Almost Timely News: ๐️ How To Improve Advertising with AI (2026-06-28)
Almost Timely News: ๐️ How To Improve Advertising with AI (2026-06-28)Claude asked me how to do this
Almost Timely News: ๐️ How To Improve Advertising with AI (2026-06-28) :: View in Browser
The Big Plug๐ My new course, GEO 201 on competitive GEO measurement, is now for sale. Content Authenticity Statement100% of this week’s newsletter was made by me, the human. In the video version, you’ll see Claude outputs. Learn why this kind of disclosure is a good idea and might be required for anyone doing business in any capacity with the EU in the near future. Watch This Newsletter On YouTube ๐บ
Click here for the video ๐บ version of this newsletter on YouTube » Click here for an MP3 audio ๐ง only version » What’s On My Mind: How To Improve Advertising with AIThis week, I was working on a new piece for Trust Insights, on AI use cases for marketing, when Claude asked me a question as it was putting together a list of things it wanted answers for. As a bit of background, I don’t have AI write for me. I have AI make up challenging questions for me to answer, and when I have large periods of downtime - like driving long distances - I wire up my voice recorder and spend time answering those questions. This way, I keep my own executive functions sharp, keep myself thinking and exploring, rather than deskilling and handing things over to the machine. Claude’s question was an interesting one: “How would you advise a marketer to use AI to make advertising more automated, especially Google Ads?“ I don’t run a ton of ads, though after thinking this process through, I may give it a try as I’m fairly comfortable that my answers are on sure footing. So let’s walk through the approach, which is underpinned by the 5P Framework from Trust Insights™. Part 1: PurposeWhy do we run ads? Early in my career, I thought ads were just about selling stuff, and they are, but we can do this in many different ways. We can buy ads to build brand awareness, to reach people early in consideration and make them aware that we have solutions to their problems, to catch people when they’re assembling their short list, to intercept people just before purchase, or to help keep customers loyal by reminding them we exist. Advertising, when done well, reaches out and captures attention of different audiences to help us reach people we might not otherwise have the means to reach, or to deepen the relationship we have with people who know us. These days, it’s harder than ever to keep attention, so ads can help with that. Which means we have to answer the question: why are we running ads? What problem are we trying to solve? I can say for sure what the wrong reason is: “because our competitors are doing it!”. That’s the wrong answer most of the time. What’s the right answer? Well, start with your analytics data. Katie Robbert did a demo of this recently for our team quite brilliantly by using Claude Code to sew together a bunch of disparate data sets and visualize our marketing operations funnel. You should do the same thing - open up the agentic coding tool of your choice like OpenAI Codex, Google Antigravity, Claude Code, Claude Cowork, etc. Grab all your data that you have available - Google Analytics, your newsletter, your CRM or Shopify cart data, whatever you’ve got. Make sure you’ve got Jesse Vincent’s superpowers installed in your environment (it’s free), and kick off by having it outline your marketing operations funnel with the brainstorming skill. Once you’ve done that, have it identify the stage in your marketing operations funnel where your transitions are weakest. That’s the area where ads might be of some use, to help nudge people out of the stuck stage. That’s our purpose, to unclog that part of our marketing operations funnel. As a sidebar, I keep calling it that because while the customer journey itself may be nonlinear, our ability to measure and manage it still very much is, especially as our organizations get larger and more complex. Everyone can’t and shouldn’t do everything after a certain point, so having a funnel that mirrors people’s responsibilities and skills makes good sense. I put my data through and about half an hour later, Claude Code identified that this newsletter was doing almost all the lifting, and my website definitely needs to be contributing more. We’ll put a pin in that. In terms of where I should be using ads, it identified a clear initiative based on my data and the limited budget I specified (USD 25 a day) - to get more, better newsletter subscribers. We now have a clear purpose for our Google Ads: get more, better subscribers. Part 2: PeopleThe next step in our process is to identify our people - both the people doing our marketing, and most especially the people we want to serve ads to. Who are these people? After all, Claude Code identified that to make the most of a limited ads budget, we should get more, better subscribers. What does this mean? To know this, we want to look at two groups of data - the people we currently have access to, and the people we’d like access to. Google Analytics, if your website has enough traffic, can give you some basic demographic data such as inferred age, gender, and most valuable, affinities and interests. This latter part is what tells us about the people in greater depth, the subjects they’re most interested in outside of our website. All of Google’s data comes from their advertising network, Doubleclick, and it’s entirely made by machines based on the places people visit. It’s not carved in stone, so treat it with the same level of surety that you’d treat any AI inference data - directional, but not rock solid. Remember that your website, if it’s big enough, will likely have different demographics by section. For me, I have a lot of marketing and business professionals who are looking for help with AI and analytics, but I also have an audience of event planners and conference organizers. These are not the same people, and they have different backgrounds. So I’d want to use Google Analytics to explore if there are significant differences in these populations based on the pages they visit on my site, like my speaking page versus this newsletter. If you want to automate ads well, you absolutely need an ideal customer profile with enough behavioral richness that you can have an LLM impersonate that audience. You can’t skip this step - and even better, if you have past ad data from the platform of your choice about who did and didn’t respond to your ads, now is the time to integrate this. After we know who we have, it’s time to dig into who we’d want in our universe. The question you have to ask yourself is, where does your ideal customer spend their time that you can get data about? For example, most B2B businesses will have their ideal customers in places like LinkedIn, but also in places like golf courses. Is there data out there that can inform your audience building qualitatively and quantitatively? One of my favorite sources for this kind of qualitative data is Reddit, especially as they’ve been doing a good job recently (perhaps too good) of cracking down on automation and AI bots farming clout. If you have a Reddit developer’s API key, you can extract data from the service in compliance with their Terms of Service, and it’s a very generous allocation. The key question to answer with our AI tools is how much of a difference is there between who we have and who we want. In my case, for my newsletter, I ideally want people who can book me to speak at their events or to hire me and Trust Insights as a consulting firm. Barring that, I’d want people who are willing to buy things like courses and books. I’d look at all the data I have plus all the data I can gather, and blend it all together. Once you have that data in hand, assemble it into an ideal customer profile or profiles, depending on how many different audiences you have. Katie built mine originally - a service Trust Insights offers - and I refresh them a couple times a year to keep them current. Part 3: ProcessWe know what we’re doing and why. We know who we’re after. Now let’s talk about how we’d go about making ads. I am not a skilled ads copywriter. It’s not my jam, and what seems fun or clever to me often falls flat with other people. I’m much less witty than I sound in my head. Any time I’m facing a domain where I know I’m not strong, my first step is to do some deep research. In this case, if we’re going with Google Ads, I’d want to commission some deep research reports on what the best practices for Google Ads are today. The platform has changed so much, and because I don’t run ads frequently, what’s in my head is badly out of date. My process for this is to use the Trust Insights CASINO deep research framework to build a thorough research prompt for Google Ads, then hand it to 3-5 different AI deep research platforms. For low risk things, I’ll use Gemini, Perplexity, and Alibaba Qwen. For high risk things - like running ads and potentially spending/wasting a lot of money - I’ll add in Claude and ChatGPT. The reason for this is that ChatGPT in particular restricts you to 10 deep research reports a month, so I try to reserve those for things that really matter. Once the platforms return their results, I do a research merge with Claude Code to unify the reports, fact check them, and then consolidate them into a primary research report. From that, I take a page from the Qwen lab - back in 2025, they showed that reinforcement learning with checklists was one of the most powerful ways to create guardrails for AI - and have Claude Code engineer a checklist from the consolidated research report. These knowledge blocks need to be kept fresh; if Google Ads is a big part of your job/business, I’d freshen them monthly. If Google Ads isn’t a big part of your job, I’d freshen them quarterly. The first thing we should do is have ourselves a little virtual focus group with our ideal customer profiles along with our data about what we’re trying to advertise. In my case, let’s say we’re going with attracting newsletter subscribers. I’d want to create three different personas from my ideal customer profile to debate each other - a detractor (someone who thinks I’m full of ๐ฉ), a neutral (someone who knows nothing about me or this newsletter), and a supporter (you!). I’d prepare a brief for them and a neutral moderator to debate about why someone would or would not subscribe to this newsletter and have the moderator lead a focus group. From that focus group, I’d get talking points to help me better understand what’s appealing (or not) about this newsletter, then distill that down into a brief. That brief is the foundation for my ads program, because it’ll contain helpful, useful insights about what people might or might not value. With that brief in hand, plus my best practices guide and checklist, plus my ICPs, I’m ready to build. Part 4: PlatformSince Claude’s specific question was about making ads more autonomous, we could approach this in a few different ways like the Google Ads API, but I tend to be very cautious with ads. Because of the high potential of wasting a bunch of money, I’d rather introduce some friction into the process and have my tool generate files I can load into Google Ads Editor. That would give me the ability to spot check the output and make sure it’s sane. If you wanted to YOLO it, you could build against the Google Ads API directly. I would personally not do that; I’ve had my share of systems that have gone rogue at one point or another and run up bills. Google’s systems have warnings about spend levels but few brakes; you can wind up with quite a large bill. Having our first system with lots of manual overrides and friction points reduces the likelihood of this happening. Okay, so what’s the next steps? An LLM will be integral to this process, because we want to make use of our ICPs, so the general workflow would go something like this:
Once the LLM finishes its work, our software can export the finished ads in TSV format to be imported into Google Ads Editor. This gives us the ability to spot check the ads as we copy and paste them, a final chance for human review. Once the ads are pasted in and we push the configuration to Google Ads, we’re ready to approve them and turn them on. For ads, the other thing we might want to do is generate multimedia for different ad formats, like images, animations, or even videos. As with the ads system itself, there are many different image generators and APIs; DeepInfra supports batch image creation, as does Google Gemini 3.5 Flash. You could programmatically generate those assets at scale; my caution here is to do some test runs first to ensure that the images look the way you want before you batch create. Getting this wrong could mean spending quite a bit of money on API calls for image generation and then having to re-do it. Part 5: PerformancePerformance is where the rubber meets the road. Once you’ve deployed your ads, then it’s time to monitor how they’re doing, what results you’re getting. In tools like Google Ads Editor, you can export your entire ads database along with all the performance measures, which is exceptionally helpful for data analysis. Most ads analysis tools simply use the data within the ads system, and one of the things we forget most is that ad systems are tuned to make money - for the ad network. Their recommendations have an implicit and sometimes explicit bias; us getting good performance out of our ads is secondary to making profits for themselves. That means that we take their recommendations with a grain of salt (since they almost always suggest spending more money) and we do our own analysis on the ads. How do we do this? One of the biggest gaps in ad systems analytics is a lack of feature engineering. Feature engineering, a data science term, is all about extracting data from the data we have so we can see if smaller pieces of data have some hidden insights for us. For example, take this date: 2026-06-28. We see this as a date, as a single piece of data. But we can decompose it into:
You might say this is unnecessary, but if we’re selling a product to consumers, things like day of week or day of month matter quite a bit - consumers get paid on regular cycles, which means sometimes people are more capable of buying than others. Even in the case of a campaign like mine, where I’m looking to acquire non-paying subscribers, people have more stress and strain at the end of the month. If I’m marketing to marketers, month end is less of a good time because lots of folks are scrambling to do month end reporting - and it’s even worse at end of quarter. Now compound that with the ways we use language - formally and informally, with different words, different topics, different everything. We could create a large number of features just from the ads data we have, then give all those features and data points to an AI system to evaluate. Which ones matter? Which ones don’t? Today’s AI systems have an encyclopedic knowledge of statistics and machine learning, and can write us the Python code we need to understand nearly any dataset. Remember that generative AI won’t be doing the analysis here, it’ll be classical AI via Python code (or the data science language of your choice, but I recommend Python to start). Generative AI will write the code to do the analysis and present us with the results - and then our generative AI tool can interpret the results and make recommendations for what steps to take next. Part 6: Wrapping UpThis first pass through the process shows what a more automated system could look like. The hard part is building all the infrastructure in advance - the knowledge blocks about Google Ads best practices, the ideal customer profiles and personas, the focus group systems, the Google Ads API or data interchanges, the ad generators and evaluators, the feature engineering and regression analysis of the data. However, once you’ve done that and you’ve battle tested each of the components, and you grow confident that the system pieces all work as intended, then you start automating them together, piece by piece. You might automate the focus group and ad copy portion first, or loop together the ad copy and regression analysis generation, with the ultimate goal of having a fully self contained system once you’re certain each subsystem works. Could you, starting from scratch, build the entire system to be fully automated in one shot? Yes, but the level of risk is much higher - and if you’re building a system that will spend your money, you’d better have amazing testing, deep subject matter expertise in Google Ads, and a high tolerance for risk. It can be done, but it’s higher risk. I’m sure Claude will enjoy this very long answer, but I hope you also found it useful. The concepts I walked through can apply to nearly any ads system, so take what’s useful and adapt it for your situation. And of course, shameless plug. If you’d like help doing this for your company, Trust Insights is happy to help. How Was This Issue?Rate this week’s newsletter issue with a single click/tap. Your feedback over time helps me figure out what content to create for you. Here’s The UnsubscribeIt took me a while to find a convenient way to link it up, but here’s how to get to the unsubscribe.
If you don’t see anything, here’s the text link to copy and paste: https://almosttimely.substack.com/action/disable_email Share With a Friend or ColleaguePlease share this newsletter with two other people. Send this URL to your friends/colleagues: https://www.christopherspenn.com/newsletter For enrolled subscribers on Substack, there are referral rewards if you refer 100, 200, or 300 other readers. Visit the Leaderboard here. ICYMI: In Case You Missed ItHere’s content from the last week in case things fell through the cracks:
On The TubesHere’s what debuted on my YouTube channel this week:
My Merch ShopI’ve been adding so much stuff that I’ve decided to bundle it all in what I call a Merch Shop, because otherwise there’s literally too much to keep track of and I run out of space in my own newsletter. So welcome to the Merch Shop! Skills for Claude and Agentic AI: Books: Courses: Subscriptions: Recent TalksThese are just a few of the classes I have available over at the Trust Insights website that you can take.
Advertisement: New GEO 201 CourseIn GEO 101, the first course I built on the basics of GEO, I taught you about presence, appearance, and relevance, the three phases of GEO, and what you need to do in each phase to align with how AI search operates. The top piece of feedback we got at Trust Insights about it was, “okay, great, but how do I tell my boss that we’re ‘winning’ at GEO?“ After I quelled my murderous rage at your boss on your behalf, Katie and I sat down and worked out a straightforward, aligned methodology for doing this. GEO 201 is based on the three phases, what you can control and what you can genuinely see - and critically, what you can’t. Because there is absolutely no way to say your brand “ranks higher” in AI search, period, end of story. But you can say and show with confidence what you’ve done and how you show up for presence, appearance, and relevance with tools you’re probably already paying for, and based on how AI search systems really work. ๐ GEO 201 is available now for USD 149. Get Back To Work!Folks who post jobs in the free Analytics for Marketers Slack community may have those jobs shared here, too. If you’re looking for work, check out these recent open positions, and check out the Slack group for the comprehensive list.
Disclosure: I source these links from LinkedIn every week on the following criteria: New in the past seven days, Easy Apply on, remote roles, USA geography. How to Stay in TouchLet’s make sure we’re connected in the places it suits you best. Here’s where you can find different content:
Listen to my theme song as a new single: Social Good: Ukraine ๐บ๐ฆ Humanitarian FundThe war to free Ukraine continues. If you’d like to support humanitarian efforts in Ukraine, the Ukrainian government has set up a special portal, United24, to help make contributing easy. The effort to free Ukraine from Russia’s illegal invasion needs your ongoing support. ๐ Donate today to the Ukraine Humanitarian Relief Fund » Events I’ll Be AtHere are the public events where I’m speaking and attending. Say hi if you’re at an event also:
There are also private events that aren’t open to the public. If you’re an event organizer, let me help your event shine. Visit my speaking page for more details. Can’t be at an event? Stop by my private Slack group instead, Analytics for Marketers. Required DisclosuresEvents with links have purchased sponsorships in this newsletter and as a result, I receive direct financial compensation for promoting them. Advertisements in this newsletter have paid to be promoted, and as a result, I receive direct financial compensation for promoting them. My company, Trust Insights, maintains business partnerships with companies including, but not limited to, Amazon, Talkwalker, MarketingProfs, Agorapulse, The Marketing AI Institute, Spin Sucks, and others. While links shared from partners are not explicit endorsements, nor do they directly financially benefit Trust Insights, a commercial relationship exists for which Trust Insights may receive indirect financial benefit, and thus I may receive indirect financial benefit from them as well. Thank YouThanks for subscribing and reading this far. I appreciate it. As always, thank you for your support, your attention, and your kindness. Please share this newsletter with two other people. See you next week, Christopher S. Penn Invite your friends and earn rewards
If you enjoy Almost Timely Newsletter, share it with your friends and earn rewards when they subscribe.
|
Comments