they teach, which is a fair question to ask and a cheap one to check. A speaker who built and deployed systems, a speaker who studies them, and a speaker who explains other people’s work are three different talks, and any of them can be the right talk for a room.
-
Treat a superlative as a claim, not as evidence.
Leading, foremost, pioneer, and the go-to are self-applied in most listings. They are free to write. A dated project, a named deployment, a peer-reviewed paper, or a sourced body of analysis is not.
-
Ask the three questions this study could not answer from the outside.
What falls outside what you cover? Which clients can I call? How did your material change in the last six months? All three are things a strong speaker answers easily and a buyer should not have to infer from a bio.
-
Absence of public evidence is not absence of experience.
Plenty of people with deep hands-on AI work cannot point at it, because it sat inside an employer or under an agreement. If a listing shows you nothing, that is a reason to ask rather than a reason to pass.
Limitations
What this study cannot tell you
- Public evidence understates real experience. People who worked under confidentiality or inside large organizations can have deep hands-on AI work that leaves no public trace. Every finding here is about what a buyer can verify, not about what a speaker has done.
- Bureau AI category pages reflect bureau marketing choices, not a definitive population of AI speakers. Speakers who represent themselves are not listed by a bureau and so are not in this study at all, which is the single largest gap in its coverage.
- Scholarly databases were unavailable during coding. OpenAlex, Semantic Scholar, Google Scholar and arXiv were blocked or rate-limited throughout, so peer-reviewed publication counts were assembled from CVs, anthologies, university pages and publisher listings instead. That almost certainly undercounts publications for some people, which would understate the study test rather than overstate it.
- Coding is judgment applied to a rubric. One researcher coded every record and the author reviewed every record; where a call was close, the rule was written down and the affected records were recoded. A reader who disagrees with a judgment can see the rule it came from and the aggregate it lands in, but cannot re-open an individual record, because record-level data is never published.
- Web archive coverage is uneven. "No archive available" is reported separately and is never treated as a negative finding.
- The rubric measures the source of knowledge. It does not measure speaking quality, audience outcomes, or value for a given event.
- The study is a snapshot of its capture window and will change with each edition.
Dropped before analysis: Public availability of a full-length recorded talk
A fourth buyer-evidence item was planned and tested: a publicly viewable, continuous talk of at least 20 minutes. It was dropped on Sept. 18, 2026, before analysis. Speakers commonly keep full talks private or host them somewhere that is not publicly searchable and send links to buyers directly, so what the measure captured was where a speaker hosts video rather than what a buyer can see. No figure is reported for it, because a low number would read as an absence of footage when it is really a hosting choice.
Method questions
Asked and answered
Who is in this study?
Individuals listed under an artificial intelligence category page on the websites of six major speaker bureaus serving U.S. corporate and association buyers, captured on one date. Every individual in the category is included, whether AI is their main topic or one of several. Organizations, panels, and entertainment acts are excluded.
Is this a ranking of AI speakers?
No. It is an aggregate study and no speaker is named in any output. It does not rank, recommend, or endorse anyone, and it does not measure speaking quality or value for a given event. It measures where a speaker’s AI knowledge comes from, as far as public evidence shows, and how much of that a buyer can check.
How is a speaker classified?
By two tests applied to public evidence. The hands-on test asks whether there is a verifiable role or project in which the person built, deployed, managed the deployment of, or professionally taught hands-on use of AI, confirmed by a third-party or dated source rather than a bio. The study test asks for a research appointment with at least three peer-reviewed AI papers, or at least 24 months of dated, sourced AI research, analysis, or reporting. Both tests passing is Practitioner-Researcher, one is Practitioner or Researcher, neither is Secondhand Explainer, and too little public information is Insufficient evidence.
Does Secondhand Explainer mean a speaker has no AI experience?
No, and that distinction matters. The category means no publicly verifiable evidence of hands-on AI work or disciplined study was found. Plenty of people with real hands-on experience work under confidentiality or inside organizations that do not let them point at anything, and they land in this category. The study reports what a buyer can verify, not what a speaker has done.
Which bureaus were included, and are they compared?
Six bureaus were included and they are not named, because this is not a bureau comparison and no bureau-level result is published. How many of the included bureaus list each person is recorded and reported only in aggregate.
Why is there no figure for publicly available full-length talks?
The measure was collected and then dropped before analysis, on Sept. 18, 2026. Many speakers keep full talks private or host them somewhere that is not publicly searchable and send links to buyers directly, so the measure captured where a speaker hosts video rather than what a buyer can see. Reporting it as a low number would have been misleading, so it is reported as a dropped measure instead.
Can I get the data?
The aggregate counts behind every chart are published as a CSV on the report page. The record-level dataset, which holds evidence links for each speaker, stays private and is retained for 24 months to support corrections and the next edition. No record-level data is ever served.
How often is this repeated?
Every six months, using the same rubric, with change-over-time reporting from the second edition onward. Each edition keeps its own URL, so a citation of this edition keeps pointing at the numbers it cited.
Change log
Every change to this edition, dated
Decisions that changed the study appear here, including the ones taken before publication, because they are what a reader checking the method wants to see.
-
Data frozen. Every record was coded, every decision point ruled, and no test was left Unclear. No exclusions were made. The report and methodology pages were rebuilt to read their figures from the frozen export rather than from transcribed numbers, and the earlier design they described, a 300-person stratified draw from a larger frame, was replaced by what was actually run: a census of all 423 listed people.
-
Report and methodology pages built, with no findings in them. The edition stays noindex until the data freeze.
-
The full-length-talk measure was dropped from the buyer-evidence score, leaving three items: stated scope limits, public client references, and a described update process. Six bureaus were confirmed as the included set.
-
The study test was tightened. Its 24-month branch now passes only on peer-reviewed research, primary research, original sourced reporting, or analyst reports; opinion columns and company marketing blogs do not pass.
Author disclosure
Ryan Vet is a working keynote speaker, including on AI-related topics, and he conducted this study. He is not in it: he appears on none of the AI category pages captured, so there was nobody to exclude, and nobody was excluded. He is measured by the same rubric anyway, and publishes his own result against these tests on his site. This study is not a ranking, a directory, or an endorsement. No speaker and no bureau is named in it, and no bureau-level result is published.
- Not a ranking, list, directory, or endorsement. No individual speaker is named in any output.
- Not a judgment of any speaker’s quality, stage skill, or value. It measures publicly verifiable evidence only.
- Not a bureau comparison. No bureau-level results are published.
Corrections
Corrections to the method or the findings are welcome, including from speakers and bureaus whose listings were captured. Every correction that changes a published number appears in the change log below with its date and reason.
