Home/Blog/PURCHASE PHASE
PURCHASE PHASE

Why brands disappear exactly when it's time to buy

A brand was present in 60.5 percent of AI answers during the awareness phase. By the purchase phase, only 31.6 percent. Weak exactly where money changes hands.

Felix Zeh·August 5, 2026·6 min read·AI-assisted
25%50%75%100%AWARENESS60.5 %CONSIDERATION64.4 %PURCHASE PHASE31.6 %
What you'll take from this article
  • Why strong visibility during awareness can be deceptive
  • How to tell whether your brand has the same problem
  • What content is missing at the purchase phase and where it lives instead
  • Six concrete steps to close the gap
60.5%presence during awareness (n=38)
64.4%presence during consideration (n=45)
31.6%presence at the purchase phase (n=79)

Some numbers look reassuring at first glance and mean the opposite on the second. This is one of those.

What exactly was measured?

In simulated buying conversations, we assigned every single AI answer to one of three phases: awareness, consideration, purchase decision. Then we checked, for each phase, in how many answers the brand under study showed up.

During awareness: 60.5 percent. During consideration: 64.4 percent. Stop reading there, and you'd see a brand that's doing well and even gaining ground.

31.6 %Presence at the purchase phase. Nearly a halving compared to consideration, and on the largest sample of the three phases.

Isn't this just too small a sample?

That's the first objection you'd expect, and it doesn't hold up here. The purchase phase, with 79 evaluated answers, is by far the largest of the three groups. Awareness and consideration together only reach 83.

The reason is the conversation mechanics: each conversation keeps going until the goal of that phase is reached. The purchase phase typically involves the longest back-and-forth, because that's where the concrete details get worked out. More answers get generated as a result. Which makes this collapse the best-supported finding in the entire analysis.

Why does presence collapse right there?

Look at the actual conversation flows and the pattern becomes clear fast. During the awareness phase, people ask about categories and basics: which battery system, what to look out for, what they even need. The AI answers with manufacturer names, because its sources support that.

At the purchase phase, the question changes. Now it's about availability, delivery time, actual prices, warranty terms, service commitments, which dealer is nearby. And for most manufacturers, exactly that information doesn't live on their own website. It lives with retailers, in reviews, in forums.

The AI does the obvious thing: it cites the source that can actually answer the question. That's usually not the manufacturer. A retailer states its price and delivery time, and a different brand may end up front and center in that answer.

Which makes the pattern doubly expensive. The brand loses visibility exactly at the phase with the highest purchase intent. And everything invested earlier in awareness and consideration runs into a wall at the final hurdle.

How to close the gap

The good news: this is one of the more solvable problems in GEO, because it's usually not a lack of authority, it's a lack of content.

  1. Check whether you're actually affected

    In a fresh chat, ask three purchase-close questions without your brand name: about price range, delivery time, the right dealer. If your brand shows up in none of them, you have the same pattern.

  2. Write down which purchase questions you don't answer

    Go through your product pages and flag what's missing: a non-binding price recommendation, delivery time, warranty length, service terms, dealer search. This list is usually surprisingly long.

  3. Publish a price range, even without an online shop

    A price band or a recommended retail price is enough. With no price information at all, your page simply drops out of the answer for every budget question, and retail takes the field.

  4. Mark up availability and delivery time so machines can read it

    Via Schema.org structured data, specifically the Offer, availability and deliveryTime fields. That's a manageable lift for IT and makes the information directly usable for models.

  5. Get service commitments out of the PDF

    Warranty terms and service commitments often exist only as a PDF, sometimes behind a login. As a plain HTML page with clear headings, they're citable; as a PDF, practically not.

  6. Structure your dealer data

    If your specialty retail network is your selling point, it needs to be findable: locations, hours, service scope, each as its own page with LocalBusiness markup instead of a search box an AI can't operate.

What you can realistically expect from this

We ranked these measures as top priority in the case we analyzed, but we didn't measure their effect. That would only be provable with a follow-up measurement after implementation, and that hadn't happened yet at the time of the analysis. Anyone who promises you a percentage here is guessing.

What we can say: the gap is about content, not technology. What's missing isn't authority, it's answers to specific questions. That's a considerably easier starting point.

Frequently asked questions

Does this also apply to service providers without product prices?

Yes, especially so. Service providers often have zero price information online, while competitors list package prices. On any budget question, whoever states a number at all wins by default. A price range or a starting-from price is enough.

We only sell through retail. Isn't that the dealers' problem?

For sales, yes. For AI visibility, no. If the purchase-close information sits exclusively with retail, the AI will, quite logically, cite retail at the purchase phase. Those answers often list an entire assortment of several brands. You've handed away control over the final decision stage.

Isn't the purchase phase lost to Amazon and the like anyway?

Maybe in classic search. In the AI conversations we measured, the picture was different: retail pages were well represented, but in the more recent survey the single most-cited source was the manufacturer's own domain, at 23 percent of all citations. The field is by no means claimed.

How do I know the measures actually worked?

Only through a follow-up measurement with the same model, the same question wording and the same number of runs. A single test after implementation proves nothing, because the swing between runs can be larger than the effect itself.

How this was measured

Every conversation continues through automatically generated follow-up questions until the goal of that phase is reached, which is why the sample sizes differ per phase: awareness n=38, consideration n=45, purchase phase n=79. Survey from July 26, 2026, model gpt-5.6-terra, 162 evaluated answers in total.

Transparency note · AI-generated content

A note on this article: research, analysis and writing were produced with the support of AI systems (the Ex Tenebris agent team) and reviewed editorially by Felix Zeh. Every figure cited comes from a real LUX/GEO analysis; the analyzed brand is anonymized to protect the client relationship.

FZ
Felix Zeh
Founder, Ex Tenebris · Stuttgart, Germany

Has spent over a decade at the intersection of marketing, data and analytics, and uses LUX to measure how brands show up in the answers of ChatGPT, Claude, Gemini and Perplexity. Questions or pushback on this article? kontakt@extenebris.de

Get in touch

How visible is your brand in the AI answer today?

The methodology behind this article is the same one we use for every assessment we run, tailored to your category, your competitors and your need states.