Why category entry points belong in every AI search strategy

For two decades, marketers have built their content around keywords. But now, AI has changed how people search. They’re able to describe situations in their own words, and that gives content teams a clearer view of the moments behind their needs. 

Marketing science calls these category entry points (CEPs): the situations that prompt a buyer to think about a category and recall possible brands.

Here’s what that means in practice. Say your team’s organic traffic is dropping. The keyword that captures this is “how to increase organic traffic.” The keyword has search volume, the SERPs are clear, the work is straightforward.

But the keyword doesn’t capture what the person is actually dealing with. They can’t yet tell what’s causing the drop: an algorithm change, AI Overviews, or their own content slipping. They’ve read articles about technical SEO and aren’t sure if that’s even the issue. They need help diagnosing before any how-to will help.

That underlying situation is the CEP. In this case, it’s “our organic traffic is dropping and we can’t tell why.” In AI search, the buyer can describe that CEP directly: “Our organic traffic has dropped 30% over six months and I can’t tell if it’s an algorithm change, AI Overviews, or our own content slipping. What can I do?”

Over the past several months, I’ve tested whether anchoring content to CEPs would change how AI systems surfaced Semrush’s work. 

The short answer is yes. One article has been cited every week for over four months. Another lifted share of voice in its target topic cluster from 15% to 26% in the week after publication.

This piece shares what I found and how you can start.

The marketing idea behind our experiment

Category entry points predate AI search by more than 15 years. The framework comes from Byron Sharp’s How Brands Grow (2010), one of the most rigorously evidenced books in marketing science. 

Sharp and his colleagues at the Ehrenberg-Bass Institute used large-scale purchase data across dozens of categories to show that brand growth depends on mental availability: being recalled in the moments that trigger category need. 

A CEP is one of those moments, and they happen all the time.

Think about driving home late at night, hungry, with most restaurants closed. McDonald’s pops into your head. Maybe Taco Bell does too. You weren’t necessarily craving either one, but the situation triggered the category, and a few brands came with it. 

That’s mental availability.

The same thing happens in B2B. For a project management tool, one CEP is the moment a small team outgrows informal coordination. A buyer in that moment might describe it as: “my team just grew past five people and coordination is breaking down.” Asana pops into their head. Maybe Monday or Trello.

For an SEO platform, a CEP might be the moment a team suspects AI search is eating their traffic but can’t confirm it. The buyer might say: “I think I’m losing traffic to AI search and I don’t know how to tell.” Semrush pops into their head. Maybe a few others.

I anchored our experiment in CEPs because they gave us a principled way to define what a content topic should be — a specific moment of need, the kind of moment a buyer might describe in an AI prompt.

Why CEPs fit AI search

CEPs fit AI search for three main reasons:

1. Prompts can give us a direct view of the situations buyers are in
2. One CEP can capture many prompts buyers use for the same situation
3. Mental availability, which CEPs are fundamentally about, is finally measurable

Prompts make CEPs visible

In AI search, buyers can describe their full situation in their own words. We can find the CEPs behind those descriptions and build content around them.

Then, when a buyer turns to AI to describe that situation, our article shows up in the answer because we wrote it for that situation.

One CEP appears in many prompts

Buyers in the same situation can phrase their prompts using different words, at different levels of specificity, and with different emotional registers.

For example, our article “Why are competitors winning AI search?” addressed the CEP we identified as: I’ve noticed my competitors showing up in AI answers and we’re not. 

Over nearly five months, AI systems retrieved the article across dozens of distinct prompts, all describing that situation in different ways. Some were highly specific (“why does [competitor] appear in ChatGPT responses for ai?”). Others were more general (“how do I get my brand in AI search results?”).

Mental availability becomes measurable

Sharp’s argument is fundamentally about mental availability: whether a brand is associated with the moment someone first thinks “I might need this kind of product.” 

That association has historically been hard to measure. We relied on surveys, unprompted recall studies, and other slow, noisy signals.

AI search now lets us see that association more directly.

The clearest signal is through a brand mention in the answer itself. Meaning your brand has been recalled at the moment of need. A softer signal is through a citation of your content as a source: the AI judged your content relevant to the moment, even without naming the brand.

Mentions and citations are both new mental availability signals. Neither was measurable before AI search. That’s one thing I thought made the experiment worth running.

How we ran the experiment

The experiment had three phases:

1. Identifying the category entry points we most needed content for, 
2. Writing articles built around those situations
3. Tracking how those articles performed across AI platforms

Identifying the CEPs

I started by mapping the prompts buyers were using in our category. The inputs came from three places: prompt data inside Semrush Enterprise AIO, conversations with our sales and customer success teams, and the kinds of questions we kept seeing in support tickets and on social.

From that mapping, I drew out the underlying situations. The moments that brought someone to an AI tool in the first place, like “I think my competitors are showing up more than us” or “I don’t know whether AI search is sending us traffic.” 

Then I filtered for situations Semrush had a right to own: places where our tools, our data, and our expertise were genuinely relevant, and where we weren’t yet well-represented in AI-generated answers.

Building the articles

For each CEP, the team wrote the article from inside the situation. 

We framed each title as the kind of question a buyer in that situation might naturally ask. “Why Are My Competitors Showing Up in AI Search and Not Us?” reads naturally because it expresses the CEP in the buyer’s own voice.

Within each article, some H2s mirrored specific prompts that fell under the CEP. Openings acknowledged the situation directly, skipping the usual definitions and category overviews.

And we built each article to address the CEP head-on, in natural language, with no marketing fluff.

Measuring AI visibility

I tracked performance usingSemrush Enterprise AIO across 1,758 prompts in our category clusters. 

For each article, I measured both signals from the previous section: citations (when our article was retrieved as a source) and brand mentions (when “Semrush” appeared in the answer itself).

I tracked five metrics:

1. Citation volume: weekly citations per article across ChatGPT, Google AI Overviews, and Google AI Mode
2. Prompt breadth: number of distinct prompts that cited each article
3. Model mix: citation distribution across the three platforms
4. Share of voice (SOV): Semrush vs. competitor mentions in each article’s topic cluster
5. Brand mentions: how often “Semrush” appeared in the AI answer when the article was cited

What changed when we anchored content to CEPs

When we anchored content to CEPs, two things changed: citation volume compounded over months on the same articles, and brand share of voice lifted in their topic clusters.

What the citation data shows

Citations compounded on the same articles for months. The articles where this happened had a clear CEP and content that covered it thoroughly.

“Why are competitors winning AI search?” peaked around week eight and held at roughly half that level for the four months that followed. 

Two more recent articles, “AI citing my site vs. third-party sources” and “Fix AI brand misinformation,” showed the same trajectory shape early in their run.

The articles that didn’t compound told me what mattered.

AI systems cited “Catch-up on AI search” across more distinct prompts than any other article in the set, then stopped citing it after five weeks. Prompt breadth alone wasn’t enough. What mattered was whether AI kept citing the article for the same prompts: whether the article was the answer to a specific, recurring situation.

We published “AI Overviews traffic loss” the same day as the top performer, and it covers a closely related topic. But it never broke into meaningful citation volume. The reason was we built it around a topic concern, not quite a CEP. The top performer started with a specific buyer situation, and that’s what AI search kept matching to.

One pattern across all articles: Google AI Overviews drove the bulk of citations on the articles that compounded. ChatGPT was the most consistent week over week. Google AI Mode was the most volatile, sometimes dominating an article’s citations and other times dropping near zero.

How citations translate to brand visibility

I also tracked share of voice and brand mentions to understand what those citations translate into.

For “AI citing my site vs. third-party sources,” Semrush mentions across the prompts that cited the article rose roughly 30% in the two weeks after publication. 

In that same article’s primary topic cluster, share of voice rose from 15% the week before publication to 26% the week after, while the broader AI Visibility benchmark moved only from 21% to 22%. 

The lift was stronger than background movement, though the post-publication window is still early.

However, the pattern doesn’t always look this clean. 

For “Why are competitors winning AI search?”, mentions across the article’s topic cluster roughly doubled in the weeks after publication. The rise had started six to eight weeks earlier, climbing through November and December 2025. Other activity in the cluster was already building momentum, and this article extended it rather than triggering a new step-change.

And, as we know, citations and mentions aren’t the same outcome. When I manually reviewed AI responses for top-cited prompts, I identified four distinct citation patterns:

1. Article cited inside the response and shown in the side panel
2. Article cited only in the side panel
3. Article cited inside the response but not shown in the side panel
4. Semrush mentioned explicitly in the answer itself

In most cases, the article served as a supporting source. 

Semrush’s name appeared in the side panel as a byproduct of the article being retrieved. Direct brand mentions in the answer body were the exception. 

Citations drive traffic and signal authority. Mentions build brand recall by putting your name in the answer itself. The two don’t always move together.

Where you can start

Start with a list of the situations that bring buyers into your category. These are your CEPs.

Sit down with your sales team, your customer success team, the people who hear what buyers actually say, and write down 20 real moments. Specific situations like: “the moment our customer first realizes they have this problem,” “the moment a competitor’s name comes up in their head,” “the moment they decide it’s worth doing something about.”

Then check your existing content against the list. Some moments will be well-covered. Others won’t. The uncovered ones are where CEP-anchored content has the most room to perform. The gap between buyer reality and what’s available is widest there.

For example: 

One of the moments we wrote down was: “I’ve noticed my competitors showing up in AI answers and we’re not.” Our existing content covered the broader topic of AI search visibility, but nothing addressed that specific situation. We wrote “Why are competitors winning AI search?” around it. The article opens with that exact moment, walks through how to diagnose it, and ends with what to do. That’s the article that compounded citations for four months straight.

Write the article you’d want to find if you were the person typing that situation into ChatGPT. Four principles matter when you start writing:

1. Frame each one around the situation itself
2. Use natural language a real person would use
3. Give each section a single clear job
4. Keep the structure scannable without sacrificing depth

These principles describe what AI search actually rewards: content built for real buyer moments, written clearly for the people in those moments.