Buyers used to research a purchase by clicking through ten blue links. Now they ask an AI model, and the model does the clicking for them. It reads the sources, builds the shortlist, weighs the tradeoffs, and hands back an answer that often names one or two brands.
That answer is not one moment. It is four. The questions a buyer asks change as they move from "what exists?" to "which one?" to "was it worth it?", and AI answers each kind of question differently, from different sources, with different amounts of confidence.
This guide explains the four stages, what AI is doing in each one, and why the brands that win early in the journey are not always the brands that win at the end.
The four stages
| Stage | Name | The buyer's question | What AI is doing |
|---|---|---|---|
| TOFU | Awareness | "What are my options?" | Mapping the category and naming the players |
| MOFU | Consideration | "How do these compare?" | Building a shortlist of 3 to 5 brands and comparing them |
| BOFU | Decision | "Which one should I pick?" | Making a recommendation against the buyer's specific constraints |
| EVFU | Extended Value | "Will this hold up after I buy?" | Judging trust, support, and what existing customers say |
Each stage has its own guide:
- TOFU: Does AI even know you exist?
- MOFU: Are you in the comparison?
- BOFU: Does AI recommend you, or the competitor?
- EVFU: Do your customers reinforce you, or talk buyers out of you?
Why the stages behave differently
Three things change as a buyer moves down the funnel.
1. The question gets narrower. A TOFU question has one or two variables: a category and maybe a region. A BOFU question can carry five or six: category, region, audience, use case, budget, timeline, and a named competitor to beat. Every added constraint shrinks the set of sources that can answer it.
2. The brand name shows up in the prompt. TOFU questions almost never name a brand. By BOFU and EVFU, most do. That sounds like good news, but it hides a trap: a brand can look strong when asked about directly and still never get chosen when the buyer asks an open question. We call that pattern the Branded Bubble, and it is one of the most common failures we see.
3. AI runs out of third-party sources. This is the most important shift, and it is covered in every stage guide under "Where AI looks for answers."
Where AI looks for answers, across the whole funnel
At the top of the funnel, AI has plenty to work with. Listicles, directories, review sites, industry publications, forums, and Wikipedia all describe categories and name the players in them. Plenty of people have written "the best X in Y."
As the buyer layers on constraints, that coverage thins out fast. Very few independent sources answer "which provider in (Region) is best for (use case) for (Target Audience) at this budget, and how does their onboarding compare to (competitor)?" Nobody has written that article.
When third-party coverage runs out, AI does one of three things:
- It falls back to the brand's own content: pricing pages, service pages, documentation, FAQs, support policies, and case studies.
- It hedges: "you should contact the provider to confirm," "pricing varies," "reviews are mixed."
- It picks the brand whose information it could find, even if that brand is a worse fit.
So the deeper a buyer goes, the more your own website stops being marketing and starts being the source of record. At BOFU and EVFU, the brand whose specifics are clear, complete, and easy for a machine to read usually wins the recommendation. The brand that says "contact us for details" usually loses it, not because it is worse, but because AI could not confirm it was better.
| Stage | Constraint density | Third-party coverage | Reliance on your own content |
|---|---|---|---|
| TOFU | Low | High | Low |
| MOFU | Medium | Medium | Medium |
| BOFU | High | Low | High |
| EVFU | High | Low (except reviews) | Very high |
How to measure your position at each stage
A single visibility score hides the stage where you are actually losing. The useful read is stage by stage:
- Run a fixed set of questions per stage. SurfaceMap uses 20 locked question templates, five per stage, each testing a different question shape. Locking them means your trendline measures movement, not prompt drift.
- Separate branded from unbranded questions. Being described well when named is not the same as being chosen when not named.
- Capture the fan-out. The sub-queries AI generates on its own tell you which sources it went looking for and what it could not find.
- Read the verbatim answers. The hedges, caveats, and "contact the provider" lines are where the fixable gaps live.
- Re-run on a cadence. Models retrain and competitors publish. A snapshot is a starting point, not a verdict.
Find your stage
Most brands are strong at one or two stages and weak at the others, and the fix for a TOFU gap looks nothing like the fix for a BOFU gap. Start with the stage guide that matches your biggest worry, or skip the guesswork:
- Take the 2-minute AI Mention State diagnostic to see which of the 12 states you are likely in.
- Run a free audit to see exactly what ChatGPT, Claude, Gemini, and Perplexity say about you at every stage. No card required.
FAQ
What are TOFU, MOFU, BOFU, and EVFU in AI visibility? They are the four stages of the AI buyer journey: Awareness (TOFU), Consideration (MOFU), Decision (BOFU), and Extended Value (EVFU). Each stage reflects a different kind of buyer question and a different way AI models gather and weigh information.
What does EVFU stand for? EVFU stands for Extended Value. It covers the post-purchase experience: implementation, support, and referral, where existing customers either reinforce or undercut how AI describes a brand.
Why does AI recommend a competitor even when my brand is well known? Often because the buyer's question contains constraints that no third-party source answers, so AI relies on whichever brand's own content states those specifics clearly. Awareness gets you into the conversation. Clear, specific, machine-readable information wins the decision.
Can a brand be visible in AI answers but still lose? Yes. A brand can appear in TOFU lists and branded questions while never being chosen in unbranded decision-stage questions. Stage-by-stage measurement is the only way to see that gap.
See where you stand at every stage
Run a free audit to see exactly what ChatGPT, Claude, Gemini, and Perplexity say about you from awareness to decision. No card required.
See Where You Stand