How to Write AI Prompts That Improve Everything in Your Marketing

There are infinite use cases for AI.

Marketers, maybe more than anyone, are very excited. In marketing, it’s mostly used for three things: to research, to write and to recommend.

• Research: generate ideas and brainstorming (this is the most popular use case for AI)
• Write: generate outlines and drafts of content, promotional assets and strategic docs
• Recommend: generate suggestions for things you’ve already created

Of those three uses, that last one might be the least common and most powerful.

AI is amazingly helpful for looking at your marketing from your prospect’s perspective and suggesting improvements. With the right prompts, it can do quick little audits and spot gaps in seconds that even an expert might miss.

Every marketer has blind spots. Yet despite our skill gaps, we still are accountable for a huge range of assets. We own the performance of every initiative across every channel. AI can make marketers stronger in those areas that aren’t our specialties. The right audit prompt can make the content strategist better at paid marketing. Or make the PPC expert better at email marketing. Or make the SEO better at social media.

That’s because AI is a “skills leveler.” It improves results a little for experts, but improves results a lot for lower skilled workers. And let’s be honest, we’re all lower skilled workers sometimes.

Source: Ethan Mollick

So many articles share lazy little prompts for making things.
But today we’re not sharing prompts, we’re sharing a way of making prompts.
And we’re not making new things, we’re making the things you already have even better.

Here’s a general framework for writing prompts – or series of prompts – for auditing anything in your marketing. This includes both public-facing assets (articles, webpages, ads) and internal documents (strategies, style guides, marketing plans).

We’ll do it by combining three elements into a single prompt:

• Our persona or ideal client profile (ICP)
• The marketing asset we want to audit
• A little summary of best practices for that asset.

AI will respond with recommendations, which we’ll then filter through to find the best ideas.

Example: Let’s have AI audit a LinkedIn Ad

I’ve never really done any paid marketing. Ads are a blind spot for me. So let’s see if AI can help fill this skills gap. I made this fake LinkedIn ad promoting a webinar about B2B lead generation. It looks pretty good to me, but Iet’s ask AI if I missed anything…

We’ll use ChatGPT but feel free to use the LLM of your choice. We’ve tested Claude and CoPilot and found that they are similar in quality for this use case.

1. Prime the AI with your persona

First, we’ll teach the AI about our audience. This is a key first step. If we don’t teach the AI who you’re targeting, responses will be generic and off-target. It’s a common mistake. Marketers write a short, low-effort prompt and expect magic to happen. But magic rarely happens for lazy marketers.

There are three ways to prime the AI on your target audience, ranging from very simple to kinda fancy:

1. Start with a persona prompt (quick)
2. Upload a PDF of your documented persona or ideal client profiles (usually better)
3. Create a CustomGPT (good for sharing it with others on your team)

In case you don’t have a documented persona nearby, feel free to use this persona prompt:

Build me a persona of a [job title] at [industry/company size/geography] with [roles/skills/responsibility]. This person is looking for help with [challenge/problem/task] and is considering [product/service].

List their hopes/dreams, fears/concerns, emotional triggers and decision criteria for hiring/contacting a [industry/category/service/product].

Look closely at the response. It’s not perfect, is it? Next, tell the AI to fix anything that’s incorrect and add anything it missed. Don’t trust the AI. Trust yourself. You have a ton of experience with your specific audience. AI does not.

Once you’ve got it dialed in, I suggest copying it into a PDF file and keeping it handy. Share it with your team. Keep it in your shared prompt library where your team can find it fast.

Or you can save this AI conversation and use it anytime you want to audit a marketing asset. You could also create a CustomGPT, which isn’t as fancy as it sounds. Just click “Explore GPTs” and follow the instructions.

For my little audit of my LinkedIn ad, I’ll use my detailed ideal client profile, which I keep as a PDF on my desktop for easy access. But if I’d used a persona prompt, I’d just stay in that conversation in the following steps.

Got your persona ready? Great. Let’s move on.

2. Give it the asset you want to audit

AI can audit your public facing assets, such as an article or ad, or your internal strategic documents, such as a marketing plan or a sales deck. Some people are squeamish about uploading things to AI, so check with your boss or your client if the asset has any proprietary information.

The format of the asset determines how you give it to the AI.

• If it’s text (article, webpage copy, email newsletter), you’ll just copy and paste it into the prompt.
• If it’s a visual (paid social ad, video thumbnail, print piece), you’ll upload the image.
• If it’s a document (powerpoint presentation, marketing plan), you’ll upload that file.

If you’re auditing a page on your website, you have options. There are three ways to give it to the AI webpage: paste in the text, upload a screenshot or give it the link. They each have pros and cons.

Example: Screenshot of LinkedIn ad

I took a screenshot of my mocked up LinkedIn ad. It’s an image on my desktop, ready to drop into the prompt. Because I need to upload this image, I’ll need the paid version of ChatGPT or the free version of Claude.

Got your asset ready? Great. Let’s move on.

3. Create a list of best practices

The final next step is to make a list of best practices for the asset that we can use in our prompt.

In my example, I need a LinkedIn ad that performs well. Before I spend money on an ad budget, I need to tune up the creative. If the ad is bad, I’m wasting time and money. So I need to create a short list of best practices for high-performing LinkedIn ads.

How do we research best practices? Find blog posts. Watch videos. Download whitepapers. Reach out to a subject matter expert. Ask the AI.

Create a list of best practices for high-performing [type of asset]. Put them in a brief bullet list that I can use in a subsequent prompt.

If you have expertise in this area, add your own best practices based on your experience. Don’t hold back. The longer the list, the more recommendations AI will provide. Many of my most effective audit prompts have 10-15 best practices.

Example: Best practices for LinkedIn ads

With less than 30 minutes of research, I collected some basic best practices for LinkedIn ads by reading articles, watching a video and talking to the AI. I’ll keep it simple here and just list three.

1. The message is tailored for the target audience and speaks directly to their pain points, using language that resonates and highlights benefits.

2. The headlines are clear and direct, indicating why they should care. 

3. The visuals are impactful, uncluttered and have minimal text.

Got your best practices ready? Great. We’re ready to build the audit prompt.

4. Create the audit prompt

We have all the elements, the persona, the asset and the best practices. We’re ready to put them together into an audit prompt. Start by telling the AI about its role and skills. Then add the list of best practices. Finally, tell it that you’re giving it the persona and asset and what you want to get back.

Here’s a template you can use.

You are a [role] with [skills].

The best [type of asset] share several common traits:

1. [best practice]

2. [best practice]

3. [best practice]

4. [best practice]

5. [best practice]

I’m giving you a persona and a [describe marketing asset].

Rate the extent to which it does or does not align with the best practices above. Add any other important best practices I missed.

Recommend changes that would make this marketing asset more effective.

I like to ask it to show how it both does and does not align with the best practices. That way it will consider positives and negatives. I also ask it to consider best practices I may have missed. I want a detailed but broad review.

Example: Audit prompt for LinkedIn ads

This is how I used the template to craft my LinkedIn ad audit prompt.

You are an expert paid social media marketing strategist, skilled at evaluating the likely performance of LinkedIn ads. The best LinkedIn ads share several common traits:

1. The message is tailored for the target audience and speaks directly to their pain points, using language that resonates and highlights benefits.
2. The headlines are clear and direct, indicating why they should care.
3. The visuals are impactful, uncluttered and have minimal text. I’m giving you a LinkedIn ad and target persona.

I’m giving you a persona and a LinkedIn ad. Rate the extent to which the ad does or does not align with the best practices above. Add any other important best practices I missed. Recommend changes that would make this ad more effective.

Got your prompt ready? Great! Fire it off and see what the AI recommended…

5. Filter through the recommendations

Patiently wait 3-5 seconds while the AI does the analysis. It’s reviewing your marketing from the perspective of your audience, using the standards for quality you provided in your best practices.

Probably, you’re looking at a mixed bag of recommendations. It’s likely you’re seeing the attributes of the persona reflected in the recommendations.

• Some of the suggestions are obvious. You’ve already considered them.
• Some of the suggestions are irrelevant. They don’t make sense for your brand
• …a few of the suggestions are really interesting.

As the expert, human marketer, it’s your job to oversee the quality of AI outputs and filter out the garbage. It’s a tool, not a magic spell. Just like the output of any tool, much of what it gives you isn’t useful. That’s fine.

Ignore many of the suggestions. That’s fine.

The hope is that one or two gives you good ideas that you hadn’t considered. Or overlooked. Or just failed to prioritize. Just glimpse your marketing through your audience’s eyes, and one or two insights may appear.

I asked an expert at UX testing, Brian Massey, for his thoughts on using AI to audit marketing. He shared this insight and a warning:

Brian Massey, Conversion Sciences “Generative AI has probably been trained on Daniel Kahneman’s “Thinking Fast, Slow”, but language models don’t have a “Fast” mode. They don’t get depleted and resort to making decisions driven by emotional experiences, recent experiences, or confirmation bias. This is an amazing capability for evaluating business communication. AI doesn’t have an agenda and does not need to take shortcuts. But as you use these tools, notice how quickly you introduce your own biases into the output. Notice how you disregard some recommendations out of hand and focus on others. Notice when you get overwhelmed by the volume of data these language models can produce. That’s a sign of depletion, which is usually followed by resorting to your biases.”

With that caution about our own biases, let’s see what the AI thought of my LinkedIn ad…

It made six recommendations, spanning aspects like messaging alignment, credibility, scarcity, visuals and the CTA. Some I liked a lot; others I found unhelpful and dismissed. Maybe I’m biased.

I thought the suggestion for a better headline was very good: change “B2B lead generation tips” to “Proven strategies to boost B2B lead conversions.” Being more specific is usually good. Specificity correlates with conversion.

I didn’t like the suggestion for the visual. My original design had a picture of my design team. Maybe I’m biased because I love those people. But the AI suggested an image of a dashboard? That sounds boring. Really, I should ask those designers. They’re the experts.

Filter the good from the bad, factoring in the level of effort and the likelihood of impact of the change. Now you can prioritize. Or keep the conversation with the AI going with another prompt or two.

✨ Bonus! Try these useful follow-up prompts

Try these prompts for deeper analysis, better performance, relative comparisons and new variants…

Make additional suggestions based on best practices I did not provide.

Align each of your recommendations with a marketing outcome or marketing metric. What results could I expect if I make these changes?

Make two variants based on the provided asset. Create one that closely aligns with the best practices. Make the second more unexpected and memorable, deviating from the best practices.

Here’s a second marketing asset for comparison. Create a color coded heatmap matrix scoring each against the best practices and alignment with the psychology of the persona.

I used that last prompt to compare my original to a revision that incorporated some of the AI’s initial suggestions. No surprise, it was pleased that I took it’s advice.

Follow up prompts are usually very effective. AI tools are most effective when you continue the conversation. Keep talking to the AI. When you run out of ideas, tell it to ask you questions. Or ask it what prompt you should use next.

Isn’t there a faster way?

Yes, you could do this faster. You could simplify the process. You could use this simple prompt.

How could this marketing piece be better?

Try it. It may work well. But test it with and without a persona and you’ll see the difference. When you include details about your audience and your success criteria, results are better.

The detailed approach we’ve shown here is more time consuming, but our goal is performance, not efficiency. You may have noticed, in marketing the difference between good and great is huge. You’ve probably seen curves like this in many marketing reports…

Source: Zipfian Distributions in Marketing

One strong performer will drive better results than a dozen low performers.

Our goal is 10x better results, not 10% better efficiency. We need performance, not speed. Spend as much time as it takes to make something excellent. That’s why we’re using AI for analysis, rather than having it crank out a big pile of medium-quality blog posts.

Don’t just use AI to make things.
You already have a lot of stuff.
Use AI to make things better.

Final Challenge: Write prompts that do this…

The term “prompt engineering” sounds snooty to me. But the best prompts are carefully constructed. See if you can construct audit prompts for any of the following use cases.

1. Write a prompt that asks the AI to audit your homepage …then compare your prompt to the one I wrote at the bottom of this article.
2. Write a prompt that audits your LinkedIn profile by comparing it to the job description of your dream job. Ask it to do a SWOT analysis of you for that job …then revise your LinkedIn proposal.
3. Write a prompt that analyzes your ads and their corresponding landing pages. Ask it to find any mismatches that might hurt bounce rates.
4. Write a prompt that asks the AI to take the role of your persona and then ask you questions after reviewing your sales material …then update your sales material

If you find a method that works well, add it to your shared prompt library and show your team how you did it. One final step: add “AI marketing analyst” to your LinkedIn profile.