The human brain is awesome. So useful for so many things!

But there are some things that our brains just don’t do well.

The reasons are often because of cognitive biases. As the brain processes information and makes decisions, it’s always trying to be efficient. It uses shortcuts, which causes biases.

One powerful and problematic bias was described by the late, great Daniel Kahneman. Kahneman was a psychologist and economist who named this bias WYSIATI, which stands for “What you see is all there is.”

Because of WYSIATI, we form judgments based on the information available. We don’t usually tell ourselves, “Well, there are still many things I don’t know.” Instead, we use the information we have in front of us. In other words…

Humans are very bad at gap analysis

It’s easy for humans to evaluate a thing based on what’s there. But it’s very difficult for humans to notice what’s missing. Even a professional with decades of experience may not catch every omission. On the other hand, AI is very good at pointing out what’s missing.

AI is amazingly good at gap analysis

This insight is a revelation for marketers. While marketers everywhere are using AI to quickly spit out draft content, a few marketers are using it to review and audit all kinds of deliverables, spot holes, and then fill them in, either by hand or with a bit of AI help.

The rest of this article is a guide for using AI to do gap analysis at various levels of detail. These are five AI methods for five types of insights for impacts at various stages of the buyer journey, from the top to the bottom of the funnel.

We’ll start with a big-picture, content strategy AI method that is helpful for building an audience through top-of-funnel content.

In-person tickets are 95% sold out; grab yours now while you can! Save $300 with our Night Owl tickets, ending on May 17.

AI can actually spot holes in the internet itself. Some topics are missing from the big blogs and media sites that rank high. They are skipped over or rarely covered but could be very interesting to your audience. They are blue ocean and differentiated, waiting to be discovered.

It’s basically impossible for a human to read everything on all the blogs in a vertical and summarize what’s not there. But this task comes naturally to an LLM.

Here are a few quick prompts to have AI quickly help show us the topics that are rarely covered in your industry.

Use these prompts to discover serious, think-piece topics:

Use this prompt to discover lighthearted, op-ed topics:

Use this prompt to discover myth-busting topics:

Use these prompts to find “missing statistics” for original research ideas:

Several of the details in the prompts can be very powerful: “counter-narrative opinions,” “common assertions,” or “common misconceptions.” Any of these could inspire topics that are just waiting to be published across channels in various formats.

We showed examples of the prompts and responses in a post about AI-driven content strategy, where we used a spaceport company as an example. What satellite launch topics are missing from the big blogs? AI nailed it with some fascinating answers on the first try:

This is one of those AI-only use cases where it isn’t just improving efficiency. It’s doing something completely new, otherwise impossible for a human.

Here is another AI-powered gap analysis for top-of-funnel outcomes.

A lot of search-optimized articles rank, but not very high, for all kinds of phrases. Often, these phrases don’t even appear in the article.

Adding these already-ranking but not-present phrases to the content on a URL will help improve the relevance of that phrase and likely improve the rankings. You may also improve the rankings for the primary phrases by improving the general semantic relevance of the URL.

You know. It’s an ontological thing. And what is an SEO if not a reverse information retrieval specialist? I digress.

How can we find all the phrases for which a URL ranks but does not rank high? Paid SEO tools do this, including the Moz Pro Site Overview report. But let’s use the free tool that has ranking data from the primary source itself. We’ll export the data from Google Search Console (GSC).

Give the CSV from GSC to a GPT for SEO tips ASAP. That was quite the alphabet soup, but you probably know exactly what I meant. Nerd.

Here’s the process: I’ll use my own Search Console account and do a keyword gap analysis on an article about social media videos.

1. From the Search Results report, make sure all four metrics are turned on: Total clicks, Total impressions, Average CTR, and Average position.

2. Create a Query filter to exclude branded (company name) queries.

3. Create a Page filter to show just the URL that you’re optimizing.

4. Export to a CSV file. It will actually give you a ZIP file containing several files. You just want Queries.csv. Open it up to confirm there’s nothing weird. Rename it gsc_queries.csv.

Using the data from our current rankings lets us skip keyword research. We already rank.

Caution: If you’re an SEO agency, talk to your clients (and check your contracts) before uploading their data. But if you’re on the brand side working with an SEO, I suggest you grant them access to upload this kind of anonymized data to AI. It’s ranking data, which is basically public anyway.

You may need a paid AI tool, such as ChatGPT Plus, to upload and analyze data. With the first prompt, you’ll upload the query data and have it clean things up a bit. The dataset may contain many very closely related phrases. You don’t need them all.

And every time you ask the AI to edit a spreadsheet, you need to download and check it carefully.

If the file has hundreds of rows and the AI is struggling, manually remove the rows with phrases that have zero clicks. How does it look? Better? Nothing weird?

With our next prompt, we’ll give it the text from the page.

The AI will do several things next, including “text preprocessing” and “keyphrase extraction.” Once it’s done, we’ll prompt it to make the SEO on-page recommendations.

And so we don’t hurt the readability through AI-powered keyword stuffing, we’ll tell it to stay focused on the human visitor. After all, our real goal isn’t to rank; it’s to connect with our audience of humans.

Here, the AI doesn’t just find the gaps; it suggests ways to fill them.

Here’s an example of the output. As mentioned, the blog post in this example is about making social media videos. The recommendations are highlighted. Decide for yourself if you like the flow, and make sure a human eye always reviews the content before changing it.

If you’ve been publishing for a while, you can use the behavior of your past readers to inform the future of your content strategy. The data is in GA4 and the analysis can be done by an LLM. If you use ChatGPT, you’ll need a paid account to upload files.

We’ll keep it simple and just export our blog post titles and some basic metrics. For more detailed analysis, you can add “session source / medium” and have AI analyze performance across channels, which can inform your content strategy. But we’ll keep it simple for now.

1. Go to the Pages and screens report.

2. Set the primary dimension (the dropdown above the first column) to “Page title and screen class.”

3. To see just the blog articles, click “Create filter +” and set the filter so “Page path and screen class” contains “blog” (or whatever folder your content is within).

It should look like this:

Now, you’re looking at a list of the blog article title tags (from which AI can infer topics) and the basic performance for each.

1. Click the share icon in the top right and export the file as a CSV.

2. Open the file and delete the nine useless comment rows at the top. While you’re in there, take a few minutes to remove any rows with irrelevant data, such as translated titles (non-English) or other strange, low-traffic rows of data.

3. Upload the file to the AI, along with a clean-up prompt:

Again, you’ll need to download the file and check it manually. Everything look ok? Anytime you have AI manipulate data, you’ll need to check the data.

Now you’re ready for the money prompt:

You should be looking at a list of great ideas that are perfect additions to your blog. These topics are likely closely related to, but not exactly like, articles you’ve written in the past.

• They should be easy to write. You’ve written pieces of them many times.

• They should be internal linking opportunities. They’re adjacent to many articles you’ve already written.

• They should be keyword opportunities. They are semantically related to your other articles.
  

Because you started with GA4 data showing results, you can ask the AI to predict the performance of these topics and even prioritize them based on the likelihood of success. Here’s the prompt:

That last prompt is fun if you’re headed to a meeting. Your peers will think you’re a sorcerer.

Next, let’s do a gap analysis in the middle of the funnel, where big content lives. Long-form formats like gated guides can show deep expertise and grow email lists. Downloadable formats can be saved and shared, helping to keep the brand top-of-mind.

But what’s missing from that long-form piece you made? Are there gaps? Can AI help you spot them?

Yes. For this example, I’ll use the biggest format we have: our book, Content Chemistry. It’s the “Illustrated Handbook for Content Marketing,” and it’s supposed to be comprehensive. Of course, it doesn’t cover everything, but maybe AI can tell us what we missed.

For this example, we had to take all the text out and put it into a simple .txt file. But for a typical PDF guide, that probably isn’t necessary. Either way, take the file and upload it to your favorite LLM along with this simple prompt:

The AI will quickly spot the topical holes. Like magic, it quickly lists everything you failed to cover.

You’ll probably already know about most of them. They may be unimportant or boring to you, but are they important to your reader? If so, consider a revision.

We now arrive at the bottom of the funnel.

Really, we should have started here because improvements at the bottom have a bigger impact on business outcomes. Visitors who start their experience on a service page are generally ten times more likely to convert into a lead than visitors who start on a blog post, at least for B2B service brands like ours.

Want to check for yourself? Head over to GA4 and create a few comparisons to the Landing page report. Then, use the dropdown under Session conversion rate to select your lead gen conversion. The difference is huge.

Why do visitors who start on service pages convert into leads? Because they have commercial intent. Every SEO learned this on day one of keyword research class. But let’s ask a question less common in SEO school…

Why DON’T visitors who start on service pages convert into leads? It’s usually because the page failed to answer a question (a lack of clarity). Or it failed to address an objection (a lack of confidence). Or it failed to support claims with evidence (a lack of trust). In other words, there are gaps.

So, let’s fire up our gap-finding AI and ask it to tell us what’s missing. But we’ll have to use a multi-prompt process because the AI doesn’t yet know our visitors. Our gaps are specific to our target audience, so we need to begin with a persona.

If you have a battle-tested ideal client profile, you can simply upload it to the AI. If you don’t, you can use the AI to create one for you. Use this handy persona prompt:

If you use ChatGPT, it will probably name your persona Alex. I have no idea why. And naturally, there will be accuracy issues. So, take time to read it closely. What’s incorrect? What’s missing? Fix it first. If you don’t, the subsequent AI responses will be bad.

You fill these gaps yourself by telling the AI to make changes. Here’s your persona-repair prompt:

It won’t be perfect, and it doesn’t have to be. But it has to be good enough to use in an evaluation of your service pages. The more accurate it is, the better the responses will be.

Next, we’ll give it the page we want to optimize. In the same chat with the persona, copy and paste the copy from a key service page (or a full-page screenshot) into the AI with the following prompt:

Does that look like a long prompt? In my experience, it is not. Some of the best prompts give the AI a little list of best practices so it knows what a successful deliverable will be. The prompt is like a micro-blog post on best practices. Use your best practices and what you know to work well based on your experience, your analytics, and your prospect.

Here’s a screenshot of a ChatGPT response. It starts by listing the ways in which you satisfied your persona’s information needs and ends with a list of the persona's information needs that were left unmet.

Look at the gaps! Did we fail to answer questions that are important to our prospect? Remember, every unanswered question and unsupported claim can hurt your conversion rates.

Now you know how to make a better page. You can do it by hand or have AI make some quick suggestions for you. Here’s a final prompt that will have AI recommend page edits based on the gap analysis.

If the page is getting traction from Google, be careful not to de-optimize for search as you optimize it for conversions. Usually, filling in gaps helps both conversions and rankings. You can improve the mousetrap without removing the cheese.

You can devise your own prompts to have the AI recommend edits that will both improve the relevance for SEO and the conversion clarity for your human visitor!

Whatever you're working on, AI can give you a quick point of view on what’s missing. It works for pretty much anything: PPC ads, email subject lines, calls to action, titles and meta tags, YouTube descriptions, and PowerPoint sales presentations.

If you like to use AI to create new things rather than audit existing things, just change the prompt. The “create draft” prompt and the “audit page” prompt are nearly the same.

The magic is in the persona and the best practices you provide above that prompt.

And remember, always proceed with caution!