The 2026 State of AI Search for Speakers is an original research report on whether AI can verify who top speakers are, what they’ve written, and what their stage work actually looks like, and what happens when it can’t.
Key Findings
I audited 94 top speaker websites and ran live tests across four AI tools. Here’s what I found:
- A third of top speakers (35%) have zero structured data on their homepage. No machine-readable link between them and their books, their credentials, or their stage work. While this doesn’t make them invisible to AI (their brands often “overcome” this), it is not helping them.
- Not one site in the sample had Book schema. Zero. That means there is currently no reliable way for AI to verify that a bestselling author actually wrote the books they’re known for.
- Only 2.5% had VideoObject schema, the tag that lets AI understand a demo reel actually shows real footage of someone speaking. Every speaker leans on their reel to get booked. Almost none of them have made it legible to the systems now involved in vetting who gets hired.
- When I asked four different AI tools the same questions about eight well-known speakers, one AI model invented three books for Malcolm Gladwell that don’t exist, and dropped his real, most recent bestseller entirely. The two AI tools that actively search the live web got it completely right, every time.
- The fix is not a redesign. It’s a well-documented, low-cost layer most sites simply haven’t added yet: telling AI (clearly) who you are and what you do, in a language it can understand.
- This is purely about the technical aspects of your website. There is much more when it comes to AI search and being recommended by AI engines for speaking gigs, but these are the fundamentals that you want to get right. As you’ll see, even big names that aren’t doing the work are having incorrect information shared about them.
The rest of this report walks through the data, the live AI test, and what a speaker should actually do about it.
Quick Glossary
A few technical terms come up in the findings below. Here’s what they mean in plain language before you hit them:
- Schema (or “structured data”): Invisible tags added to a webpage’s code that tell computers, in a standardized format, exactly what’s on the page. Without it, a computer has to guess from ordinary text, the same way a human would.
- JSON-LD: The specific technical format most schema is written in today. It is simply the language structured data speaks.
- robots.txt: A small file on every website that tells automated visitors, including search engines and AI crawlers, what they’re allowed to look at. A site can accidentally block AI tools from reading it entirely without anyone noticing.
- llms.txt: A newer, optional file some websites now offer: a short, curated map of a site’s key pages, written specifically to be easy for AI systems to read.
- RAG (Retrieval-Augmented Generation): Describes an AI tool actively searching the live web before answering, rather than relying only on what it learned during training. This is the single biggest predictor of whether an AI gets information right.
- Hallucination: When an AI states something false with total confidence. It is the most dangerous kind of AI error, because it is often indistinguishable from a correct answer unless checked.
The Website That Started This
I was reviewing a well-known speaker’s website and noticed something odd. This person has published multiple books, a real, substantial catalog built over years. But nothing on the page told a computer what those books actually were: there was no structured data, no way for an AI system to confirm that this person wrote what they say they wrote.
And once I started checking further, this turned out to be a much larger problem than I expected, close to universal, in fact.
Why This Matters for Booking, Not Just Search
It’s tempting to file this under “SEO problem.” It’s not that.
If you find yourself saying, “my agency takes care of that”, you should check! If you’re wrong, you may be missing opportunities.
More event planners, in-house talent teams, and executive assistants are now using AI tools to shortlist and vet speakers before ever picking up the phone, and that shift is only going to accelerate. This will only continue to evolve, as planners start using dedicated agents as “scouts” to cut down on their workload of identifying prospects and keeping up with the conversation.
If an AI assistant can’t verify a speaker’s books, credentials, or stage footage from their own website, one of two things happens: it falls back on outdated, third-party, or incomplete information, or, as this report shows, it sometimes just makes something up.
For many speakers starting out? Not having this in place increases the odds you don’t show up when someone asks an LLM, “Who do you recommend for a speaker?” These speakers have huge brands and you are still building yours.
You want to influence (as much as possible) the moment a booking decision is being explored. You want to be visible and recommended.
Jonathan’s report puts data behind what I’ve been telling speakers for months: AI is now part of how you get booked, and it isn’t reading your website the way a human does. In a market flooded with polished fakes, the speakers who can be independently verified are the ones who’ll survive the shift.
Methodology
I audited 94 speaker websites, drawn from Global Gurus category rankings (Leadership, Motivational, Coaching, and related lists) plus additional widely recognized names across the professional speaking industry. 79 sites returned usable data; the other 15 failed outright due to broken or unreachable domains, itself a small but telling finding on its own.
Methodology note: This audit checked homepage-level structured data only. Some speakers may have Book or VideoObject schema on dedicated sub-pages not captured here, so these figures likely represent a best-case reading of what a first-time visitor or AI crawler encounters, not an exhaustive, site-wide audit.
For each site, I checked six categories of structured data, chosen because each one answers a specific question an AI system, or a human using one, would actually ask about a speaker:
- Person schema confirms who someone is: their name, role, credentials, and professional identity, in a format a machine can verify rather than infer from prose.
- Book schema connects a named person to a specific published book, complete with fields like ISBN, publisher, and publication date, which is the only reliable way an AI system can confirm that a book actually belongs to the speaker claiming it rather than to someone else entirely, or to no one.
- CreativeWork schema is the broader tag for authored work that isn’t a formal book: articles, courses, podcast episodes, whitepapers, even a named framework or methodology. This one matters just as much as Book schema, arguably more for speakers who don’t have a published book yet but do have a course, a podcast, or original IP worth crediting to their name.
- VideoObject schema tells an AI system that a video embed is genuine footage of a specific person speaking, rather than an opaque file it has no way to interpret.
- Event schema verifies that a specific speaking engagement, past or upcoming, actually happened or is actually scheduled, which matters enormously to a planner trying to confirm someone is active and in demand. Unlike Book or Organization schema, Event data doesn’t need to link to a live web page to be valid, since conference and agenda URLs change or disappear year to year, so the date, location, and named organizer carry the verification weight instead.
- Organization schema links a speaker to the company, agency, or brand they’re formally affiliated with, whether as a member, employee, or founder, which matters for credibility and for connecting a speaker’s site to any other properties they run. This is distinct from simply having spoken at a company’s event, which belongs in Event schema instead.
I also checked whether each site’s robots.txt file blocks major AI crawlers such as GPTBot, ClaudeBot, PerplexityBot, or Google-Extended, and whether each site has a genuine llms.txt file, a newer format explained in the glossary below. I validated the llms.txt results carefully to rule out false positives: several sites initially appeared to have one, but were actually serving a generic “page not found” page that returns a false success signal, or simply redirecting the request back to the homepage.
Separately, I ran a live test: the same two questions, asked of four different AI tools, about eight speakers chosen for contrast between heavy names with little or no schema and a few with some already in place.
Finding One: A Third of Top Speakers Are Invisible to AI at the Structural Level
| Schema type | % of sites with it | Count |
|---|---|---|
| Organization | 35.4% | 28 / 79 |
| Person | 21.5% | 17 / 79 |
| VideoObject | 2.5% | 2 / 79 |
| Event | 1.3% | 1 / 79 |
| Book | 0.0% | 0 / 79 |
| CreativeWork | 0.0% | 0 / 79 |
More than a third of the sites audited carry zero structured data of any kind, and not one of the 79 sites had Book or CreativeWork schema present. That means there is, at the homepage level, no machine-readable link between these speakers and the books that built their reputations, and several instantly recognizable, bestselling names sit in this group.
Finding Two: The Video Blind Spot, and Why It’s a Huge Miss
This is the finding that should sting the most, because it’s the one closest to the actual product a keynote speaker sells.
Only 2 of 79 sites had VideoObject schema, the structured data that tells an AI system this is real footage of this person speaking. Every one of these speakers leans on a demo reel to get booked, and reels get built, edited, and proudly featured on homepages precisely because they’re supposed to do the selling. Yet 97.5% of the time, that reel is functionally invisible to the AI systems increasingly involved in vetting who gets booked, because AI cannot watch a video the way a person can. Without VideoObject schema, and ideally a written transcript alongside it, a speaker’s most persuasive asset, the thing that actually shows what they’re like on stage, might as well not exist to an AI system, no matter how good the footage is.
The one genuinely encouraging data point in this whole dataset: the one site in the sample with VideoObject schema properly implemented (Adam Mendler’s) showed every single AI tool I tested consistently finding and citing his specific, real video page by name. That’s a small sample, but it’s a repeatable, verifiable signal that this particular fix does exactly what it’s supposed to when someone actually implements it.
Adam Mendler, whose site is the one from this study where it clearly worked, put it this way:
I’ve spent years interviewing thousands of the most successful leaders and sharing what I’ve learned with audiences around the world. My goal is to positively impact as many people as possible through Thirty Minute Mentors and through my keynote speaking, so it is great to know that my site is helping my work reach people effectively.
Finding Three: AI Crawler Access, Mostly Open, With Pockets of Self-Inflicted Invisibility
Of 61 sites with a detectable robots.txt file, 9 (14.8%) block one or more major AI crawlers outright. This may be an accident, a security plugin default, or a hosting setting nobody ever reviewed, rather than a deliberate choice, but the effect is identical either way: these speakers don’t have live AI visibility for those Large Language Models from their website.
Finding Four: llms.txt, Early, Genuinely Useful, Not a Silver Bullet
15.2% of sites (12 of 79) have a real llms.txt file, a curated, plain-text map of a site’s key content built specifically for AI tools to read. The best example in this dataset is Simon Sinek‘s site, which cleanly separates content into labeled sections for books, courses, and podcast episodes, giving an AI system an organized map instead of a wall of undifferentiated links. Here’s what it looks like:
It’s worth being precise about what this file currently does. Independent research auditing over 1,400 llms.txt files found they genuinely do get indexed, sometimes rank in Google, and can surface directly in AI answers. Google has since clarified its own position too: it doesn’t use llms.txt as a ranking signal, but it’s “completely fine” to maintain one for other AI tools that do use it, a softer stance than its earlier, more dismissive guidance.
While you’ll get a variety of opinions, I’ll give you my perspective: it’s worth doing regardless, and not because it changes what gets cited today. As AI agents become a normal part of how people research, plan, and book, the sites that already speak their language will be the ones agents route through cleanly, while everyone else gets skipped or misread. This is future-proofing work.
This kind of trend has direction and an opportunity cost. The time investment is small, and it may put you ahead of those who wait.
Both humans and AI systems are going to need a fast, accurate way to understand who you are, and the sites building that now are simply going to be ready sooner, assuming LLMs.txt becomes more widely adopted.
The Real Test: What Happens When You Actually Ask AI About These Speakers
Numbers about schema are one thing. What actually happens when a person asks an AI tool about a speaker is another. So I ran a live test: the same two questions, tell me about this person’s books, and what do they speak about and where can I find video of them, asked of four different AI tools, for eight speakers chosen to contrast heavy names with little or no schema against a few with some already in place.
The Headline Result: it’s not Really About Schema Alone: it’s About How the AI Tool Works.
| Speaker | A general-purpose AI (lighter model) | A general-purpose AI (flagship model) | AI search tool #1 (live web search) | AI search tool #2 (live web search) |
|---|---|---|---|---|
| Tony Robbins | Accurate | Accurate | Accurate | Accurate |
| Brené Brown | Accurate | Accurate | Accurate, missed her newest 2025 release | Accurate, missed her newest 2025 release |
| Malcolm Gladwell | Invented three books that don’t exist, omitted his real 2024 bestseller | Omitted his real 2024 book, but clearly flagged uncertainty rather than inventing anything | Fully accurate, included the 2024 book | Fully accurate, included the 2024 book |
| Daniel Pink | Accurate | One likely invented minor title, omitted a major real book | Fully accurate | Fully accurate |
| Angela Duckworth | Accurate | Accurate | Accurate, missed her newest 2026 release (Situated, Sept 2026) | Accurate, missed her newest 2026 release (Situated, Sept 2026) |
| Simon Sinek | Accurate | Accurate | Accurate | Accurate |
| Grant Cardone | Accurate | Accurate | Accurate | Accurate |
| Adam Mendler (rare VideoObject schema) | Correctly recognized he isn’t a traditional book author | Same | Same, and correctly surfaced his specific real video page by name | Same, and correctly surfaced his specific real video page by name |
In the original test, the general purpose (lighter model) invented three Malcolm Gladwell books that don’t exist (“The World According to Story,” “The Disinformation Age,” “The Breakthrough Idea”) and omitted his real 2024 bestseller, Revenge of the Tipping Point. In a follow-up test, the same model made a different error: it attributed The Glass Hotel, a 2020 novel by Emily St. John Mandel, to Gladwell, and again left out Revenge of the Tipping Point. Two tests, two different confident errors, the same real book missing both times.
The clearest, most useful finding in this entire report is that the tools which actively searched the live web before answering were consistently the most accurate across the board, including one case where an AI tool caught and corrected a piece of bad information circulating elsewhere online, a book wrongly attributed to Adam Mendler by a third-party bureau site, which the tool identified and corrected on its own.
The Gladwell example above is the clearest illustration of this, and a second attempt to reproduce it later surfaced a different but equally confident error involving the same author.
Omission, leaving out something true, usually a very recent release, is not dangerous, just incomplete. Fabrication, stating something false with full confidence, is the dangerous one.
The practical takeaway for any speaker is that you don’t get to choose which AI tool a meeting planner happens to use. Some tools are simply safer than others by design, and a speaker has zero control over that.
Structured data is one thing a speaker can control, and if you can help influence how you show up (or not), why wouldn’t you?
Moreover: left to its own devices, AI will make up things. It’s not magic, and doesn’t just “figure out speakers”: you have to set it up for success as much as possible.
What Good Actually Looks Like
Two sites in this dataset stood out for having a real, well-built llms.txt file rather than a generated afterthought. Simon Sinek’s site separates content into clearly labeled sections, Books, Optimism Press titles, on-demand courses, podcast, giving an AI system an organized map rather than a wall of undifferentiated links. Dorie Clark‘s site does the same, with a clear one-sentence description of who she is and what she does at the top of the file. Neither is doing anything exotic. They’re doing the basics, and in this sample, the basics put them ahead of roughly 85% of their peers.
What This Means
This isn’t a story about a handful of speakers having messy websites.
This is a story about an entire profession that sells personal credibility for a living, most of which has, so far, treated its own website as a brochure rather than as the machine-readable record an AI system now needs in order to recommend someone with confidence.
That gap is real, but it isn’t hard to close – it takes time, understanding, investment and commitment.
The parts measured in this report (Person, Book, Event, and VideoObject schema) are well-documented, low-cost additions to any website, not a rebuild, not a redesign, just a layer added underneath what’s already there (assuming you have a proper, multiple page website).
Schema is one piece of a larger picture, not the whole fix.
How consistently a speaker’s name, credentials, and books show up across bureau listings, third-party profiles, and citations elsewhere on the web matters just as much as what’s on their own homepage, and this report didn’t measure that side of it directly. Though, I see the impact that it has with my clients through the Speaker AI Visibility Audit and beyond.
In the one case in this sample where VideoObject schema was properly implemented, every AI tool tested consistently found and cited the real video page by name. That’s a single data point, not a trend line, but it’s a repeatable, verifiable one; and it suggests speakers who close this gap first may get a real, if temporary, edge over peers who are trusting reputation alone to carry them into an AI-mediated future.
Done as part of a broader effort to help AI understand and recommend you, it can be a powerful catapult. You’re likely already doing the work, the question is if you’re capitalizing on all your time and effort in a way that AI understands.
Potential Limitations
A few limits worth stating plainly:
- Homepage-only audit. Some sites may have Book or VideoObject schema on dedicated sub-pages not captured here. These figures represent a best-case reading of what a first-time visitor or AI crawler encounters, not an exhaustive, site-wide audit.
- Single point in time. AI models and their behavior change. A result recorded here may not reproduce identically if the same question is asked again later.
- Curated sample, not a random one. The 79 sites came from Global Gurus category rankings and additional widely recognized industry names, not a random sample of all professional speakers.
- Schema measures machine-readability, not outcomes. This report shows what AI tools can and cannot verify from a site’s structured data. It does not measure booking rates, recommendation frequency, or downstream business impact directly.
So What Does This Mean For You?
If you’re an established, well-known speaker, the risk is complacency. A recognizable name and a strong reel feel like enough, and for a human audience, they mostly are.
But this report shows that reputation alone doesn’t automatically translate into something an AI system can verify, and a handful of the biggest, most bankable names in this dataset carry zero structured data at all.
A big brand may not be enough. The Gladwell example is the clearest warning here: fame didn’t protect him from being misrepresented by one of the tools tested.
Being well-known gets you into the conversation, but it doesn’t guarantee AI will describe you accurately.
If you’re a speaker in the $2,500 to $30,000 range, this is where the finding actually gets interesting, because it’s one of the few places in this industry where the playing field is closer to level than it looks.
You’re not going to out-brand Tony Robbins or Malcolm Gladwell, and you don’t need to.
What this data shows is that most of your competition, at every fee level, hasn’t done this work either. Structured data doesn’t care how famous you are.
A well-tagged Person schema, a properly linked Book or Event schema, a transcript sitting behind your demo reel, all of it reads the same to an AI system whether you’ve given ten keynotes or ten thousand.
You’re very likely already doing the underlying work: building a reel, writing a bio, listing your speaking topics, collecting testimonials.
The problem is that the work you’re doing likely isn’t being translated into a form AI can actually read and cite.
Treat this as one of your standing visibility lanes rather than a one-time fix, because the AI tools sitting between you and your next booking are only going to matter more, not less, and right now, showing up well here costs very little and almost nobody else at your level is doing it.
Thank you for reading this report. It is my hope that you can take the learnings and apply it. Learn more about how I help speakers grow their business.
Need more opportunities?
See how you show up in AI search.
Jonathan Dunnett helps keynote speakers get booked on purpose. He is a Council of Competitive Intelligence Fellow and brings systematic, intelligence-driven frameworks to speaker business development. Serendipity isn’t a strategy.
