For twenty years, entrepreneurs have constructed their content material round key phrases. However now, AI has modified how folks search. They’re capable of describe conditions in their very own phrases, and that offers content material groups a clearer view of the moments behind their wants.
Advertising and marketing science calls these class entry factors (CEPs): the conditions that immediate a purchaser to consider a class and recall attainable manufacturers.
Here is what meaning in observe. Say your crew’s natural visitors is dropping. The key phrase that captures that is “how one can improve natural visitors.” The key phrase has search quantity, the SERPs are clear, the work is easy.
However the key phrase does not seize what the individual is definitely coping with. They can not but inform what’s inflicting the drop: an algorithm change, AI Overviews, or their very own content material slipping. They’ve learn articles about technical search engine marketing and are not positive if that is even the difficulty. They need assistance diagnosing earlier than any how-to will assist.
That underlying scenario is the CEP. On this case, it’s “our natural visitors is dropping and we will not inform why.” In AI search, the customer can describe that CEP instantly: “Our natural visitors has dropped 30% over six months and I am unable to inform if it is an algorithm change, AI Overviews, or our personal content material slipping. What can I do?”
Over the previous a number of months, I’ve examined whether or not anchoring content material to CEPs would change how AI techniques surfaced Semrush’s work.
The brief reply is sure. One article has been cited each week for over 4 months. One other lifted share of voice in its goal subject cluster from 15% to 26% within the week after publication.
This piece shares what I discovered and how one can begin.
The advertising and marketing thought behind our experiment
Class entry factors predate AI search by greater than 15 years. The framework comes from Byron Sharp’s How Brands Grow (2010), one of the vital rigorously evidenced books in advertising and marketing science.
Sharp and his colleagues on the Ehrenberg-Bass Institute used large-scale buy knowledge throughout dozens of classes to indicate that model development is determined by psychological availability: being recalled within the moments that set off class want.
A CEP is a type of moments, and so they occur on a regular basis.
Take into consideration driving dwelling late at evening, hungry, with most eating places closed. McDonald’s pops into your head. Perhaps Taco Bell does too. You were not essentially craving both one, however the scenario triggered the class, and some manufacturers got here with it.
That is psychological availability.
The identical factor occurs in B2B. For a venture administration software, one CEP is the second a small crew outgrows casual coordination. A purchaser in that second would possibly describe it as: “my crew simply grew previous 5 folks and coordination is breaking down.” Asana pops into their head. Perhaps Monday or Trello.
For an search engine marketing platform, a CEP may be the second a crew suspects AI search is consuming their visitors however cannot affirm it. The client would possibly say: “I feel I am shedding visitors to AI search and I do not know how one can inform.” Semrush pops into their head. Perhaps just a few others.
I anchored our experiment in CEPs as a result of they gave us a principled option to outline what a content material subject must be — a selected second of want, the form of second a purchaser would possibly describe in an AI immediate.
Why CEPs match AI search
CEPs match AI seek for three primary causes:
- Prompts may give us a direct view of the conditions patrons are in
- One CEP can seize many prompts patrons use for a similar scenario
- Psychological availability, which CEPs are essentially about, is lastly measurable
Prompts make CEPs seen
In AI search, patrons can describe their full scenario in their very own phrases. We are able to discover the CEPs behind these descriptions and construct content material round them.
Then, when a purchaser turns to AI to explain that scenario, our article exhibits up within the reply as a result of we wrote it for that scenario.
One CEP seems in lots of prompts
Patrons in the identical scenario can phrase their prompts utilizing completely different phrases, at completely different ranges of specificity, and with completely different emotional registers.
For instance, our article “Why are competitors winning AI search?” addressed the CEP we recognized as: I’ve seen my rivals displaying up in AI solutions and we’re not.
Over practically 5 months, AI techniques retrieved the article throughout dozens of distinct prompts, all describing that scenario in several methods. Some have been extremely particular (“why does [competitor] seem in ChatGPT responses for ai?”). Others have been extra basic (“how do I get my model in AI search outcomes?”).
Psychological availability turns into measurable
Sharp’s argument is essentially about psychological availability: whether or not a model is related to the second somebody first thinks “I would want this type of product.”
That affiliation has traditionally been onerous to measure. We relied on surveys, unprompted recall research, and different gradual, noisy alerts.
AI search now lets us see that affiliation extra instantly.
The clearest sign is thru a brand mention within the reply itself. Which means your model has been recalled in the intervening time of want. A softer sign is thru a citation of your content material as a supply: the AI judged your content material related to the second, even with out naming the model.
Mentions and citations are each new psychological availability alerts. Neither was measurable earlier than AI search. That is one factor I believed made the experiment value working.
How we ran the experiment
The experiment had three phases:
- Figuring out the class entry factors we most wanted content material for,
- Writing articles constructed round these conditions
- Monitoring how these articles carried out throughout AI platforms
Figuring out the CEPs
I began by mapping the prompts patrons have been utilizing in our class. The inputs got here from three locations: immediate knowledge inside Semrush Enterprise AIO, conversations with our gross sales and buyer success groups, and the sorts of questions we saved seeing in help tickets and on social.
From that mapping, I drew out the underlying conditions. The moments that introduced somebody to an AI software within the first place, like “I feel my rivals are displaying up greater than us” or “I do not know whether or not AI search is sending us visitors.”
Then I filtered for conditions Semrush had a proper to personal: locations the place our instruments, our knowledge, and our experience have been genuinely related, and the place we weren’t but well-represented in AI-generated solutions.
Constructing the articles
For every CEP, the crew wrote the article from contained in the scenario.
We framed every title because the form of query a purchaser in that scenario would possibly naturally ask. “Why Are My Rivals Displaying Up in AI Search and Not Us?” reads naturally as a result of it expresses the CEP within the purchaser’s personal voice.
Inside every article, some H2s mirrored particular prompts that fell below the CEP. Openings acknowledged the scenario instantly, skipping the same old definitions and class overviews.
And we constructed every article to deal with the CEP head-on, in pure language, with no advertising and marketing fluff.
Measuring AI visibility
I tracked efficiency usingSemrush Enterprise AIO throughout 1,758 prompts in our class clusters.
For every article, I measured each alerts from the earlier part: citations (when our article was retrieved as a supply) and model mentions (when “Semrush” appeared within the reply itself).
I tracked 5 metrics:
- Quotation quantity: weekly citations per article throughout ChatGPT, Google AI Overviews, and Google AI Mode
- Immediate breadth: variety of distinct prompts that cited every article
- Mannequin combine: quotation distribution throughout the three platforms
- Share of voice (SOV): Semrush vs. competitor mentions in every article’s subject cluster
- Model mentions: how typically “Semrush” appeared within the AI reply when the article was cited
What modified after we anchored content material to CEPs
After we anchored content material to CEPs, two issues modified: quotation quantity compounded over months on the identical articles, and model share of voice lifted of their subject clusters.
What the quotation knowledge exhibits
Citations compounded on the identical articles for months. The articles the place this occurred had a transparent CEP and content material that coated it completely.
“Why are rivals successful AI search?” peaked round week eight and held at roughly half that degree for the 4 months that adopted.
Two more moderen articles, “AI citing my website vs. third-party sources” and “Repair AI model misinformation,” confirmed the identical trajectory form early of their run.
The articles that did not compound informed me what mattered.
AI techniques cited “Catch-up on AI search” throughout extra distinct prompts than another article within the set, then stopped citing it after 5 weeks. Immediate breadth alone wasn’t sufficient. What mattered was whether or not AI saved citing the article for a similar prompts: whether or not the article was the reply to a selected, recurring scenario.
We revealed “AI Overviews visitors loss” the identical day as the highest performer, and it covers a intently associated subject. Nevertheless it by no means broke into significant quotation quantity. The rationale was we constructed it round a subject concern, not fairly a CEP. The highest performer began with a selected purchaser scenario, and that is what AI search saved matching to.
One sample throughout all articles: Google AI Overviews drove the majority of citations on the articles that compounded. ChatGPT was essentially the most constant week over week. Google AI Mode was essentially the most risky, typically dominating an article’s citations and different occasions dropping close to zero.
How citations translate to model visibility
I additionally tracked share of voice and model mentions to grasp what these citations translate into.
For “AI citing my website vs. third-party sources,” Semrush mentions throughout the prompts that cited the article rose roughly 30% within the two weeks after publication.
In that very same article’s major subject cluster, share of voice rose from 15% the week earlier than publication to 26% the week after, whereas the broader AI Visibility benchmark moved solely from 21% to 22%.
The carry was stronger than background motion, although the post-publication window remains to be early.
Nevertheless, the sample does not all the time look this clear.
For “Why are rivals successful AI search?”, mentions throughout the article’s subject cluster roughly doubled within the weeks after publication. The rise had began six to eight weeks earlier, climbing via November and December 2025. Different exercise within the cluster was already constructing momentum, and this text prolonged it reasonably than triggering a brand new step-change.
And, as we all know, citations and mentions aren’t the identical end result. Once I manually reviewed AI responses for top-cited prompts, I recognized 4 distinct quotation patterns:
- Article cited contained in the response and proven within the aspect panel
- Article cited solely within the aspect panel
- Article cited contained in the response however not proven within the aspect panel
- Semrush talked about explicitly within the reply itself
Typically, the article served as a supporting supply.
Semrush’s identify appeared within the aspect panel as a byproduct of the article being retrieved. Direct model mentions within the reply physique have been the exception.
Citations drive visitors and sign authority. Mentions construct model recall by placing your identify within the reply itself. The 2 do not all the time transfer collectively.
The place you can begin
Begin with a listing of the conditions that convey patrons into your class. These are your CEPs.
Sit down along with your gross sales crew, your buyer success crew, the individuals who hear what patrons truly say, and write down 20 actual moments. Particular conditions like: “the second our buyer first realizes they’ve this drawback,” “the second a competitor’s identify comes up of their head,” “the second they determine it is value doing one thing about.”
Then examine your present content material towards the listing. Some moments will likely be well-covered. Others will not. The uncovered ones are the place CEP-anchored content material has essentially the most room to carry out. The hole between purchaser actuality and what’s obtainable is widest there.
For instance:
One of many moments we wrote down was: “I’ve seen my rivals displaying up in AI solutions and we’re not.” Our present content material coated the broader subject of AI search visibility, however nothing addressed that particular scenario. We wrote “Why are rivals successful AI search?” round it. The article opens with that actual second, walks via how one can diagnose it, and ends with what to do. That is the article that compounded citations for 4 months straight.
Write the article you’d need to discover in case you have been the individual typing that scenario into ChatGPT. 4 rules matter whenever you begin writing:
- Body every one across the scenario itself
- Use pure language an actual individual would use
- Give every part a single clear job
- Hold the construction scannable with out sacrificing depth
These rules describe what AI search truly rewards: content material constructed for actual purchaser moments, written clearly for the folks in these moments.
