
Blog Post
Search Engine Optimization
Two phases aren't enough: Why GEO prompts need to map the customer journey in three stages
Table of Contents
Two Stages Aren’t Enough: Why GEO Prompts Must Map the Customer Journey in Three Steps
If you want to measure a brand's visibility in ChatGPT, Perplexity, or Google AI Mode, you need a prompt set that reflects the customer journey across at least three distinct phases: Awareness, Consideration, and Decision (often called Acquisition or Transactional in performance marketing). Two phases simply aren't enough. And that's not just marketing theory being pedantic—it's because that missing middle stage is precisely where modern AI systems recommend, compare, and filter out providers.

The Debate That Comes Up in Every Single GEO Audit
Every GEO audit kicks off with the same question: Which prompts are we actually measuring? Unlike traditional keywords, there are no search volumes, no steady rankings, and no static results. That makes your prompt set the single most important methodological choice of the entire audit—it determines what can even show up on your radar in the first place.
Our team recently had an open discussion about this. The suggestion on the table was to simplify things down to just two phases: Awareness (informational) and Acquisition (transactional). It’s an honest, understandable point. Drawing sharp lines between stages is genuinely tricky in day-to-day practice, and filling two buckets is naturally faster than managing three or four. Anyone who has ever tried neatly sorting a hundred prompts knows all about the borderline cases.
The counter-argument, however: Simpler and faster still has to be just as effective. And with only two phases, it simply isn't. This article explores why search intent research has recognized a middle tier for over 20 years, why consumer behavior studies pinpoint it as the sweet spot for brand building, and why fresh data from AI search confirms both. To wrap things up, we’ll share a practical classification rule to help you easily solve that categorization challenge.
What Does Search Intent Research Say About the Number of Stages?
Search intent research has recognized three core types since 2002, expanding in practice to four around 2015. A simple binary split between informational and transactional has never existed in the literature.
Andrei Broder, then CTO of AltaVista, laid the foundation with his Taxonomy of Web Search. He demonstrated that behind a query lies not just a need for information, but often the desire to reach a specific site (navigational) or perform an action (transactional). Building on this in 2008, Jansen, Booth, and Spink developed a three-level hierarchical classification, discovering that over 80 percent of all web queries were informational, while navigational and transactional each accounted for roughly 10 percent.
Modern SEO practice eventually rounded out this taxonomy with a fourth category: Commercial Investigation. Today, tools like Semrush, Ahrefs, and Sistrix consistently classify keywords into informational, navigational, commercial, and transactional. The reason is empirical. Between "What is X" and "Buy X now" sits a massive block of queries comparing solutions, criteria, and vendors: "best method for," "X vs. Y," "reviews of," "providers for." These queries aren't pure informational questions, nor are they finalized transactions. They represent the exact moment a shortlist is formed.
A quick note on precision: Google itself uses the primary categories Know (with the subcategory Know Simple), Do, Website, and Visit-in-person in its Quality Rater Guidelines. The term Commercial Investigation actually comes from SEO tools, not Google. In practice, however, both systems describe the exact same reality: There is an intermediate step between knowing and doing, and it is substantial enough to warrant its own category.
This taxonomy is the ideal benchmark for prompt sets, because a prompt is essentially a natural-language search intent. Cutting out the middle isn't just departing from academic convention—it means moving away from the very model that has best described search behavior for over two decades.
What Does Behavioral Research Say: Where Is Brand Preference Actually Formed?
Brand preference is built in the middle of the journey, not at the finish line. Three independent lines of research consistently prove this.
Google’s Messy Middle (2020). Google's consumer insights team, led by Alistair Rennie and Jonny Protheroe, partnered with behavioral economists to analyze what really happens between the initial trigger and the ultimate purchase. What they found wasn't a neat, linear funnel, but a loop. Users continually bounce between two mental modes: exploration as an expansive activity and evaluation as a reductive one, cycling through both until they’re ready to decide. Google dubbed this space the "Messy Middle," calling it the place where customers are won or lost. A two-phase model completely erases this space.
Bain’s Day 1 List (2022 and 2026). When it comes to B2B decisions, Bain & Company has observed the same pattern for years. Somewhere between 80 and 90 percent of procurement projects begin with a vendor shortlist the decision-maker already has in mind. About 90 percent ultimately buy from a vendor on that initial Day 1 List. A study by Bain and LinkedIn surveying 750 procurement leaders (February 2026) reinforces this 90 percent figure, while revealing that hidden influencers within the buying center drive half the impact on this list. Bain puts it bluntly: If you aren't on the Day 1 List, you've lost before the sales process even starts. And brands don't earn a spot on that list during the transaction phase—they earn it during the research leading up to it.
Gartner and 6sense on Independent Research. For complex B2B purchases, Gartner found that buying teams spend only a tiny fraction of their time interacting with vendors, dedicating the lion's share to independent research. 6sense estimates that around 70 percent of the journey is completed before the first vendor contact ever occurs. This research is pure, unadulterated Consideration—and it is migrating rapidly into AI systems.
All three research perspectives arrive at the same truth: If your brand only shows up in the Decision phase, statistically, you're already too late. A prompt set that doesn't measure this middle phase separately leaves the audit's most crucial question unanswered: Is our client already being mentioned during the research phase, before the user ever explicitly asks for vendor recommendations?
Why Not Just Stick with AIDA or Kotler's 5A Framework?
Models like AIDA and the 5As are hierarchy-of-effects frameworks, not intent frameworks. They describe a person's psychological state and how marketing communication influences them. A prompt set, on the other hand, models expressed intent—the actual words someone types into a search bar. Because these serve different purposes, the search intent taxonomy provides a far better fit.
You can clearly see this in the earliest stages. AIDA separates Attention and Interest. But when someone consults a search engine or an AI tool, they already have both attention and interest. When searching, those two initial steps merge into one: Awareness, in the sense of "I want to understand what my issue is." Consolidating the early stages makes total theoretical sense. The real flaw of the two-phase approach isn't at the beginning; it's right in the middle, where AIDA’s Desire and the 5A’s Ask represent the very stage that got cut.
The fifth stage of the 5As (Advocate) and the Loyalty phase of many journey frameworks certainly exist in AI search as well. Prompts like "How do I cancel X?" or "How do I properly use product Y?" reflect real, everyday queries. But for an audit aimed at measuring new customer acquisition, that phase works best as an optional add-on rather than part of the core set. More on that in a moment.
The clean takeaway: AIDA and the 5As give us the logic that a journey has multiple distinct steps. Broder and modern search intent research provide the stages we can actually observe in search queries. Both viewpoints naturally converge on three clear phases for acquisition: Awareness, Consideration, and Decision.
What Are LLMs Doing in the Middle of the Funnel? The Data from 2025 and 2026
AI systems have essentially taken over the Consideration phase. In traditional Google searches, users in this stage had to sift through ten blue links, build comparison tables, and weigh criteria themselves. ChatGPT, Perplexity, and Google AI Mode take on all that cognitive heavy lifting. Four data points bring this to light.
1. Awareness prompts usually end without a click and without brand mentions. In early 2025, Bain & Company found that roughly 60 percent of search queries ended without a click to a destination website, with 80 percent of users relying on AI summaries for at least 40 percent of their searches. For pure informational questions ("What is the difference between brownout and burnout?"), AI models rarely mention specific service providers or brands in the response. If your audit only checks Awareness and Acquisition, your phase-one measurements will be dominated by prompts where brand mentions are practically impossible by design.
2. AI answers heavily dominate evaluation prompts. Peec AI analyzed 500,000 business prompts in April 2026. Google AI Overviews surfaced for 86.7 percent of all queries showing purchase or evaluation intent, up from 56.9 percent in April 2025. For Decision-stage prompts, that figure rose to 88.5 percent. Coverage also climbed alongside prompt length: from 64.6 percent for two-word queries to about 89 percent for prompts spanning 11 to 15 words. These longer, consultative prompts are almost exclusively Consideration queries. Context matters here: This dataset focuses on commercial queries, so the percentage doesn't apply to all of Google Search, but specifically to the high-value segment marketers care about. In the EU, the rate hit 76 percent.
3. Comparison content earns disproportionate citations, but formats are evolving. Peec AI's Citation Benchmarks, drawn from over a million citations, show that 52 percent of listicles achieve a high citation rate on ChatGPT, even though they represent under a fifth of all cited URLs. This clearly signals that AI actively seeks out comparative and evaluative content when users are weighing options. At the same time, the data reveals that format preferences aren't set in stone. Following the GPT-5.6 update in early August 2026 (based on Peec AI's before-and-after comparison of identical prompts), listicles' share of ChatGPT citations dropped from 15.8 to 7.8 percent, and comparison sites fell from 9.1 to 6.2 percent, while product pages gained ground. The takeaway isn't that Consideration is losing relevance, but rather that it is shifting away from generic lists toward substantive, standalone answers. Without measuring this phase separately, you'll miss these dynamics entirely.
4. AI traffic converts better because the consultation already happened. Adobe Digital Insights has been tracking AI-referred traffic to US retail sites since October 2024. By March 2026, AI visitors were converting 42 percent better than non-AI traffic—a dramatic turnaround from a year prior, when they converted 38 percent worse. By May 2026, that advantage grew to 54 percent, accompanied by 53 percent longer session times and 23 percent more pages viewed per visit. Adobe's take: These users have already completed their research right in the chat and arrive pre-qualified. The entire Consideration phase played out in their conversation with the AI. While it's true that AI referrals still make up a modest share of total traffic, their high intent is clear proof of the value created in that middle stage.
A Real-World Example: A Managing Director Asks ChatGPT
A two-phase model quickly breaks down when faced with a realistic, three-step conversation. Picture a managing director of a mid-sized company in Germany's Lower Rhine region noticing a sharp dip in team morale. Their journey through AI search typically looks something like this:
Step 1: Awareness. They ask: "What is the difference between brownout and burnout, and what symptoms do employees show?" The AI responds with occupational health definitions, symptom checklists, and maybe a note on leadership habits. It won't recommend a boutique consultancy in their local area. A consulting firm's visibility score here will almost certainly be zero for brand mentions, with informational articles serving as sources at best.
Step 2: Consideration. They follow up: "What measures effectively counter brownout and 'quiet quitting' in mid-sized businesses? Is external organizational consulting worth it, or should we focus on internal leadership coaching?" Now, the AI acts as an advisor. It compares methods, outlines selection criteria for consultants, references proven frameworks, and cites consulting firms that have published whitepapers or case studies on this exact topic. The AI casually brings up brands as real-world examples, study authors, or methodology pioneers. This is where the user leans toward a specific type of solution, and this is where that mental Day 1 List takes shape—well before they ever search for specific vendors.
Step 3: Decision. Finally, they ask: "Which management consultancies in the Lower Rhine region specialize in quiet quitting, brownout prevention, and executive development?" The AI serves up a local shortlist of three to five providers along with their positioning and contact channels, pulling from service pages, business directories, review sites, and PR coverage. In the world of LLMs, this is the equivalent of Transactional Intent: People rarely book a high-touch consulting engagement directly in a chat, but they do ask for that final list of vendors to contact.
In a two-phase model, Step 2 gets awkwardly shoehorned into either Awareness (diluting the zero-brand baseline of definition queries) or Acquisition (mixing it with local vendor queries that rely on completely different sources and tactics). Either way, the audit fails to evaluate whether the client's website is recognized by the AI as an authoritative methodological resource. And that is precisely the factor that decides who makes the shortlist.
Interestingly, the two-phase approach often cites prompts like "Who consults companies on quiet quitting in the Lower Rhine?" as examples of Acquisition. For LLMs, that classification actually makes sense, since direct transaction prompts ("book now," "order") are rare, making the provider shortlist the final touchpoint in the chat. The real issue lies elsewhere: Between the informational query in Step 1 and the vendor query in Step 3, a two-phase model leaves zero room for Step 2. Yet that exact question—"consulting vs. coaching, and what criteria matter"—is where the user's shortlist of memorable brands is decided.
How Do You Solve the Boundary Issue Between Phases?
Blurred lines between phases are completely normal, but the solution isn't having fewer phases—it's having a clear, operational classification rule. Trimming down to two stages doesn't eliminate edge cases; it just leaves you with one boundary instead of two, and that single line runs right through your most critical cluster of prompts.
The most effective method is focusing on the expected response type. Don't ask: Which phase is the user in? Instead, ask: What is the AI expected to deliver?
Phase | The User Expects | Typical Prompt Patterns | What the AI Cites |
Awareness | An explanation | What is, how does it work, why, difference between | Encyclopedias, industry portals, guides |
Consideration | A list, comparison, or set of criteria | Which options, what to look for, is it worth it, X vs. Y, best method for | Comparisons, case studies, trade articles, forums |
Decision | A specific provider or actionable next step | Which provider in, who offers, pricing for, contact for, buy | Service pages, directories, reviews |
Using this simple guideline, roughly 90 percent of prompts can be categorized in seconds. For any remaining edge cases, simply assign a primary phase and move forward. The goal isn't absolute, debate-free perfection for every single prompt; it's consistent categorization that yields clear, actionable insights for each segment.
Two best practices from daily tracking workflows are worth keeping in mind. First, branded prompts—queries containing your client's brand name—should always be tracked separately. Once a brand is explicitly named in the prompt, visibility is practically guaranteed, which will heavily skew your general category benchmarks. Second, avoid blending everything into a single, combined visibility score. Mixing 50 broad Awareness prompts with 10 Decision prompts washes out the nuance and hides what's happening right near the point of purchase.
What Does a Solid Prompt Set for a GEO Audit Look Like?
A robust prompt set is structured as a matrix combining the three phases with the client's core offerings, keeping branded prompts in their own distinct group. Industry practice is steadily aligning around this setup: SE Ranking, Semrush, Search Atlas, and the default setups of several tracking platforms actively use Awareness, Consideration, and Decision, while Peec AI's prompt discovery defaults to a 25/50/25 distribution across informational, commercial, and transactional intent.
Element | Recommendation | Rationale |
Phases | Awareness, Consideration, Decision | Rooted in search intent research and the Messy Middle |
Second Dimension | One column per product or service | Visibility often varies dramatically across different offerings |
Weighting | Roughly 30%, 50%, and 20% | Slightly favors Awareness compared to the 25/50/25 baseline, reflecting fewer direct transaction prompts in LLMs |
Scope | 15 to 20 prompts per core service, 50 to 150 for a full business audit | Using only a single prompt per phase leaves findings vulnerable to normal LLM response variance |
Branded Prompts | Dedicated block, around 10% to 20% | Prevents category scores from becoming artificially inflated |
Reporting | Evaluated per phase, never as a single blended score | Blended metrics conceal critical performance bottlenecks |
The real beauty of using three phases becomes evident during analysis: Each phase uncovers distinct issues and points to clear, tailored solutions.
Underperforming in Awareness? The client likely lacks glossaries, research studies, clear definitions, and educational foundational content. The primary goal here is earning source citations rather than direct brand mentions.
Underperforming in Consideration? The client needs stronger method comparisons, real-world case studies, practical checklists, buyer's guides, and solution frameworks. The goal here is securing brand mentions in an advisory context—establishing the brand as a capable solution provider well before someone asks for vendor recommendations. This is what puts a client on that coveted Day 1 List.
Underperforming in Decision? The client probably lacks dedicated service pages, local entity signals, structured data markup, verified customer reviews, and listings in reputable directories. The goal here is winning direct recommendations.
This structured approach shows why the "two phases are faster" mindset is a false economy. Any time saved during initial setup is paid back tenfold in confusing reporting, as strategic questions and local vendor lookups get lumped together, resulting in vague recommendations. An audit that can't pinpoint whether the root issue lies in informational content or entity authority simply misses the mark.
Does a GEO Audit Need a Loyalty Stage Too?
For a purely acquisition-focused audit, no. But for clients with subscription or recurring-revenue models, it serves as a valuable fourth module. Post-purchase prompts are common in everyday AI interactions: "How do I configure product X?", "How do I cancel service Y?", "Is upgrading to plan Z worth it?" These queries directly impact churn, support overhead, and advocacy.
That said, they don't belong in the core prompt set because they answer an entirely different question. An acquisition audit evaluates how easily new customers can discover your brand. A loyalty module assesses whether current customers are getting accurate, helpful support. Both questions are valuable, but they call for different prompts, distinct benchmarks, and usually different stakeholders on the client side.
The key is to clarify this scope explicitly in the audit rather than simply skipping the phase without comment. A quick note does the trick: "This audit focuses on the three stages of new customer acquisition; post-purchase prompts can easily be incorporated as an additional module."
Frequently Asked Questions About Customer Journey Prompts
How many stages does a GEO prompt set need? Three for net-new customer acquisition: Awareness, Consideration, and Decision. This aligns with search intent literature dating back to Broder (2002), the four-part framework commonly used by SEO platforms (setting aside navigational queries, which rarely apply to LLMs), and Google's Messy Middle concept. A fourth stage for existing customers can be added as needed.
Why aren't two stages enough? Because that omitted middle stage is precisely where AI systems evaluate providers and construct shortlists. Pure informational queries rarely prompt AI to recommend brands, and direct transaction queries often lead users out of the chat. Genuine brand preference is built right in between.
How do I categorize a prompt into the right stage? Focus on the expected response. If the user expects an explanation, it belongs in Awareness. If they're looking for a comparison, a list, or criteria, it's Consideration. If they want a specific vendor recommendation or an actionable next step, it's Decision.
How many prompts should a GEO audit track? Aim for 15 to 20 prompts per core service and target market, balanced across all three stages with extra weight on Consideration, accompanied by a dedicated block of branded prompts. For a company with a handful of core services, that totals roughly 50 to 150 prompts. Using fewer makes it difficult to account for normal variations in AI responses.
Is AIDA the wrong model for prompt design? AIDA is a behavioral model designed to capture human psychological states. Prompts, on the other hand, reflect articulated intent. That makes search intent taxonomy a much better fit for the job. Still, both models arrive at the same conclusion: There is a critical, influential stage between knowing and doing.
What is the difference between Commercial and Transactional Intent in LLMs? Commercial intent revolves around exploring options, criteria, and comparisons ("Is external consulting or internal coaching the better investment?"). Transactional intent focuses on finding a specific provider or taking action ("Find consultancies near me"). In conversational LLMs, commercial queries appear far more frequently than purely transactional ones.
Bottom Line: Simpler Still Needs to Be Just as Good
Collapsing your framework into two phases might save an hour during prompt setup. But in exchange, you lose the ability to answer the most important question of the audit: Is your brand being recommended by AI while prospective buyers are weighing their options—before they ever ask for a list of vendors?
Using three phases isn't academic perfectionism. It's the standard model recognized by search intent research since 2002, proven empirically by Google's Messy Middle research, demonstrated in B2B by Bain's Day 1 List findings, and affirmed by recent AI search data from Peec AI and Adobe. You can easily solve the classification challenge with a simple rule based on the expected answer type. And breaking down your analysis by stage gives clients clear, diagnostic insights they can immediately put into practice.
If you're serious about GEO, you need to measure where choices are actually made. And those choices happen right in the middle.
Sources
Broder, A. (2002): A Taxonomy of Web Search, SIGIR Forum 36(2)
Jansen, B. J., Booth, D. L., Spink, A. (2008): Determining the informational, navigational, and transactional intent of Web queries, Information Processing & Management
Google, Rennie, A., Protheroe, J. (2020): Decoding Decisions: Navigating purchase behavior and decision making
Bain & Company (2026): Raising the Odds on a Deal: How Likelihood to Buy Rewires B2B Growth
Bain & Company (2025): Goodbye Clicks, Hello AI: Zero-Click Search Redefines Marketing
Peec AI (2026): What Does a Good Citation Rate Look Like? Benchmarks From Over 1 Million AI Citations
Search Engine Journal (2026): Google AI Overview Data Looks Different For Commercial Queries
Digital Commerce 360 (2026): Adobe: AI-referred traffic to retail sites doubles in a year
SE Ranking (2026): How to find and choose the right prompts to track for AI search visibility
Semrush (2026): How to decide which AI search prompts to track
Search Engine Land (2026): How to map AI search prompts to every stage of the sales funnel
Kotler, P., Kartajaya, H., Setiawan, I. (2017): Marketing 4.0. Moving from Traditional to Digital, Wiley (5A Framework)
Gartner: The B2B Buying Journey (Proportion of time buying groups spend with providers)
6sense: Buyer Experience Report (Portion of the journey completed before initial vendor contact)
Comments on the post
no comments yet
Write a comment
Your email address will not be published. Required fields are marked with *
