WEBVTT 00:00:00.000 --> 00:00:04.880 Say I run a business. How many customers did ChatGPT send me last month? 00:00:04.880 --> 00:00:07.760 If you ask your website analytics? About zero. 00:00:07.760 --> 00:00:09.360 And the real number? 00:00:09.360 --> 00:00:12.080 Honestly? Nobody can see it from where you're sitting. 00:00:12.080 --> 00:00:13.600 That's a terrible answer. 00:00:13.600 --> 00:00:20.560 It is. But the people an assistant sends you mostly don't click anything. They call you. Or they type your name into Google. 00:00:20.560 --> 00:00:23.920 So they show up, and my website never hears about it. 00:00:23.920 --> 00:00:26.160 Right. It just looks like somebody found you. 00:00:26.160 --> 00:00:28.400 So my dashboard's been lying to me. 00:00:28.400 --> 00:00:30.640 Not lying. Counting the wrong door. 00:00:42.514 --> 00:00:52.914 This is After the Demo. I'm John Arndt. I run about fifteen small software products out of Sacramento, most of them built and run with AI agents, plus an AI consulting practice. 00:00:52.914 --> 00:00:53.954 And I'm Kate. 00:00:53.954 --> 00:01:02.114 Each episode, one question business owners are actually asking about AI, answered from the middle of the work, and something you can do about it this week. 00:01:02.114 --> 00:01:10.354 This time it's the one you're already asking. Are these AI assistants sending people to my business or not? And how would I even know? 00:01:10.354 --> 00:01:14.194 By the end you'll know how to check it yourself, and what to fix first. 00:01:14.194 --> 00:01:16.834 And if he gets abstract, I stop him. 00:01:16.834 --> 00:01:21.794 She will. Start with who's asking, because it's more people than you think. 00:01:25.104 --> 00:01:36.624 So, who's asking. There's a consumer survey from BrightLocal, from February of this year. About a thousand adults in the US. They looked at where people go when they want a recommendation for a local business. 00:01:36.624 --> 00:01:38.704 And the answer is Google. 00:01:38.704 --> 00:01:43.584 Google's still first. Seventy-one percent. Facebook's second, at sixty. 00:01:43.584 --> 00:01:45.824 Okay. So nothing's changed. 00:01:45.824 --> 00:01:50.384 Third place is ChatGPT and the other AI tools. Forty-five percent. 00:01:50.384 --> 00:01:54.544 Forty-five percent of people asked a chatbot to find them a plumber? 00:01:54.544 --> 00:01:59.824 Used AI for local recommendations, yeah. And the year before, that number was six. 00:01:59.824 --> 00:02:00.784 Six. 00:02:00.784 --> 00:02:04.304 Six percent to forty-five percent. In one year. 00:02:04.304 --> 00:02:07.584 Okay, that's the kind of jump where I want to see the survey. 00:02:07.584 --> 00:02:14.144 Fair. It's in the show notes. About a thousand people. One survey, but that's a big move for any survey. 00:02:14.144 --> 00:02:19.824 And it's still third. Behind Google and Facebook. So people didn't stop googling. 00:02:19.824 --> 00:02:25.184 Most of them didn't. They added a step. Some of them ask the assistant first, then go look you up. 00:02:25.184 --> 00:02:26.704 Look you up how? 00:02:26.704 --> 00:02:31.104 Hold that. It's a big part of why your analytics say zero, and we're getting there. 00:02:31.104 --> 00:02:37.344 So the owner at the top of the show, asking how many customers ChatGPT sent them. That's not paranoia. 00:02:37.344 --> 00:02:53.304 No. That's paying attention. But the number I actually care about is in the follow-up. BrightLocal went back in March and surveyed just AI users, four hundred fifty-five of them. Ninety-seven percent said they sometimes double-check what the AI recommended against real reviews. 00:02:53.304 --> 00:02:56.704 Ninety-seven. So basically everybody. 00:02:56.704 --> 00:03:05.704 Basically everybody. Sometimes, to be fair. But think about what that means. The assistant hands them a shortlist. Then they go read your reviews before they pick. 00:03:05.704 --> 00:03:09.344 Hang on. Sometimes is doing a lot of work in that sentence. 00:03:09.344 --> 00:03:13.824 It is. Sometimes isn't always. The rest of the time, they don't check. 00:03:13.824 --> 00:03:17.384 So for those people, the list is the whole decision. 00:03:17.384 --> 00:03:20.024 Right. Which is why you want to be on it either way. 00:03:20.024 --> 00:03:21.984 So the robot gets you in the room. 00:03:21.984 --> 00:03:26.487 And your reviews close it. The AI builds the list. It doesn't make the sale. 00:03:26.487 --> 00:03:31.024 Which is kind of a relief. The stuff a business already did still counts. 00:03:31.024 --> 00:03:37.144 It still counts. But you have to make the list first. If you're not on it, that person never gets to your reviews. 00:03:37.144 --> 00:03:43.904 And if the AI and the reviews don't agree? The assistant says you're great, and the reviews say you never call anybody back. 00:03:43.904 --> 00:03:48.944 Then the assistant just sent a customer over to read why they shouldn't hire you. The reviews win that one. 00:03:48.944 --> 00:03:58.224 Okay, let me push on this. Say I run a dental practice, and say my patients are seventy. They are not asking a robot for a dentist. 00:03:58.224 --> 00:03:59.984 You're probably right about them. 00:03:59.984 --> 00:04:01.024 Probably? 00:04:01.024 --> 00:04:12.704 Mostly right. That same March report broke it out by age. People thirty to forty-four, sixty-four percent have asked an AI for a business recommendation. Sixty and over, twenty-four percent. 00:04:12.704 --> 00:04:14.224 So I win. 00:04:14.224 --> 00:04:17.824 About your patient, yes. Now, who picked the dentist? 00:04:17.824 --> 00:04:18.784 The patient. 00:04:18.784 --> 00:04:27.424 Sometimes. Take senior living. Say the mom's seventy-eight. Often she's not the one researching places. Her daughter is, and say the daughter's forty-two. 00:04:27.424 --> 00:04:30.624 Which puts her in the age group where it's sixty-four percent. 00:04:30.624 --> 00:04:37.824 Right in it. Same thing with a business that sells to other businesses. Often the owner signs, but the office manager is the one who went looking. 00:04:37.824 --> 00:04:42.384 Come on. The office manager is asking ChatGPT for a vendor? 00:04:42.384 --> 00:04:51.264 If she's thirty-five, she's in the same age group as the daughter, so odds are she's asked an AI to recommend a business. For work specifically, the survey doesn't say. 00:04:51.264 --> 00:04:55.424 So the customer and the person doing the research aren't always the same person. 00:04:55.424 --> 00:04:59.864 A lot of the time they're not. And the researcher is the one who'd be asking the AI. 00:04:59.864 --> 00:05:04.837 Okay. Split decision. My seventy-year-olds mostly aren't asking. Their kids are. 00:05:04.837 --> 00:05:06.104 I'll take a split. 00:05:06.104 --> 00:05:08.984 So what does somebody actually do with that this week? 00:05:08.984 --> 00:05:20.344 Grab a piece of paper. Write down who actually researches your business before anybody calls you. The buyer, their kid, their office manager. Be honest about it, because it's often not who you picture. 00:05:20.344 --> 00:05:22.704 Why paper? Can't I just think about it? 00:05:22.704 --> 00:05:30.144 In your head it's always the buyer. On paper you notice it's the daughter, or the office manager. Put a rough age next to each one, too. 00:05:30.144 --> 00:05:32.144 Because of the sixty-and-over thing. 00:05:32.144 --> 00:05:36.064 Yeah. If the person on your list is forty, don't plan around the seventy-year-old. 00:05:36.064 --> 00:05:36.944 And then? 00:05:36.944 --> 00:05:43.197 Then assume that person asked an AI first. Before they saw your website. Before they read a single review. 00:05:43.197 --> 00:05:45.184 That's it? A list? 00:05:45.075 --> 00:05:51.715 That's it. Once you know who's asking, you know what question they typed. And that's the question you'll want to check later. 00:05:51.715 --> 00:05:56.835 Fine. So if they're asking, why does my website say nobody came? 00:06:01.723 --> 00:06:06.843 Because your website can only count the people who come through your website. And a lot of them don't. 00:06:06.843 --> 00:06:12.763 You said that at the top. They call, or they type my name in. But somebody has to be tracking this. 00:06:12.763 --> 00:06:29.323 People are trying. At the end of September, Cody Schneider posted that most companies have no idea AI search is already sending them traffic. And he built a dashboard out of Google Analytics and Search Console that rolls ChatGPT, Perplexity and Gemini traffic into one number. 00:06:29.323 --> 00:06:30.523 That sounds useful. 00:06:30.523 --> 00:06:37.083 It is. I think he's right that most companies don't know, and putting it in one number is a good idea. I'd build that. 00:06:37.083 --> 00:06:38.923 I hear a but. 00:06:38.923 --> 00:06:44.683 But it can only count the people who clicked. So it's a floor. It isn't the scoreboard. 00:06:44.683 --> 00:06:47.723 How low a floor? How many people skip the click? 00:06:47.723 --> 00:06:59.843 The best number I have is about Google, not ChatGPT. It's a Pew Research study from July twenty twenty-five. They looked at real Google searches from nine hundred adults in the US, back in March of that year. 00:06:59.843 --> 00:07:01.123 Okay. And? 00:07:01.123 --> 00:07:09.083 When there was no AI summary at the top, people clicked a regular result on fifteen percent of visits. When the summary showed up, eight percent. 00:07:09.083 --> 00:07:10.483 That's roughly half. 00:07:10.483 --> 00:07:16.683 Roughly. And the links inside the summary itself? People clicked one of those on one percent of visits. 00:07:16.683 --> 00:07:18.043 One percent. 00:07:18.043 --> 00:07:18.683 One. 00:07:18.683 --> 00:07:23.363 The answer's sitting right there, and hardly anybody clicks through. 00:07:23.363 --> 00:07:31.163 On Google, with the summary showing, a lot fewer people clicked. I don't have that study for ChatGPT, so I won't pretend it's the same number. 00:07:31.163 --> 00:07:34.403 Fine. The ones who don't click. Where do they go? 00:07:34.403 --> 00:07:44.043 I keep a research page on this, and the link's in the show notes. For a local business, the people who get your name from an AI answer mostly call you, search your name, or find you in maps. 00:07:44.043 --> 00:07:45.283 The wrong door. 00:07:45.283 --> 00:07:53.574 The wrong door. None of that comes in labeled ChatGPT. The call is just a call. The name search looks like somebody who already knew you. 00:07:53.574 --> 00:07:56.763 And the people from before, the ones who double-check the reviews. 00:07:56.763 --> 00:08:03.083 Probably typing your name in to find the reviews. So even the careful ones look like they found you on their own. 00:08:03.083 --> 00:08:06.803 Okay, but Google has to know something. It's their summary. 00:08:06.803 --> 00:08:15.003 There's something there now. Google's Search Console has a report for impressions from AI Overviews and AI Mode. Search Engine Journal wrote it up in August. 00:08:15.003 --> 00:08:17.643 Impressions meaning what, in English? 00:08:17.643 --> 00:08:24.523 How often your pages showed up in those AI answers. You can split it by page, by country, by device and by date. 00:08:24.523 --> 00:08:25.563 That's the number, then. 00:08:25.563 --> 00:08:32.843 It's a number. There are no clicks in it. There are no searches, so you can't see what anybody typed. Just how often you showed up. 00:08:32.843 --> 00:08:36.683 It tells me I was at the party. Not who I talked to. 00:08:36.683 --> 00:08:40.043 Pretty much. Good to have. Still not customers. 00:08:40.043 --> 00:08:47.803 Okay, I'm going to be the annoying one. Say I run a business and my accountant wants a number. They called is not a number. 00:08:47.803 --> 00:08:49.163 You're right. It isn't. 00:08:49.163 --> 00:08:49.963 Thank you. 00:08:49.963 --> 00:08:57.403 And keep the dashboards. Cody's version, the Google report, whatever you've got. They're real floors, and floors are worth watching. 00:08:57.403 --> 00:08:58.283 But. 00:08:58.283 --> 00:09:03.883 But the number your accountant wants is at the phone and the intake form. You get it by asking one question. 00:09:03.883 --> 00:09:05.403 How did you hear about us. 00:09:05.403 --> 00:09:17.563 Which you might already ask. It's one of the checks on that research page, too. The change is the list of answers. Add one that says, an AI assistant, like ChatGPT. On the phone script, and on the intake form. 00:09:17.563 --> 00:09:20.683 Why spell out ChatGPT? Just put AI. 00:09:20.683 --> 00:09:27.763 Because the person on the phone might not think of it as AI. They might just think of it as ChatGPT. Give them the word they'd use. 00:09:27.763 --> 00:09:29.363 And then what, I count it? 00:09:29.363 --> 00:09:33.803 Count it for thirty days. Every time somebody picks that answer, it goes on the tally. 00:09:33.803 --> 00:09:35.723 So the number is a tally sheet. 00:09:35.723 --> 00:09:36.576 Yes. 00:09:36.576 --> 00:09:40.803 That's a very cheap answer to a very expensive-sounding problem. 00:09:40.803 --> 00:09:49.003 It's also the first number in this whole conversation that's actually yours. The surveys are about everybody. The tally is about your business. 00:09:49.003 --> 00:09:51.963 Does whoever answers the phone need to know why they're asking? 00:09:51.963 --> 00:09:58.443 Tell them. Otherwise somebody says ChatGPT told me about you, it sounds like small talk, and it never gets written down. 00:09:58.443 --> 00:10:01.043 And thirty days from now, I've got a real count. 00:10:01.043 --> 00:10:09.723 A real count. Maybe a low one, since people don't always remember how they found you. But it's yours, and it beats a zero from a dashboard watching the wrong door. 00:10:09.723 --> 00:10:13.963 Okay, but I'd rather know before thirty days. Can I just ask it myself? 00:10:18.697 --> 00:10:24.137 Yes. Today. It takes about ten minutes, and you don't need anything you don't already have. 00:10:24.137 --> 00:10:28.537 Okay. Walk me through it like I've never opened one of these things. 00:10:28.537 --> 00:10:35.817 Start with the list from earlier. Who researches your business, and what would they type? Say it's the daughter, looking for a plumber for her mom. 00:10:35.817 --> 00:10:39.177 She types something like, who's a good plumber near me. 00:10:39.177 --> 00:10:44.537 That exact question. Type it into ChatGPT, then into a couple of other assistants. 00:10:44.537 --> 00:10:47.817 Do I put my business name in it? Because I'd be tempted. 00:10:47.817 --> 00:10:54.857 Don't. That's the one rule. If you name yourself, of course it talks about you. You want to know if it brings you up on its own. 00:10:54.857 --> 00:10:57.177 And is that what you do, just fancier? 00:10:57.177 --> 00:11:05.257 Pretty much. I ask the same fixed questions on a schedule, never with the business's name in them, and I save the full answers so I can compare them later. 00:11:05.257 --> 00:11:08.537 Why the same questions every time? I'd want to try a bunch. 00:11:08.537 --> 00:11:12.857 Because if you change the question, you can't tell whether the answer moved or you did. 00:11:12.857 --> 00:11:16.137 Fair. And what you see is what my customer sees? 00:11:16.137 --> 00:11:27.817 No, and I want to be plain about this. I ask the engines directly, the same ones behind the apps. The apps can answer differently. And no tool can show you exactly what one person saw on their phone. 00:11:27.817 --> 00:11:30.617 Same engine, then. Not the same screen. 00:11:30.617 --> 00:11:42.857 Right. So what does it actually find? I keep a public leaderboard for plumbers in Sacramento. Last run was September seventeenth. Two assistants, ChatGPT and Gemini. Five questions a customer would ask. 00:11:42.857 --> 00:11:44.457 So ten answers. 00:11:44.457 --> 00:11:49.897 Ten answers. Eight different plumbers got named across them. And one of them showed up in five. 00:11:49.897 --> 00:11:54.417 One plumber in half the answers. And the other seven get the leftovers. 00:11:54.417 --> 00:11:55.600 Less often, yeah. 00:11:55.600 --> 00:11:59.977 Hold on, though. Five questions, ten answers? That's your data? 00:11:59.977 --> 00:12:01.337 That's the data. 00:12:01.337 --> 00:12:09.364 Because that's a pretty small pile to hang a plumber's reputation on. Say I'm plumber number eight. I'm not losing sleep over ten answers. 00:12:09.364 --> 00:12:13.777 You're right. It's small. I wouldn't call it a verdict. It's a thermometer. 00:12:13.777 --> 00:12:19.577 Meaning I take it again next month. One reading is today. A few readings tell me which way it's moving. 00:12:19.577 --> 00:12:20.230 That's it. 00:12:20.230 --> 00:12:22.217 So I win this one. 00:12:22.217 --> 00:12:27.017 You win this one. It's a small sample, you say so out loud, and you measure again. 00:12:27.017 --> 00:12:28.817 Okay. Got a bigger pile? 00:12:28.817 --> 00:12:41.177 I do. On September twenty-sixth I ran a study on credit unions. Four assistants, ChatGPT, Gemini, Claude and Perplexity. Eight metros around the US. Four hundred eighty answers. 00:12:41.177 --> 00:12:42.937 Okay, that's a real pile. 00:12:42.937 --> 00:12:49.097 In each metro, the biggest local credit union has about half the members. Fifty-one percent, on average. 00:12:49.097 --> 00:12:51.417 Then it should come up about half the time. 00:12:51.417 --> 00:12:56.537 You'd think. It got about a third of the local credit union mentions. Thirty-four percent. 00:12:56.537 --> 00:13:00.777 Being the biggest doesn't even get you your share. That seems like the headline. 00:13:00.777 --> 00:13:02.377 That's the small problem. 00:13:02.377 --> 00:13:03.977 That's the small one? 00:13:03.977 --> 00:13:10.937 One hundred fourteen of those four hundred eighty answers used a name the credit union doesn't use anymore. Mostly after a merger. 00:13:10.937 --> 00:13:13.337 Wait, that's almost one in four. 00:13:13.337 --> 00:13:25.497 Nearly one in four. Then eligibility, meaning who's allowed to join. There were fifty-four statements about that, each one checked against the credit union's own membership page. Eighteen didn't match. 00:13:25.497 --> 00:13:28.537 One in three. On who can even join. 00:13:28.537 --> 00:13:31.817 One in three didn't match the credit union's own page. 00:13:31.817 --> 00:13:40.777 Take the daughter again. Say she's finding a credit union for her mom, and the answer hands her an old name. Maybe she's one of the people who double-checks. 00:13:40.777 --> 00:13:48.777 Maybe. But now she's double-checking a name the credit union stopped using. Invisible is the small problem. Wrong is the big one. 00:13:48.777 --> 00:13:55.097 Okay. I ask it myself this week. Then what, I do this by hand every month for the rest of my life? 00:13:55.097 --> 00:14:08.497 You can. That's why I built Mentioned On: it asks the same questions on a schedule, without your name in them, and keeps the answers, so a change is something you can check. There's a free scan, no signup, and it can't promise you a spot in the answers. 00:14:08.497 --> 00:14:16.697 Good. I'd trust it less if it did. Okay, the move this week. Three assistants, the question my customer would ask, no name in it. 00:14:16.697 --> 00:14:19.977 Save the answers with the date. And circle anything that's wrong. 00:14:19.977 --> 00:14:21.017 Wrong like what? 00:14:21.017 --> 00:14:26.777 An old name, or the wrong area you serve. Anything you'd correct if a customer said it to you on the phone. 00:14:26.777 --> 00:14:28.617 And if it doesn't mention me at all? 00:14:28.617 --> 00:14:32.697 Write that down too, with the date. That's your first reading on the thermometer. 00:14:32.697 --> 00:14:36.937 So say I find something wrong. Do I need some new AI trick to fix it? 00:14:41.679 --> 00:14:54.879 Depends who you ask. At the start of October, Chris Long posted, in all caps, that people should make an AI instructions page now, because it can influence what Google's AI Overviews say when somebody searches your brand name. 00:14:54.879 --> 00:14:56.479 In all caps. 00:14:56.479 --> 00:14:57.599 All caps. 00:14:57.599 --> 00:14:58.879 Does it work? 00:14:58.879 --> 00:15:04.239 I don't know. I haven't tested it, and I'm not telling anybody to go build one. It's the hack of the week. 00:15:04.239 --> 00:15:24.959 There's always a hack of the week. And I saw the other side of that one. Stacy Muur posted around the same time that the algorithms behind ChatGPT, Claude and AI Overviews change faster than you can adapt. A trick works, people abuse it, it gets changed. So she'd just focus on SEO, which helps the AI answers anyway. 00:15:24.959 --> 00:15:26.799 That's a fair read of what she said. 00:15:26.799 --> 00:15:36.559 And I'm with Stacy. Say I run a business, and I've got maybe two hours a week for marketing. I'm not spending them on a trick that might get patched before I finish building it. 00:15:36.559 --> 00:15:38.559 Let me argue the other side for a minute. 00:15:38.559 --> 00:15:40.079 Of course you will. 00:15:40.079 --> 00:15:54.719 One SEO, Gagan Ghotra, says he's seeing Google test AI Overviews with no source links at all, on some best-of searches, especially local ones. Some searches, not all. That's one person's observation, and I can't confirm it. 00:15:54.719 --> 00:16:00.799 Okay. And if he's right, then on those searches, even that one percent from earlier has nothing to click. 00:16:00.799 --> 00:16:08.319 Right. On those searches there's nothing to click and nothing to count. If that spreads, being named in the answer is the whole game. 00:16:08.319 --> 00:16:16.959 Fine. That's a reason to keep checking what it says. I'll give you checking. What I want to know is what makes it say my name, and say it right. 00:16:16.959 --> 00:16:29.559 Fair. So I looked at my own data. I keep a page on where AI answers get their links, last updated October third. Thirty-four thousand citations, from about nine hundred fifty websites, scanned since June. 00:16:29.559 --> 00:16:35.679 That's a lot of links. So where do they point? Some secret AI directory I've never heard of? 00:16:35.679 --> 00:16:40.919 About eighty-six percent pointed at a business's own website. Or a product's own website. 00:16:40.919 --> 00:16:44.879 Their own website. The thing they already have. 00:16:44.879 --> 00:16:49.399 And directories and review sites, all of them combined, came to about ten percent. 00:16:49.399 --> 00:16:52.559 Hang on. Before I start gloating, what's the catch? 00:16:52.559 --> 00:17:00.439 The catch is those are businesses that got scanned through my own service, so it leans toward local services. Your industry might not look like that. 00:17:00.439 --> 00:17:03.759 Noted. Is there anybody else's data, or just yours? 00:17:03.759 --> 00:17:20.879 BrightLocal did a study in August. About thirteen hundred business locations, across Google's AI Overviews, AI Mode and ChatGPT. The single biggest cited source was the Google Business Profile. Twenty-eight and a half percent of all the citations, so a bit over a quarter. 00:17:20.879 --> 00:17:24.479 The Google profile. The one with your hours and your photos on it. 00:17:24.479 --> 00:17:36.079 That one. Different study, different sample, but both point at stuff you already control. And BrightLocal's own line, roughly, was that there isn't one source you can optimize for and call it done. 00:17:36.079 --> 00:17:42.959 Okay. Let me say it back to you. The new AI strategy is fix your website and fix your Google profile. 00:17:42.959 --> 00:17:43.519 Yeah. 00:17:43.519 --> 00:17:45.119 That's the old strategy. 00:17:45.119 --> 00:17:49.839 Yeah. You win that one. The work's the same. What changed is that you can check it. 00:17:49.839 --> 00:17:51.359 And Stacy wins too. 00:17:51.359 --> 00:17:53.919 And Stacy wins too. Credit where it's due. 00:17:53.919 --> 00:18:01.279 The stuff I circled earlier. The wrong answers. Say I'm the credit union with the old name. Do I start on my own site? 00:18:01.279 --> 00:18:10.079 Own site first, because that's where most of those links pointed. Make sure it says the current name, plainly. Then make your Google profile say the same thing. 00:18:10.079 --> 00:18:20.879 And they have to match because people double-check. You said that at the start. If my website says one name and my Google profile says another, the person checking just got two answers. 00:18:20.879 --> 00:18:22.559 And one of them is wrong. 00:18:22.559 --> 00:18:30.239 Okay. The list for this week. My homepage says, in plain words, what I do, where I do it, and who it's for. 00:18:30.239 --> 00:18:34.079 Plain words. If your neighbor couldn't repeat it back, rewrite it. 00:18:34.079 --> 00:18:38.079 So not, innovative solutions for a changing world. 00:18:38.079 --> 00:18:45.519 Not that. Then fix the wrong thing you circled, on your own site. Then make your Google profile match it. 00:18:45.519 --> 00:18:51.759 And next month I take the temperature again. Same questions, same assistants, and I see if the thermometer moved. 00:18:51.759 --> 00:18:55.599 That's the whole loop. None of it's a trick, so there's nothing to get patched. 00:18:55.599 --> 00:18:57.359 And the AI instructions page? 00:18:57.359 --> 00:19:02.559 I haven't tested it, so I'm not telling you to build one. If I test it and it works, I'll say so. 00:19:02.559 --> 00:19:06.799 Fine. Give me the one thing. Tomorrow morning, coffee in hand. 00:19:07.017 --> 00:19:13.897 Open ChatGPT and two other assistants. Ask each one the question your customer would ask, without your name in it. 00:19:13.897 --> 00:19:16.217 Save the answers, with today's date. 00:19:16.217 --> 00:19:21.337 Then set a calendar reminder to ask the same questions next month. That's the one thing. 00:19:21.337 --> 00:19:24.457 That's it? That's barely a cup of coffee. 00:19:24.457 --> 00:19:32.697 That's it. And if you've got two more minutes, add an AI assistant, like ChatGPT, to your how did you hear about us list. Start the tally. 00:19:32.697 --> 00:19:38.457 And send this to whoever answers your phones. They're the ones who'll hear ChatGPT told me about you first. 00:19:38.457 --> 00:19:43.017 Links are in the show notes. The studies, the posts we talked about, and my research pages. 00:19:43.017 --> 00:19:47.417 Back to the top, then. How many customers did ChatGPT send me last month? 00:19:47.417 --> 00:19:52.777 Last month, nobody can tell you. A month from now, you'll have a tally and two sets of answers. 00:19:52.777 --> 00:19:55.017 Which beats a zero from a dashboard. 00:19:55.017 --> 00:19:59.777 A zero from the wrong door. This has been After the Demo. I'm John Arndt. 00:19:59.777 --> 00:20:02.217 And I'm Kate. Go watch the right door.