Joshua Jacoby warns marketers to feed AI tools real analytics data, not just prompts, to avoid hallucinated decisions. Don't rely on AI alone for copywriting, publish no more than one blog and one landing page weekly to dodge thin-content penalties, and never use AI-retouched photos on Google My Business, which Google may flag as deceptive.
Hello, my name is Joshua Jacoby and I am the owner of My Favorite Web Designs. We're a digital marketing and website design agency located in East Mesa. We've been in business for over 17 years and I've personally overseen or hand-developed thousands of websites and done marketing for hundreds of brands. And I wanted to give a little perspective today about what AI is doing well and where it's missing the mark in terms of marketing for your business. So a lot of people are trying to use Claude Desktop or different projects through OpenAI to manage the marketing on their business. And one of the first things you need to look for is how much data you've given the project before you ask it to make decisions for you. The less actual hard number data that you give the project, the less realistic their decisions that they're making for your brand are going to be because they're not going to be based in actual facts. So give it more data, give it to Google Search Console, give it to Google Analytics, give it the query data. If you have conversion data, give it all the actual specific conversion data so that it can make decisions based off of actual information and not dubbing information, which it does. The next thing I'll say is in terms of marketing, you need to not use the AI to do all the writing. AI's writing can be really, really good and really, really impressive. OpenAI is better than Claude writing, in my opinion. But all that writing is still recognizable as AI. So you have to have extensive projects to guide the writing of the AI so it's not recognizable as AI and that it actually has your own brand voice and that it has the facts based off your business. SEO is a good example. So if you're talking to AI about the SEO for your website, you might ask it its opinion on something like a backlink that you're going to build in a directory. Yeah. And AI is going to go search for recent articles on the top that can give you a decision based just off of a few recent articles. Because while it might have access to all the information in humanity, it can't keep it all in its context window at the same time. So it's going to select certain bits of information in order to answer your question. And when you're asking it on something ambiguous, then all of a sudden the answer is going to be different constantly. And that happens all the time in marketing. And you have to be very cautious about asking it to make hard decisions because there's often more than one answer to the question. And even amongst the community answers, you're going to get way different answers from user-generated content than you will from the official documentation. So you have to be able to control what sources it's actually drawing its decisions from when it's doing the marketing for your business. One of the next things that we see people do is going too fast, especially with organic SEO. It's tempting to lay down a huge amount of content in a small time, but if you lay down a lot of content on an existing website and you don't give it time to rank, then it can start automatically ranking those things thin pages and you can get into a situation where your newly published content never ranks because Google thinks your entire site is thin or being manipulated by AI because you're publishing too quick. A general safe cadence for publishing, in our opinion, would be about one blog and one landing page a week. For some larger brands, you can easily get away with more. For smaller brands, you could do less, and even that cadence might be a lot for a brand new website. But generally that's a safe-ish type cadence that we see out there, and it's easy to get over, go over the top with the content and start producing it too fast. And that can be problematic as well. Some of the other things we see in marketing is criticism of existing marketing decisions and dramatic framing of marketing problems. So let's take, for example, an AdWords campaign with a company that I was running recently on exact match. The company was getting dozens and dozens of phone calls and experiencing really good return on investment from the Google Ads that it was getting in, but the attribution was broken due to one of the backend tagging features. So when it was coming in, it wasn't attributing to a specific campaign. And we tried explaining that to the client that they needed to correct their call tracking module. But the AI, when it analyzed that same situation, said that these ad campaigns drove zero conversions because it wasn't actually able to see the phone calls attributed to them. And even though we tried to explain to him that you need to explain to that AI project that the attribution was incorrect and there were 39 phone calls based off $1,000 in spend or whatever it might have been, if it doesn't actually have that hard data, it's just going to hallucinate and it's going to frame it dramatically. So it's going to say something like, oh, here's the real picture. This $3,000 in spend netted you zero phone calls and you need to instantly stop that. So Whenever you're working with an AI, you gotta kind of treat it like a power tool. It can do tons of good things, but it can do tons of bad things and you gotta steer it to do exactly what you need it to do. I could go on kind of forever about what you can and can't use AI with, but I'll leave you with one last fun tip that I think a lot of people are missing right now, which is don't use AI to retouch images of people. And if you do use AI to retouch images of people, definitely don't put those images onto your GMB or your landing pages, but especially your GMB. Google will find that be deceptive or fraudulent and not like that. So authentic, especially for GMB, will always be edited right now. And we pretty much recommend going directly from cell phone to GMB because that's kind of a pattern that they can see, like the direct device data going right onto the uploaded GMB with the same IP and all that location right there. So authenticity matters on the backend and on the frontend. And just be careful about where you're using AI because it is getting wildly smart. But it is still going to be wildly hallucinating for forever until we all understand exactly how to work with it and train it and feed it what it needs. Thank you.
