Subscribe
Share
Share
Embed
Zero Click Marketing is a marketing strategy podcast about content marketing, audience research, and how brands grow when clicks matter less. Hosted by Amanda Natividad, Chief Evangelist at SparkToro, the show explores how marketers reach audiences, build influence, and earn attention in a zero-click internet. New to the show? Start with Episode 2: What Zero Click Marketing Actually Is.
[00:00:00] Kevin Indig: trust is not just like a, you know, like a, like a conversion booster, if you will. I think it is the most important good in this AI world since content is easier and faster to create than ever before, and we're competing with more synthetic content. So if you are a person, a creator, or a, a company, and I, I I think you wanna be very strategic and systematic about building and, maintaining trust.
[00:00:26] Amanda Natividad: I'm Amanda Natividad and welcome to Zero Click Marketing.
[00:00:32] Amanda Natividad: Kevin, we've been talking about GEO, AEO, LLM, AI stuff for a while on this podcast. What a great intro, right? I think we're at the point though where everyone thinks they're supposed to care about it. Some people don't care, some people do care a lot. But I think here's my point of view on it.
[00:00:57] I think if nothing else, a lot of what generative engine optimization touches on is the stuff like good content marketing and good digital PR. So for me, it's a little bit of like, I guess we're finally getting executives to really care a lot about relevant, resonant marketing.
[00:01:17] That's one way I see it. And anyway, you've been doing a ton of research in this space. I've been reading all of it. I guess one question would be how do you do all this research? No, but one, I think, could you take us a little bit through, I guess, the cliff notes? Like, what have you been learning this past year as you've been diving deep into all of this?
[00:01:39] Kevin Indig: Yeah. Thank you so much, first of all. it's good to be here, good to talk to so here's kind of the 360 degree, overview of what's going on. I'll try to not turn it into a TED Talk, so first of all, I wanna say, like, I have trouble picking a name. I'm not happy with any of the names that we give it, so just know that. second of all, the way that I think about... I'll just call it AO, but everybody should know this. I don't know of any of the names, but, I'm, we're gona pick one so we can have a conversation that is not too meta. the way that I think about AO is basically a 10,000 piece puzzle that has no blueprint and that changes parts every week or at least every month, right? So big black box, a lot of things that we don't know, a lot of constant change Now, across my research, and I've done like basically two types of research. One is data analysis, where I was generously provided, with large data sets, you know, um, citations, prompts, all that good stuff. And then on the other hand, I have also conducted user behavior research, where we recruited panels of people, average people, not technical people or marketers, and we observed their behavior. For example, when they prompt an AI mode or when they search on Google, and we recorded them, we had them speak out their thoughts loudly, so we can, we can track it and trace it and evaluate it, and then basically, do an actual user research like you would do in a lab. and so what the kind of the gist or the cliff notes of where we are today based on that research is that, A, people don't click. Uh, that probably fits the name of the podcast pretty well, but we really are steering towards a zero click landscape. And that means not clicking outside of LMs or out of Google. clicking through to other websites is what I mean when I say they don't click. they will still click buy or purchase buttons, but, not only has the number of clicks decreased steadily, but research from me, but also from institutes like Pew show that people generally don't click citations, right?
[00:03:41] They don't click sources, only in very few cases. they basically get the answer directly now as opposed to having to click to a bunch of links and then, like filter sources directly themselves. So that's number one. Two, trust is really the most sacred ingredient in all of this. trust matters profoundly because it decides what people believe from the AI answer, and it also decides on whether people, for example, double check AI answers on YouTube or Reddit. and then lastly, there's so much variance in how models work, across the common surfaces, right? So Perplexity works very differently than ChatGPT, works differently than AI mode, et cetera, But also within the models, right, every new version has different behavior, and whether you, for example, use a high reasoning or low reasoning also leads to different results.
[00:04:34] So the black box factor is incredibly high. And with that, I'll finish my, or pause my TED Talk for now.
[00:04:42] Amanda Natividad: No, that's super helpful. That's a really good high level, I think, of everything you've been researching. So we have nobody's clicking, trust is the most sacred ingredient, and there is so much variance, not even just from tool to tool, but within each tool. Which makes sense, right? Because even if you think about your own ChatGPT behavior, you might be using a different model in the app on your phone compared to on your desktop, because on your desktop you might be like, "Oh, I want critical thinking here.
[00:05:11] I want the long form stuff." I want to talk about ghost citations, because this dovetails nicely with a nobody is clicking piece. what are ghost citations? And would you say ... Well, this is a leading question. Maybe I'll just say my little, little intro here is I feel like ghost citations are basically like the AI version of zero click influence, where you're getting the, the brand is getting some of the exposure, but very weak attribution.
[00:05:42] So tell us about ghost citations.
[00:05:44] Kevin Indig: so essentially ghost citations is when your brand is cited as a source but not mentioned in the AI answer. so when you take a step back in terms of how LLMs work, there's different steps that they go through, provided that a user asks a question that they need sources for, right? When they need sources, they first retrieve the source, meaning they, they crawl the page, they index the content, et cetera,they process it. Then they decide to cite a source, and then they mention it in the AI answer. Those are the theoretical steps. Now, n- not every brand, or better said, every source is mentioned as-- i- is mentioned in the AI answer, and not every retrieved result is even cited, right?
[00:06:24] So, when we talk about that process, AI will retrieve a lot of content to, like Google, figure out which,content has the best answer. and then it will decide to cite certain sources, and then it will decide what to mention in the AI answer. and none of these are the same. As a matter of fact, I found that, most of the time, I think it was only 13% of brands get both cited and mentioned, okay? There is a, a distinct difference in business model that matters here. So a, a company like G2, for example, which is a, a software review platform, they have very low chances to get mentioned. They have incredibly high chances to get cited, when it comes to software prompts, because most of the time when people prompt something like, "What is the best CRM?" most of the time, models will not say, "According to G2." They will just say the best CRMs are, I don't know, HubSpot, Salesforce, et cetera. So these are the brands that are mentioned. But they will a lot of times cite G2 because it, it looks a lot at G2 to form its answer. So for a business like G2, being cited a lot and not mentioned as much makes a lot of sense.
[00:07:30] But if you were a SaaS company or if you were a direct to consumer, shopping company, you wanna be mentioned as much as possible because the research shows pretty clearly that less than a percent of people even click on citations or pay attention to them. Only when we talk about high risk prompts, right?
[00:07:49] When you think about medical prompts or insurance or finance or s- things like that, that's when people fact check a bit more. But across the board, people just don't click much on citations, but they very much pay attention to what's being mentioned in the, in the answer.
[00:08:04] Amanda Natividad: Oh, that's super interesting. Thank you for laying it out that way, because I was about to ask for an example, and the G2 one is a really good example of a brand that is cited a ton because it's getting scraped essentially for the actual answer, but it's not getting mentioned a lot because in the context of the query, what's the best CRM software, the users thinking about the CRM software, not the source from which they would find the answer.
[00:08:32] Like, it's just not how people think. Huh.
[00:08:35] Kevin Indig: Exactly. I mean, sure, Salesforce uh, will probably be mentioned and cited, but to a much, much lesser degree than a, than a G2. and at the same time, you know, there are domains that, that get, used a ton for the answer that get barely any visibility like Reddit. Reddit gets some citations for sure, but they're being retrieved so many more times they're getting citations for, and they're barely mentioned in the answer at all
[00:09:00] Amanda Natividad: I mean, that's, a, that's similar to LinkedIn, right? Like, they're getting retrieved a lot, not mentioned a lot because it's the content itself that's getting mentioned or cited.
[00:09:12] Kevin Indig: exactly. there's also difference in prompt length. So,
[00:09:16] for short prompts, we see a lot of brand mentions. For long prompts, we don't see a lot of brand mentions. It probably has to do with the fact that longer prompts indicate more of what you're looking for, so AI can be much more targeted with its recommendations.
[00:09:27] Whereas if you have a short prompt, like just best CRM or something, the AI wants to show you a lot of different options so you can find the best one
[00:09:35] Amanda Natividad: Cool. All right.
[00:09:36] I have a question here on the way AI reads or takes an information. this also comes from the research you did where you analyzed over a million search results to understand where and how AI systems seem to read content. And so the framing seems to be like AI behaves more like an editor, like a busy editor than a patient student.
[00:10:01] So I guess my question is, if that's true, that framing's correct, if AI reads like a busy editor, what should marketers stop doing now or start doing instead?
[00:10:13] Kevin Indig: So you're correct. AI reads more like an editor, meaning there's, there's a lot of top heaviness. The intro matters a lot. and then there's another bump in attention for the conclusion. But essentially, AI, I mean, it ingests so many tokens, and there's so many kind of ways that AI has to make sure that the answer it gives is ideal and optimal, that it will often scan the content, first of all, by markers to see if the content has what it needs, because retrieval and extraction is so expensive. second, it will pay most attention to the, to the beginning of the content because that usually sets the tone for the rest of the content, right? And so the way you wanna think about ideal writing is you wanna lead with a punchline, right? Like classic journalism.
[00:11:00] lead with a punchline and bring it straight to the point. AI writing is not big on storytelling and fluff, right? Uh, the ideal AI writing is just a table of facts, maybe with some context around it. So pretty bare bones, not too much catch up, just the hot dog. you can work with things like, you know, TLDRs, executive summaries, summaries, key takeaways, uh, to make the content even more interesting.
[00:11:26] and then there are these other interesting markers. I mean, again, like, like brevity, getting straight to the point, that matters hugely. And then also being very, deterministic in your writing, meaning it should be very clear to understand what the subject, object, predicate, et cetera, are.
[00:11:41] Like, the sentence should be very clear to understand. No passive wording. Those type of markers and maybe make it very clear to the AI what the relationship between the, concepts is that you cover in your content, right? And This is challenging because, I think if, if you write in a too clean, too dry way, it's just not interesting for humans, right?
[00:12:01] And that still matters. but on the other hand, if you're on the other end of the spectrum, if it's all storytelling, it's like a lot of memey type of writing, very hard to understand, without context, then for AI, it's just not that interesting because it wants the nuggets. It wants the, the,the kind of the key takeaways.
[00:12:18] Amanda Natividad: This is super helpful. I think what's helpful about this too is that it, I think this gives people a little bit of structure to think through how they're creating their content. But at the same time, like, well, other thing I'll say is I feel like this is also getting at AI trying to replicate how people take in information.
[00:12:39] Like, most people don't like too much fluff, and we're, we are not smart, you know, we're not smart. we do better with simple active sentences.
[00:12:49] Kevin Indig: We do. We do. You know, it's, it's, uh, you're absolutely right. I mean, you have to look at what AI is trained on to understand AI's preferences. But then also there's a technical aspect to AI understanding of relationship between entities and your content. and again, like I, I, I think, you know, at the end of the day, there is an, an, an interesting implied meta question here, which is, am I writing for AI or for humans? Because this is not classic Google anymore, where you would write for Google in a sense so that more humans come to your content, right? When you care about AI writing, you care about showing up in AI and not much about humans, right? So I think there's an interesting argument to be made of like should you even care about it or
[00:13:31] should you, you know, write for both of them separately. and then on the other hand, I think there is a, style that good writers use that is not ideal for AI, but so much better for humans to read. So for example, Malcolm Gladwell, one of my favorite writers, he intentionally withholds information so he can reveal it later, right? There's a story arc to it.
[00:13:54] there's story twists, there is surprise, there is new reveals. All-- None of that is good for AI, right? If you wanna write for AI, it needs to be Wikipedia. facts clearly structured, no storytelling, just all the most important information in the, in the beginning.
[00:14:09] Amanda Natividad: Oh, that's, that's a really good example. You know I've been seeing a lot lately? I don't know if you're on Threads much, but I've been seeing a lot of AI generated AI slop in the form of these, like, multi-part threads. And I mean, I can't prove it, right? You can't, you can't really prove when something is written by ChatGPT or whatever.
[00:14:31] But I read some of this and I'm like, I really, really think this is written by ChatGPT." And, like, it goes viral, and there are lots of comments. And then I'll see even people I know who are pretty anti-AI who are like, "Oh my gosh, this story." And I'm like, "Oh, I'm pretty sure that's AI." I guess, like, this isn't even a question, I'm just reacting.
[00:14:53] Like, that kind of does crush my spirit a little bit when I see that.
[00:14:57] Kevin Indig: Yeah. No, look, there, there is a, there is an anchor here to research. it, it crushes my spirit too, and I think you can lose a lot of trust by, letting some of the AI writing shine. I think it's incredibly dangerous. And trust is the most important ingredient in this AI world. It is the most important ingredient. We see this in, in the research in two ways. First of all, there is a distinct pattern that when Google shows AI overview, so AI answers at the top of the search results, the number of people that click on Reddit and YouTube goes up by two-thirds, So almost double. And at first I was like, "Okay, sure, it makes sense because, uh, Reddit and YouTube are cited in, in, in the AI overview." And then I realized, wait, no, people don't click citations that much and, and, and it's not the case. Even though now YouTube is probably the most, cited domain across most models. But, uh, when we did the research, that was not the case. What was the case is that when we looked at the recordings from our research, we saw that people read the AI answer and they were like, "Yeah, that makes sense, but I wanna see if a human says the same thing." So then they go to Reddit and they see-- that they search or they go to Red, uh, like they, click on Reddit results or they search on Reddit to see if that AI answer is supported by what they perceive as humans. so there is a fact checking mechanism and that has to do with trust. And then second, we saw that when users, get a shortlist in AI mode, so for example, they wanna buy a laptop or they wanna buy insurance, they go to AI mode and they, they do the research, and then they get a short list here like the three best solutions. 75% of the time, they will go with the first one. So interestingly, the being first in a short list is still important even though we're not talking about classic Templu links anymore. But in the context of AI mode and other, LMs being the first is still important. The only time or the only f- factor that is more powerful than the rank order is trust. So when people get a shortlist and they see a brand that they trust, they will always pick that brand. They will not go with the first rank result, right? So trust is not just like a, you know, like a, like a conversion booster, if you will.
[00:17:16] I think it is the most important good in this AI world since content is easier and faster to create than ever before, and we're competing with more synthetic content. So if you are a person, a creator, or a, a company, and I, I I think you wanna be very strategic and systematic about building and, maintaining trust.
[00:17:37] Amanda Natividad: you mentioned AI ranking here, and we did some research at SparkToro, and it was in January, so maybe in LLM world that's really old. And we found that AI ranking was kind of random. Like, visibility is one thing, but ranking was just a lot more inconsistent, unpredictable. Do you think that's still the case or not so much?
[00:18:03] Kevin Indig: It depends on how I measure it, I would say,
[00:18:05] and it depends on the model. Some models Are
[00:18:07] just much more noisier than others. it seems that Google, for example, tries to get a certain consistency into their rankings. But, one thing that I found is that there-- the, the volatility or the variance, better said, of the answer generally from prompts is just so high. So, to, to make that more tangible, there were two research pieces that I co-authored, where in both cases we ran prompts several times consecutively, and one researcher was three times, other, uh, researcher was five times. in both times, we noticed that the, the part of the answer and the citations that remain across every of the three or five runs is incredibly small. We're talking about a few percentage points, right? So there is this concept of,model bias or model consistency, which basically means, how confident is the model in a certain part of the answer or a certain source, and that is the one that will maintain over time. And that's why I'm saying, like, it depends in my mind, about how and when you measure it, right?
[00:19:07] It changes so quickly. models are-- have different preferences. and so if you, if you track a prompt, several times, I think there is a higher chance to understand what remains consistent. But if you ju- if you just track a single run, then yeah, it's, the results can be completely different.
[00:19:24] Amanda Natividad: Hmm. Are there any tools right now that you kind of like for tracking some of this stuff?
[00:19:28] Kevin Indig: I'm biased. I work with AirOps, for example. I, uh, but I think there, there are many tools that are doing, a good and interesting job out there, right? I think you have to mention, Peak and Profound and, Gauge, Promptwatch, early. I can probably go on for a couple of days. but I will also say that I think there's still a lot of work on prompt tracking to be done. and right now there's a lot of, uh, almost like a universal approach to prompt tracking, and I would love to see more kind of new innovative approaches. Again, I think the kind of consecutive run thing, that's one important ingredient. Then there's also something important to be said about personas.
[00:20:09] And, and these days more tools ha- include personas, but it's incredibly important to track prompts, in the context of a certain audience profile because the personalization degree is so, so high with AI search, right? With classic Google, 99% of the results were the same for people living in the same country. with AI search, everybody has a unique experience essentially, and AI has a different degree of context about you. So you need to think about jobs to be done and about, personas, ideally syn- synthetic personas, to write the prompts that, that you're tracking. And even then, right, prompts are a, a proxy.
[00:20:47] There is no way to understand what somebody, what, what somebody prompts. You could just approximate it.
[00:20:52] Amanda Natividad: Oh, totally. And I, I've done like, I mean, we, this was part of the research we did at SparkToro, but I've also done like very, very many little versions of this where I've asked friends on united under a common interest, like I asked 10 people like, "How would you look for this thing?" And then I was like, "Send me your prompt."
[00:21:10] And they all had very different prompts. Like I asked about, "How would you find a basketball league for your kid?" And I I got prompts that ranged from, "My kid is not a beginner to sports, but I really want them to learn like good sportsmanship." Another person's prompt was just, "Best basketball leagues near me."
[00:21:32] Which, hey, they're both valid, right? But it's just people can have the same intention or the same intent and still have very different prompts.
[00:21:41] Kevin Indig: 100%. 100%. That, that's why I think intent, jobs to be done, those are kind of frameworks that make a lot more sense. but I mean, also the length of the prom- prompt or query decides how many people search for the same thing by default, right? And so I think as soon as you overstep 10, 15, 20 words, the chance that that, that you're the only person in the world using that specific prompt is pretty high
[00:22:03] Amanda Natividad: Yeah. Oh, that's interesting. Okay, so this is a good segue, I think, into the user behavior research you've been doing. I know you've already been, talking about some of that here, but I think this is also a lot of the research you've been doing, right, with Amanda Johnson. well, I guess I just wanna ask, like, when you're seeing users actually use these tools, I guess what has been surprising to you, and, like, were there assumptions that you had that you th- that, like, you asumed they would do but didn't do?
[00:22:32] Kevin Indig: Yeah. So, um, a lot. so Am- Amanda is my, my editor,she helps me a ton with, with like, not just editing, but also just thinking about pieces and, and like creating this content. and then Eric, Van Buskirk is my business partner in crime to help like with operations, et cetera, of these studies. and so, you know, like on fascinating, shockingly fascinating behavior that I saw very consistently in one of the research studies that we did is that people accept the AI answer pretty outright. so we're talking about AI answers specifically in AI mode, and like Gemini and and ChatGPT, et cetera, But whenever it's like a an AI chatbot, people really take the answer as the word.
[00:23:17] They, they don't do a lot of fact checking and like probing. on Google, when we talk about AI overviews, that is very different. people show a very different comparison behavior, much more probing, fact checking, et cetera. So first, what shocked me is how many people actually take the AI answer as the thing.
[00:23:32] But maybe it shouldn't shock me because, you know, I think the convenience factor is so high and, and the majority of people don't know how these things work, right? the second behavior that was very interesting, is about, how engaged people are. So people really spend a lot of time, usually about the time that mo- the average person brushes their teeth.
[00:23:52] That's how long people spend on an AI mode answer, right? So they really read this stuff. it's less skimming. It's really like reading top to bottom, and then following this relatively closely. and the third point that I found very interesting is just again, like how few people actually click.
[00:24:06] they maybe click when it comes to purchases, when they wanna buy a product. But as the action of buying moves more into AI as well, you know, like I think we're pretty close to being able to check out in AI mode directly. That's also gonna go away. So click outs, again, there's-- they still happen, of course, today.
[00:24:24] Not everybody uses AI mode, right? But if I look into the future, I really do, uh, uh, see like a a zero click future on the horizon. So these, these ecosystems become more closed, right? Google becomes a lot more closed, um, since less, um, at some point, probably no traffic out anymore. so we need to change our ways of measuring this type of stuff, about thinking of success, et cetera. but these are the three behaviors where I was like, "Okay, that's different."
[00:24:50] Amanda Natividad: That's super interesting. Okay, so what are, let's say, you know, a company or a brand is still relatively new in tracking their success in AEO, what are some of the things that you think someone should measure? Like, let's say it's first 90 days of marketing. I don't know. How would you recommend people measure progress here?
[00:25:12] Kevin Indig: First of all, I think it's im- important to, like, think about what makes most sense for your business. If you are more of an aggregator, again, like review business, aggregator type of business, meaning if you have, like, a large inventory of either a product or something else, citation...
[00:25:25] You're more likely to be cited. in every other case, I think the money really is in AI mentions. I think citations don't matter much, to be honest. the most attention goes to AI mentions. And so you wanna m- you you wanna monitor your, mention rate, or, or sometimes called share of voice, which basically measures how often you are mentioned in answers you're tracking against the total available mentions, right?
[00:25:51] Like, all the brands that are mentioned, how often are you mentioned? Second, and, and, and I'm not gonna repeat, like, all the, the prompt tracking things that we have already been through. I, I'll take, I'll set that as, like, a default. but second, you want to understand when you are in a shortlist and when you're not in a shortlist. So shortlist rank and how many prompts return a shortlist, absolutely critical. that is, to me, the high, the, highest point of conversion, most bottom of the funnel. That's when people are most likely to take action. and then third, you wanna get a sense for sentiment. We're still in a stage with prompt tracking where we just look at volume, but we don't really look a lot at how we're being portrayed.
[00:26:28] if we need to go much, much further beyond sentiment. We also need to understand what attributes we're being connected to. So are we being connected to being reliable, cheap, trustworthy, effective, et cetera? all that is-- It's not easy to parse these days. You probably need some sort of a custom solution.
[00:26:45] But that is what, what's necessary because, again, people engage with AI answers. They really trust them in many, most cases. so the way that you're being described is very, very important and needs to be tracked over time. So I'd say, like, these three things minimals, share of voice,your overall sentiment, and then the attributes that you're being connected to.
[00:27:03] Amanda Natividad: That's super interesting. I think it's ... Like, I'm kind of laughing at this because I still remember sitting in a conference room like, gosh, maybe 12 years ago with our PR team and with executives, and the PR team was talking about, "We've really increased our share of voice this quarter. Sentiment is really positive about our product."
[00:27:26] And the executives were like, "What does that even mean? That sounds like PR fluff." But here we are.
[00:27:32] Yeah. brand recall, for example. That is something you can perfectly transfer over to, uh, to AI and AI search. actually think, like, I have to take that like marketing died and was replaced by performance marketing, and over the last two decades, performance marketing has worked so well that modern marketers basically just look at numbers and dashboards, and nobody knows how to talk to people anymore. But that is really what it takes these days, right? If you wanna be successful in AI search, you need a constant stream of user research, user interviews to understand what the questions are that they raise, what they think about you, even if, if they even know your brand, right? Because AI mirrors or tries to mirror reality, right?
[00:28:16] Kevin Indig: If the common consensus is that your brand is, I don't know, uh, untrustworthy, right? If your last pass, and sorry to, to dump on you, last pass, but if you're last passing at this massive security breach, right, that will stick around forever in, in AI search. And so you need to, start at the base, which is your target audience, and you need to redevelop that atrophied muscle that is actually talking to customers.
[00:28:40] Amanda Natividad: I mean, 100%. That is the, the heart of audience research, customer research. You go where your audience is, not where you hope they would be. You build relationships there. You influence them where they are, no matter where that is. And then over time, you kind of build that goodwill. You build that positive sentiment so that when they're ready, then they will become a loyal customer.
[00:29:05] Kevin Indig: Exactly,
[00:29:06] exactly.it's trust, right? It's, uh, and, and this is hard to build and easy to lose. but yeah, we need to get off the dashboards. Like Indiana Jones saying, "Get out of the library," I, I kind of wanna ... This is my Harrison Ford
[00:29:18] take. There you
[00:29:18] Amanda Natividad: I love that. I love that. this takes us to a good wrap-up point. I guess, uh, what is, what's the call to action now? What's the best way for people to learn from you?
[00:29:26] Kevin Indig: thanks so much. So my call to action is to subscribe to my newsletter. It goes out to 20,007 people now. it's called Growth Memo, so growth-memo.com, which is Google Growth Memo, you'll find it. that is my call to action. Again, like I, I try to be a, you know, a humble disciple and learn this stuff as it constantly changes.
[00:29:47] So I share as much as I find and as I can. that's probably the best way to learn with me
[00:29:52] Amanda Natividad: Awesome. Kevin, you're the best. I have learned so much from you. You are so thorough in the research that you do. it still kind of blows my mind. I'm like, "I don't know how he does it," but thank you for the work that you do, because you make all this stuff a little bit more accessible to the rest of us
[00:30:08] Kevin Indig: It Means a lot, Amanda. Thank you so much
[00:30:11] Amanda: Thank you for listening or watching Zero Click Marketing. We'll see you next week.