I have written about GEO as a discipline before — what separates Generative Engine Optimization from SEO and which levers work. This piece is the practical counterpart: What actually happens on the customer's side when search runs through a conversation instead of a results page? And what decides who the AI recommends?
The new customer journey has three steps — and no click path
The classic journey was measurable: search, click, website, form. The AI-search journey looks different:
- The question. Not "real estate agent Bonn", but: "I want to sell my house in Bad Godesberg — how do I find a good agent, and who would you recommend?" The question is longer, more specific, and carries the context nobody used to provide.
- The recommendation. The assistant names two or three providers — with reasons. Whoever appears here makes the shortlist. Whoever doesn't appear does not exist in this conversation, no matter how well they rank on Google.
- The direct contact. The prospect types the name directly or simply calls. In your analytics this shows up as direct traffic or brand search — the actual trigger stays invisible.
That explains an effect many companies are seeing right now: organic clicks stagnate, but the quality of inquiries goes up. Someone who arrives via an AI recommendation has already done the comparison.
Why real estate feels it first
It is no coincidence that clients now find me via "real estate + AI marketing". The industry has exactly the traits that make people ask an AI for advice instead of skimming a list:
- High stakes, rare decision. Selling a house happens once or twice in a lifetime. Nobody has routine; everybody wants a recommendation.
- A question of trust, not a product comparison. "Who should I choose?" is a question about people and evidence — exactly the format assistants are good at.
- Local authority counts. The AI justifies its recommendation with what it can verify: neighbourhood expertise, references, specialisation. Whoever shows that in machine-readable form gets cited.
The same logic applies to every advice-heavy service — tax advisors, agencies, fractional CMOs. Real estate is just the early indicator.
What an AI needs in order to recommend you
An AI does not recommend a brand it cannot understand. Four things decide whether you appear in the answer:
1. An entity it can grasp
Who are you, where are you based, what do you do, since when? That belongs on your website as Organization or Person markup — and it has to read the same everywhere: website, LinkedIn, directories. Contradictory facts cost you the recommendation.
2. Evidence instead of claims
"Leading agency" is worthless to a language model. "2,500 users in the first month after go-live" is quotable. Concrete numbers, named references and verifiable results are the currency an AI uses to justify its recommendation.
3. Answers to real questions
Assistants assemble answers from passages that can stand on their own. An FAQ section that fully answers "What does … cost", "How long does … take", "Does it work with …" delivers exactly those passages — as visible text and as FAQPage markup.
4. An open door for the crawlers
An llms.txt that describes what you stand for, and a robots.txt that doesn't accidentally lock out GPTBot, ClaudeBot and PerplexityBot. Sounds trivial — it is the most common reason technically solid sites are missing from AI answers.
The self-test: 30 minutes, no tools
How to find out where you stand
- Ask the customer's question. Open ChatGPT, Perplexity and Claude and ask the way your customer would: "I'm looking for [your service] in [your city] — who do you recommend and why?" Don't mention your name.
- Log the answers. Who gets named? With which reasoning? The reasoning reveals which evidence the AI found — and which you are missing.
- Ask about yourself. "What do you know about [your company]?" If the answer is thin or wrong, you know the cause: the machine cannot grasp you.
- Repeat monthly. Same questions, same rhythm, short log. Not a lab-grade metric, but a reliable trend — and the only one this channel offers.
Measuring what doesn't click
AI search barely shows up in analytics. Three sources replace the missing click path:
- The source question in your intro call. "How did you find me?" — asked consistently and written down. For me, this has become the single most informative marketing metric.
- Referrer traces. Referrals from
chatgpt.comandperplexity.aiappear when someone does click. Small in volume, big as a signal. - Server logs. Visits from GPTBot & co. show whether your content is being collected at all. No crawlers, no recommendations.
The honest conclusion
AI search rewards exactly what good work looks like anyway: clear positioning, verifiable results, complete answers. There is no trick that turns weak substance into a recommendation — but there is a lot of wasted visibility among companies whose substance would easily be enough. The machine just cannot read it.
If you want to see what this looks like in practice: for real estate companies the full approach is described on Real Estate Marketing with AI (German), the technical side lives on Propstack & onOffice Integration (German) — and across industries, the AI Marketing Use Cases (German) show which systems run in my own production setup.