11% AI Referrals. Christopher Penn on GEO and video.

Key Takeaways

• 11% of Trust Insights' new business now comes from AI referrals. That's the last six months, and Christopher Penn calls the number astonishingly high.

• Anyone selling you AI visibility tracking software is lying. Google personalizes AI search by Gmail, calendar, photos, and YouTube views, so brand recommendations are impossible to predict.

• The only real GEO strategy is to publish video daily. Use questions from your inbox, customer service line, and old webinars as source material.

• Measure AI visibility with a free-form "How did you hear about us?" text field. Analyze it monthly with AI of your choice.

• YouTube trains Gemini, AI Overviews, AI Mode, and Vertex API. If you didn't make the video, you're not in the game.

Why AI visibility software is a $5,000-a-month scam

Christopher Penn says the AI visibility tracking industry is selling something that does not exist. The reason is personalization.

Google now personalizes its AI search results, AI Overviews, and AI Mode based on the rest of your Google account, your Gmail, your calendar, your Google Photos, even what you watch on YouTube. Garrett Sussman demonstrated this at SEO Week by setting up synthetic Gmail accounts with different subscription patterns. The brands Google recommended varied wildly from account to account.

If you are paying five to ten thousand dollars a month for AI visibility software, cancel the subscription. The data the software is pulling is not predictive of what any individual searcher will see.

The only real GEO strategy is daily video

When asked what a mid-market B2B company with 50 to 200 employees should do, Christopher's answer is one word, daily. Publish video every day.

The objection is always the same. We do not have time. We cannot make that much video. Christopher's response is that the source material already exists. It is sitting in your inbox, your customer service queue, your webinar archives, and the list of questions you ran out of time to answer at the end of every webinar you ever ran.

Pull those questions into a Google Doc. Get out your phone. Clip on a lavalier mic. Answer one question per video. That is the strategy. It is also the entire premise of Marcus Sheridan's book They Ask, You Answer, which Christopher recommends so highly he says you do not even need to read it past the title.

How to actually measure AI visibility (since the software won't)

If the dashboards are lying, how do you know if any of this is working? Christopher's answer is to skip the dashboards entirely and add one field to your contact form.

A free-form text field. The label reads, "How did you hear about us?" No drop-down. No multiple choice. Let people type or voice dictate whatever they want. Then every month, run the responses through the AI of your choice and look for one thing, the names of AI products showing up as referral sources.

ChatGPT. Claude. Google. Perplexity. DeepSeek. Alibaba's Qwen. If the count is zero, your GEO strategy is not working. If the count is climbing, it is.

For Trust Insights, that number is now 11% of new business in the last six months. The free-form field is the proof.

YouTube is the data source for every Google AI product

When Google trains its AI, YouTube is the well it draws from. You cannot opt out, because Google owns YouTube and uses it directly.

Look at the third-party AI training opt-out controls in your Google account. YouTube is not listed there. Look inside Gemini's AI Studio at the grounding sources, YouTube is in there. Look at the Vertex API, Google's enterprise search system, YouTube is in there. AI Overviews and AI Mode pull from your YouTube views as part of personalization.

YouTube is also the second largest search engine on the planet. People type "how do I fix this leaking faucet" or "how do I connect my QuickBooks to my CRM" into YouTube every day.

One small action you can take today, in YouTube Studio go to Settings → Channel → Advanced settings, and toggle on "Allow third-party models to train on my channel." When the AI vendors come asking, say yes.

iPhone footage is good enough (with one exception)

For daily content, volume beats polish. But there is one caveat, and it is about audio, not video.

David Tamés, the legendary video teacher at MassArt, said in 2006 that good audio is what great video is made of. Christopher still quotes it today. A 1970s bar TV with working sound holds an audience watching soccer. A silent 100-inch video wall does not. Sound carries attention.

The exception is brand fit. If your brand is built on production quality, like a video agency or a high-end speaker reel, then phone footage will hurt you. People will wonder, should I hire him if his video looks like that? But if your brand sells angular contact ball bearings, the video can look like anything you want, as long as the bearings are visible. Nobody is choosing ball bearings based on video polish.

Get a decent microphone. Match the production quality to your brand promise. Then publish.

Frequently Asked Questions

How do you measure AI visibility for your business?

Add a free-form "How did you hear about us?" text field on your contact form, call center intake, and email channel. Every month, run the responses through AI and look for ChatGPT, Claude, Google, Perplexity, DeepSeek, or Qwen showing up as referral sources. If the count is zero, your GEO strategy is not working.

Can you predict what AI search engines will recommend?

No. Google personalizes AI search results based on the rest of your Google account, including Gmail, calendar, Google Photos, and YouTube views. Two people running the same query get different brand recommendations. Christopher Penn says anyone selling software that claims to predict your AI visibility is lying.

How often should B2B companies publish video for GEO?

Daily. Christopher Penn's answer for mid-market B2B is the same as for everyone else. Pull questions from your inbox, customer service line, and old webinar archives, then answer one per video with your phone and a lavalier mic.

Why does YouTube matter for AI training?

YouTube is owned by Google, and Google uses YouTube as training data for Gemini, AI Overviews, AI Mode, and Vertex API. You cannot opt out. YouTube is also the second largest search engine, where people ask how-to questions every day. If you do not make those videos, your competitors will.

Is iPhone video quality good enough for B2B?

Yes, for most B2B brands. The exception is if your brand is built on production quality, like a video agency or a high-end speaker. For everyone else, get a decent microphone, since good audio matters more than video resolution, and focus on velocity instead of polish.

Watch the shorts

Each short answers one specific question from the interview:

• How do you measure AI visibility for your business?

• Can you predict what AI search will recommend?

• Why does YouTube matter for AI visibility?

Full Interview Transcript

Dane: Hey everybody, my name is Dane Frederiksen. I am CEO of Digital Accomplice, a video creation company in Oakland, California. And I'm joined today by Christopher Penn, who is co-founder and chief data scientist of Trust Insights. So Christopher, what else do we need to know about you and what you're up to these days?

Christopher: Most of my focus these days is on artificial intelligence. I've been working in the field since 2013, back in the days when it was called machine learning, or what we now call classical AI. And obviously, ever since November 2022, when ChatGPT came out, suddenly everyone's an AI expert. You know, a lot of us have been in this space for quite some time.

Dane: So what is your take on what we're supposed to call GEO, AEO? What are we doing calling this thing?

Christopher: It depends on who you ask. There's GEO, AIO, AEO, AIO, do AEO, GAEO, I've seen XEO. I think a lot of the industry has standardized on GEO. That was one of the first terms that came out from a paper in Princeton in 2023. So that's probably the oldest of the terms. It's probably one of the ones folks recognize the most. But honestly, everyone and their cousin's got a different name for it because everyone's trying to find a place to hang their hat as something they can own. And the landscape there has gotten more complex than ever.

Dane: But you didn't answer my question. What's your go-to? What are you calling it?

Christopher: We use GEO because it's probably the oldest of the terms.

Dane: Yeah, okay. So in the world of GEO, with me as a video expert, I'm on this mission to try and figure out definitively, like for the non-technical people, people in leadership positions, what do they need to know about video and how that can intersect with GEO so they can make good decisions about investment and prioritization. Like high level, what do you think that means for them right now?

Christopher: So a few things. First, a paper came out, or study came out in January of this year by SparkToro and Gumshu, I believe it was, that showcased among other things the ability for AI models to recommend brands in the same order is basically zero. They're entirely probabilistic tools. At SEO Week this past Monday and Tuesday, Garrett Sussman did a great piece showing how Google, which is still 93 to 95% of all search and is now like 100% AI search thanks to AI Overviews and AI Mode, Google is heavily influenced as they increase personalization by the rest of your Google account. So if you have personalization for AI for search turned on with AI Overviews and with AI Mode, it starts pulling from things like your Gmail, your calendar, your Google Photos, your YouTube views, etc. And that influences what shows up in your search results. So this has an important meaning for a lot of people. It means this, you cannot predict what brands are going to get recommended. Period. End of story. Anyone who is selling you software saying we can tell you what your AI visibility is, zero. They have zero insight now because the way language models work is, when you Google something now, Google said this in summer of 24, Google does the Googling, and the more personalization data it brings in, the more distortion is in that. In Garrett's talk, he was showing how he set up synthetic Gmail accounts, like five or six fake people, and gave them different YouTube subscriptions, gave them different Gmail subscriptions and things like that. And the brands that Google recommended varied from account to account based on the personalization in there. So anyone saying we know what your AI visibility is is just lying. There's just no polite way of saying they're lying. And if you're spending money on enterprise AI visibility software, cancel the subscription. Save yourself the five grand to ten grand. It's worthless because of that level of personalization. Now, if you are a video producer or in the video industry, the single biggest thing you should be doing right now is loading up your YouTube account as much as you can as often as you can with rich data, video with audio, with closed captions, with solid descriptions, with tags. And then in your YouTube channel settings, there's a — if you go to settings, channel, advanced settings, scroll all the way to the bottom, there's a little thing that says allow third-party models to train on my channel. You want to say yes. It's like I remember the scene from Ghostbusters, the next time someone asks you if you're a god, you say yes. In this case, next time someone asks if they can train on your channel, you say yes. And you allow companies that build models to train on your YouTube data. If you are not — this is what I tell people — the cadence at which you should be loading video to your YouTube channel, if you can, is daily. That's how much video you should be creating.

Dane: So what does this mean for like a typical B2B brand? Maybe not enterprise as much, or small startups, like, I don't know, 50 to 200 employees, just sort of like middle of the road B2B company. What should their video strategy look like right now for GEO?

Christopher: Be publishing daily. That's the answer. And here's the thing, in terms of people like, oh my god, what do I do? I don't know how to create that much video. Like that seems impossible. No, it's not. Here's how you do it. You go to your inbox. You go to your customer service line. You go to wherever it is people ask you questions. You extract out the hundreds of questions that you inevitably get. Hold a webinar. Look at your webinar archives. Look at all the questions you didn't get to in your webinar. You put this all in a big-ass Google Doc, and then you get out your phone or your camera or whatever, you've got clap-on, stick on your lavalier mic, or if you want to be a youngster, hold it in your hand, which is exactly how it was not designed. And then you answer those questions. And once you put up one video a day, today, "Someone asked me this question today, Dane asks how can enterprise B2B companies, or how can mid-market B2B companies be using video for GEO?" And then you answer the question. This goes back to a book that Marcus Sheridan published, I think about 10 years ago now, which is so good you don't have to read it. The title of the book was They Ask, You Answer. Like, that's it. You don't need to read the rest of the book. I mean, buy it to support Marcus, but you don't need to read the rest of the book. That is your strategy. If you are any brand, look at all the questions you're asked, get out the video device that you have, and answer the questions.

Dane: So then how do we, if we go about doing that, and if it's impossible to know how we're going to show up in AI search, how do we measure the effectiveness of what we're doing? Are we just taking it on faith that this is best practices, eat your vegetables, it'll work eventually?

Christopher: No. The number one, the gold standard for measuring AI visibility is this. On your points of intake, your contact form, your call center, your email box, or whatever, there's a little box that says, "How did you hear about us?" Let it be a free-form text field. Don't give people a drop-down. Give them a little free-form text field. And then using the AI of your choice, analyze the results every month. And if the number of people who said ChatGPT or Google or whatever the answer is zero, then your GEO strategy is not working. But you should start to see Claude and ChatGPT and Alibaba's Qwen and DeepSeek and Google as referral sources. And if you let people physically type or voice dictate into that, you will get more insight into how they're doing it. 11% of Trust Insights business comes from AI referrals now. That was in the last six months. It is astonishingly high.

Dane: So do you have a sense of like, for the companies that are like, "yes, we want to do more with video, but I can't get my subject matter experts on camera, they don't have time," is there some sort of data-based argument we can put in front of them to be like, "hey, I know you don't want to go to the dentist, but you're going to have 30% less cavities if you do"? Or is there something we can point to about why you should do this?

Christopher: Other than the obvious, which is this is literally how AI models — and feel free to send them this interview — but that's fundamentally, we know where Google gets its data from. And now with a lot of these new revelations about how Google in particular with AI Mode is getting its answers, it comes from things in Google properties. So you should be sending people, with their permission, email. And guess what? When you put a YouTube link in an email — I'm sure you've seen this in Gmail — what does it show at the bottom of the email? A link to the YouTube video with a preview of it, which of course is pulling from YouTube's metadata because Google owns it. When you look at that third-party AI training thing in Google, guess which company isn't listed there? YouTube. Google. You can opt out of it — there's no way to opt out of it because Google uses that as training data. When you look inside Gemini in AI Studio, where Google grounding sources come from, guess what? YouTube's in there. You look in the Vertex APIs, which is Google's enterprise search system, guess what's in there? YouTube's in there. So there's a lot of different contact points where you can illustrate, here, we know Google is getting its data from YouTube, which is, by the way, the second largest search engine on the planet, and is where people go to things like, "how do I fix this leaking faucet?" Right? "How do I do this? How do we that?" I guarantee you, if you are in any company of any kind, someone is answering those questions, because that's exactly what a searcher on YouTube is going to type in. "How do I connect my QuickBooks to my CRM?" Right? Someone is answering the question. If that's what your company does and you didn't make that video, you're not in the game.

Dane: Yeah. So then let's talk about how much. How much can we say video is a part of that game? So let's say I've got a million dollars to invest in content. How much of that should I put towards video and how much should I put towards writing and AI tools and other stuff? Like, is there a way to think about that? Like high level, how many wheels of the car is that?

Christopher: Here's the thing about video. Video is the highest density information source. If I sit down and I work, as I do, like when I make my videos, I make my daily videos in batches, and make them typically batches of 10 to 20. On Sunday, what I do is I take my phone, which has a little magnetic backing, I stick it to my refrigerator, I clip on a little lav mic, and as I'm cooking the Sunday meal, I just answer piles of questions. When I do that, I'm creating a lot of very rich data. So that goes on YouTube. But I can also, and I do, I use a tool like FFmpeg to rip the audio track out of that. And guess what? That goes up as a podcast. I put the audio track into Parakeet MLX, which transcribes it, and guess what? There's a blog post. And then the compendium of those blog posts is what I put into Claude code to say, "okay, assemble a book out of this. You have to use all my original words, but assemble a book out of this." So it's not AI slop. It's literally my words just re-sequenced better. I can take that transcript and I can put it into this Claude coworker, whatever, say, "okay, you have 1,200 words, because that's the length of a LinkedIn post, turn this into a LinkedIn post," and so on and so forth. So if you start with video, if you put like half your budget into video production at velocity, you will create so much stuff that you can then spin into other formats without having to reinvent the wheel constantly. Now, the key here is velocity is volume. It is tempting, and there's a time and place for super high quality polished video. You do a formal shoot. Like for example, I'm a professional speaker. I do a lot of keynote talks. There is 100% a place for someone with a RED camera to be doing a polished speaker reel for me. But for your daily day-to-day video, this is good enough. The camera on this thing is good enough for that. And in fact, in some algorithms like Instagram, believe it or not, the crappy video — they have a problem with a lot of AI slop, and so they're starting to tune their algorithm to focus more on quote, "authentic video," like as if AI can't generate that too. But if you want to produce at scale, that should be your focus, not just high quality, but good useful content with high velocity, with high frequency.

Dane: As a closing thought on that point, is there any data to support this idea of, like, iPhone footage is good enough, like it won't hurt your brand? Because a lot of people will say, "hey, this isn't good enough, you have to have high enough polish, otherwise we're going to look bad or it's going to hurt us somehow." Is there any data to counteract that thinking?

Christopher: So Instagram's algorithm actually is starting to favor that. But here's the thing. Well, there's two things. Number one, David Tamés, who is a legendary video teacher, teaches at MassArt, said in 2006, and I still take this lesson to heart today, "good audio is what great video is made of." If you go to a bar and there's a crappy TV from the 1970s but the sound is working, great, people will watch soccer on a fuzzy TV because they can hear it. If on the other hand, you have a giant 100-inch video wall and no sound, people will look at it, but then they will just not pay attention to it. And so I always tell people, get yourself some kind of decent microphone. And then the video quality depends on the rest of your brand. So for you, Dane, as a person who is a professional videographer and video creator, shitty video is probably not going to help your brand, right? Because people are like, "well, should I hire him if his video looks like that?" Right? You want to be capturing a raw log and you want to be editing in DaVinci Studio on the quarter million dollar deck and stuff like that, and have, you know, what is it, 10-bit 8K log footage and all that stuff. On the other hand, if you are a B2B company that makes angular contact ball bearings, your video can look like whatever you want, as long as people can see the ball bearings. Because no one is buying ball bearings from you because of your video quality. People are buying ball bearings because your ball bearings are better than someone else's.

Dane: Yeah, that's great advice. Well, Christopher, thank you so much for your insights today. I appreciate it.

Christopher: Thank you for having me.