AI search introduces a second marketing funnel between the information brands create and the recommendations consumers receive.
For decades, brands have optimized a traditional marketing funnel to move consumers from awareness to purchase.
Today, a second marketing funnel is emerging that explains how AI models discover, process, and share recommendations with users.
The emergence of AEO appears to add a second marketing funnel to the user journey. While traditional marketing funnels guide users toward conversions, a selection funnel helps determine where a brand disappears between the information available to an AI model and the recommendation a consumer receives.
The selection funnel adds a second journey ahead of the traditional marketing funnel.
How AI Models Eliminate Brands From Answers
A user asks ChatGPT: “What is the best mineral sunscreen for sensitive skin?” The model's answer includes a handful of mineral sunscreens relevant to the user's search. What’s invisible to most users and brands is everything that happened before the sunscreen options appeared. What product information was available to the model? Which sunscreen products were included as potential candidates but eliminated in the model's final answer? Which product attributes influenced the model's final sunscreen recommendations?
Google documented one version of the upstream retrieval process. Google's generative search systems often use a “query fan-out,” generating multiple related searches to retrieve additional information before developing a response. The fan-out process produces an internal series of searches for relevant terms to the user's query. Retrieval varies by platform, but most AI search experiences use query fan-out or similar multi-query and iterative search processes for user requests.
Most retrieved candidates are eliminated before the user sees the answer.
For the prompt “What is the best mineral sunscreen for sensitive skin?” the fan-out searches explore different parts of the user's request: mineral sunscreen, sensitive skin, ingredients, dermatologist recommendations, reviews, product attributes, and other relevant criteria. The search results create opportunities to retrieve information about different products. A brand's product page, retailer listing, review, editorial article, or other source may appear because the source helps answer one or more parts of the question.
But retrieval does not ensure a recommendation. Retrieval gives the model information to evaluate, but the model must then determine which products are relevant to the user's request and whether claims about an eligible product are consistent and credible. The evaluation process means the majority of the sunscreen candidates retrieved won’t be recommended. The model might find a product candidate lacks enough evidence connecting the sunscreen to sensitive skin or conflicting information about the formula. Another product might have a dermatologist endorsement for the requested attributes. AI models weigh variables like dermatologist endorsements and ingredient consistency differently.
The evaluation process is central to the selection funnel:
- Eligibility: Can the system access usable brand information, like ingredient accessibility?
- Retrieval: Does information about the brand surface when the system searches for material relevant to the customer’s question?
- Consideration: Does the brand survive dynamic filtering processes and enter the set of possible products being evaluated?
- Representation: Does the available evidence accurately reflect the brand, product offerings, and competitive differentiation?
- Recommendation: Does the evidence support putting the brand in the final answer?
Each stage represents a point where a brand can move forward or disappear from a model’s recommendation.
High AI Visibility Can Mask Weak Recommendation Performance
Most AEO measurements start with visibility. Did the brand appear in the output? How often was the brand mentioned? Was the brand cited? What percentage of the model's answers included the brand? Visibility metrics tell us what happened at the end of the model’s recommendation process, but they don’t explain why it left a brand out.
A visibility gap can originate at any of the five funnel stages.
Assume a sunscreen appears in 5% of AI recommendations while a leading competitor appears in 35%. The visibility number identifies a gap, but the number doesn’t explain why the gap exists. The brand could be failing at any of the five funnel stages, from lacking accessible ingredient data to losing the final recommendation because a competing product better satisfies the query requirements. Calling each potential outcome an “AI visibility problem” collapses several different challenges into one metric.
The Selection Funnel Explained
The funnel sequence is critical because retrieval doesn’t guarantee retrieved information will influence the model's answer.
Each selection-funnel stage answers one diagnostic question.
Anthropic provides a useful example: its web-search architecture supports dynamic filtering that processes search results before the information reaches Claude’s context window. Information can surface during search and be filtered out before it reaches the material Claude uses for the response. Separating retrieval from consideration and analyzing the stages in order prevents brands from spending time and money on problems that don’t exist.
A brand that is consistently retrieved does not need a technical audit. A product that is consistently misclassified needs a content audit, not another generic blog post. A product that is represented accurately but consistently loses recommendations needs a different investigation.
OpenAI’s shopping documentation explains how ChatGPT crawls the internet for relevant information, including price, availability, reviews, and product specifications. OpenAI also measures product accuracy by whether recommended products meet a shopper’s requirements, such as price, material, and specifications. Following that guidance, the hypothetical sunscreen can be represented accurately, but it can still lose the recommendation because another product better matches the customer’s requirements. Losing a recommendation is not the same failure as being absent from retrieval.
The Selection Funnel Feeds the Traditional Marketing Funnel
A selection funnel doesn’t replace the traditional marketing funnel; it sits in front of it. The recommendation stage is where the selection funnel hands the customer back to the traditional marketing funnel. At the recommendation stage, a customer can receive the model's recommendation and reject the product because of price, reviews, familiarity, availability, retailer preference, or another purchase consideration. The handoff leaves marketers with at least two connected user journeys to measure.
The recommendation stage hands the customer back to the marketing funnel.
Selection Funnel
Eligibility → Retrieval → Consideration → Representation → Recommendation
Combined with:
Traditional Marketing Funnel
Awareness/Consideration → Preference → Selection → Purchase → Loyalty
The framework helps marketers answer a question a visibility score can’t: Was a brand eliminated from an answer before the model recommended the brand, or did the customer see the recommendation and choose something else? The breakdown may occur after the selection funnel has done the assigned job.
No Single Marketing Discipline Influences the Selection Funnel
The selection funnel doesn’t operate independently from traditional marketing.
Evidence from any discipline can shape an AI-generated answer.
PR coverage provides third-party evidence. SEO impacts how many pages are accessible and discoverable online. Product pages establish first-party specifications and claims. Retailer pages provide category, pricing, and availability information. Reviews document customer experiences. Merchant feeds provide structured product information. Editorial coverage creates comparisons, evaluations, and outside perspectives. Brand positioning influences how a company describes the brand category and competitive set.
AEO cuts across marketing disciplines because the evidence influencing an AI-generated answer may come from any discipline.
That relationship looks something like this:
Marketing creates evidence, which AI systems retrieve, filter, and interpret, leading to answers that consumers receive and use to make decisions.
The sequence is one reason we don’t treat AEO as rebranded SEO. SEO remains and will continue to be a relevant part of the AI search equation. But a number one SEO rank in organic search on Google or Bing can’t explain why a brand was retrieved but excluded from an AI model's consideration. Therefore, brands should approach AEO as the discipline of understanding what happens when a machine becomes an intermediary between marketing activity and consumer consideration.
A Brand Can Win One Funnel and Lose the Other
The two-funnel model can make traditional visibility reporting misleading. A brand can have strong consumer awareness and weak AI retrieval. A brand can rank on the first page of Google and be absent from ChatGPT’s recommendations for a similar question. A brand can appear frequently in AI answers but rarely receive an endorsement. A brand can earn the recommendation and still lose the sale.
A brand can pass one funnel and be eliminated in the other.
The question most brands ask regarding AI visibility right now is, “How visible are we in AI?”
The better question to evaluate a brand's performance in an AI model is, “How far does our brand make it through the AI search funnel? Where does the brand get eliminated, and why does the elimination happen?”
Visibility is an important checkpoint, but recommendation and ultimately selection should be the goal of a successful selection funnel.
AEO gives brands a new customer journey to diagnose: what happens between the evidence a brand puts into the market and the AI recommendations the brand receives. The traditional marketing funnel tells us whether customers choose a brand, whereas a selection funnel tells us if a brand has the opportunity to be considered at all. Marketing teams have spent decades measuring the first journey; now we need to measure the second.
