A lot of people are using AI like a chatbot, just having it agree with them and giving vague instructions. But there’s a better way. Here are the exact phrases and frameworks I use to get the best results from AI agents.
Start with the goal, not the how
The first thing I do is talk about what I want the AI to do, not how I want it done. If I want to show it how something should be done, I’ll reference specific examples. I’ll say things like “according to the BlitzMetrics blog guidelines” or “according to the many examples you’ve seen us publish on how we optimize Facebook ad campaigns.”
If you’ve got lots of examples of how you’ve achieved a certain result, whether it’s published to a vault like Obsidian, stored in a database, sitting in your CRM, or blogged about, then the AI can reference how it’s being done. That way you can focus on the result.
Be very specific. Instead of saying “improve my SEO,” say something like “I want to increase the number of calls I’m getting for residential roofing jobs over the next 90 days via local service ads and Facebook ads.” When the goal is specific, just like it would be if you’re working with one of our agencies, the agent can backtrack through how to achieve it.
Set the context by talking, not typing
This is where everyone seems to miss. The number one mistake is typing to AI instead of talking to it. When you talk to the AI like you’re having a conversation, empathetically, you’re able to give it a couple minutes of context on all kinds of little details. What it can use, what it can pull from, the different components of relationships, proof, credibility, specifics, your offer.
If you don’t tell it where to look, the AI is just going to guess and give you a generic answer. It’s a word pattern matcher, so it will just match the most obvious likely pattern. Instead, say things like “here’s what’s going on, this is the situation, this is who we’re working with” and reference specific videos, emails, or Zoom call transcripts so it understands the background.
Force it to disagree with you
Once you’ve defined the goal and provided the context, make sure you don’t have a blind spot. I love to say “be the devil’s advocate. What are things that I’m not thinking about that are important? What would Dennis poke holes in? What assumptions am I making? Don’t just agree with me.”
Unless you tell it explicitly not to just agree with you, it’s going to say “oh, that’s fantastic, you should definitely do that.” You don’t want that. I find with all the major AIs, you still have to tell it every time: don’t just agree with me, be critical, be cynical. I want to hear things that are in my blind spot.
Document everything with meta articles
As your agent is doing work, it has to document what it’s doing. I see a lot of people go wild and loose, with all these agents doing different things and overlapping each other because none of them are keeping track.
What we like to do is write to a Google Doc. I’ll say “as you are doing this work, document what you’re doing step by step so another agent can come in and inspect the work.” Then we audit it according to our specific guidelines, whether that’s blog post posting guidelines, Facebook video ad optimization, or local service ads.
We take it a step further by turning these into what we call meta articles. A meta article is a step-by-step document of what the agent did for that particular project. It doesn’t have to be published externally. It could live on your laptop, Google Drive, Dropbox, wherever. But this way, if you ever need to revert, if a new model comes out and you don’t like the output anymore, you can always go back.
QA your work and set up weekly cycles
Once a task is done, ask the AI “did you do it completely?” and have it QA its own work according to your QA guidelines. You’ll find it spawns another round of work because most agents will do a few minutes of work and stop. Ask it again, “can you make it better?” and it will find more to do.
You can batch this process too. For example, if someone has a YouTube channel with 200 videos, I’ll say repurpose those into articles. It’ll do the first 10, and then I’ll say batch it, and every time you finish a batch, update the meta article.
The most valuable thing is setting up a weekly scheduled job. You can do this inside Claude, Gemini, or ChatGPT, where every week it wakes up on Friday and audits according to what we call an MAA Cycle, which stands for Metrics Analysis Action. I’ll prompt it by saying “now I want you to do MAA according to the guidelines on BlitzMetrics.com so we can continue to optimize and increase revenue.”
Now you’ve got the system looking over all the different components, troubleshooting and telling you what’s going on. You’re in the cockpit rather than being a worker. Whether you are hiring a digital marketer, trying to be an AI expert, or hiring an AI person, this is something you need to understand.
If you want to learn more, come check out my YouTube channel.
Dennis Yu
Dennis Yu is the CEO of Local Service Spotlight, a platform that amplifies the reputations of contractors and local service businesses using the Content Factory process. He is a former search engine engineer who has spent a billion dollars on Google and Facebook ads for Nike, Quiznos, Ashley Furniture, Red Bull, State Farm, and other brands.
Dennis has achieved 25% of his goal of creating a million digital marketing jobs by partnering with universities, professional organizations, and agencies. Through Local Service Spotlight, he teaches the Dollar a Day strategy and Content Factory training to help local service businesses enhance their existing local reputation and make the phone ring.
Dennis coaches young adult agency owners serving plumbers, AC technicians, landscapers, roofers, electricians, and believes there should be a standard in measuring local marketing efforts, much like doctors and plumbers must be certified.