AI Search in March 2026: What Customers Really Ask You

In March 2026, customers ask AI assistants like ChatGPT and Perplexity differently than Google: They describe their situation completely, mentioning budget, location, and desired outcome in one sentence. Those who only provide short standard answers get overlooked in important decisions. This article shows you how to collect real customer questions, build complete answers, and check your most important pages in 60 minutes—without special software and without GEO voodoo.

Here's the thing: Your customers no longer just google "tax advisor Munich". They open ChatGPT and write: "I'm self-employed, earn irregularly, and need tax advice in Munich. What does that cost roughly and what should I watch out for in the initial consultation?"

One single question. Five concerns. If your website only covers the keyword, you're out.

Google began the official rollout of AI Overviews in Germany, Austria, and Switzerland at the end of March 2025. Since then, users see AI-generated summaries at the top of search results. For you, this means: Visibility isn't just created through classic rankings, but also through whether your content is used as evidence for an AI answer.

A compilation published in June 2026 reports that 70% of respondents increased their use of AI for search in the past year; 25% indicated significantly stronger use, 45% somewhat stronger use (source). This isn't a distant future anymore. This is now.

Why Customers Ask AI Differently Than Google

Classic search queries are short. "Tax advisor Munich costs." Done. Google delivers ten blue links. You click through.

AI queries are dialogic and context-rich. People describe their problem as if talking to a colleague: "I'm self-employed, earn irregularly, and need tax advice in Munich. What does that cost roughly and what should I watch out for in the initial consultation?"

A single question combines multiple classic search intentions: information, comparison, costs, location, suitability, and next steps. For you, this means: Don't just collect keywords, but document question, situation, decision criterion, and desired action.

A study conducted in January 2026 with 524 people who had used an LLM search tool in the previous 30 days shows: AI search research shouldn't just measure finished search terms, but also evaluate real, freely formulated prompts (source). That's exactly what you should do for your own business.

Which Questions Particularly Often Need Advice

Not every question is equally valuable. Some lead directly to a decision, others are just curious. The most important questions have three characteristics: high personal risk, multiple selection criteria, and local or temporal dependency.

Questions with High Personal Risk

"What's the right solution for my situation?" Customers ask this question when a lot is at stake: health, money, legal consequences, reputation. A standard answer won't do here. This requires context, examples, and evidence.

"Which option fits my budget?" Customers don't want a price list. They want to know if they can afford your service and what happens if they pay less.

"What happens if something goes wrong?" This is the question nobody asks out loud, but everyone thinks. Whoever answers it wins trust.

Questions with Multiple Selection Criteria

Customers don't just compare prices. They compare: price, location, availability, duration, qualifications, guarantee or warranty. If your website only answers one of these, you lose to someone who covers all six.

Questions with Local or Temporal Dependency

"Who offers this in my city?" "How quickly can I get an appointment?" "What documents do I need in my state?" These questions are gold for local service providers, tradespeople, and healthcare professionals. Whoever answers them concretely gets recommended.

Collecting Customer Questions from Conversations and Emails

You don't need a survey. You already have the answers. They're in your email archive, in conversation notes, in contact forms, in reviews, and in the questions your staff answer every day.

Here's how it works:

1. Search support and sales emails for recurring phrases.
2. Evaluate conversation notes from initial consultations and phone calls.
3. Collect questions from contact forms, reviews, and comments.
4. Ask your staff regularly: "What question did you answer three times this week?"
5. Record questions in five columns: Original question, Customer situation, Objection or uncertainty, Required evidence, Matching website page.

Don't prematurely rewrite original phrases into marketing language. Everyday language shows how customers actually describe their problem. Remove personal data before publication or anonymize it.

An example: A physiotherapist gets asked three times a week: "Does health insurance cover treatment if I don't have a referral?" This isn't an SEO question. This is a real uncertainty that keeps customers from booking appointments. Whoever answers it on their website gets more inquiries.

Turning One Question into a Helpful Website Answer

A good answer has seven elements: direct answer, prerequisites, process, orientation, example, source, and next action. Sounds like a lot. It's not. Look:

Direct answer: "Yes, health insurance covers the costs for physiotherapy with a referral. Without a referral, you pay yourself."

Prerequisites and exceptions: "The referral must come from a general practitioner or specialist. In emergencies, the practice can offer private treatment that you can submit to insurance afterwards—but the insurance decides that, not us."

Process: "You bring the referral to the first appointment. We bill directly with insurance. You only pay the statutory co-payment of €10 plus 10% of treatment costs, maximum €10 per prescription."

Orientation: "A prescription usually includes six treatments. The co-payment is between €10 and €20, depending on the type of treatment."

Example: "A patient came with back pain and a referral for six treatments. He paid a total of €18 co-payment. Insurance covered the rest."

Source: "The basis is § 32 SGB V. You'll find details in your health insurance's remedy catalog."

Next action: "Book an appointment online or call 089/123456."

That's 150 words. Five minutes of writing time. But it doesn't just answer "What is this?", but also: "Is this suitable for me?", "What does it cost?", "How does it work?", "What are the alternatives?" and "How do I recognize quality?"

That's exactly what an AI answer needs to recommend you. Learn more in our article GEO/LLM Visibility.

Adding Evidence, Examples, and Local Information

An answer without evidence is a claim. AI systems prefer verifiable content. This doesn't mean you need to footnote every sentence. But it means: Name specific service areas instead of general advertising statements.

Name qualifications, certifications, and memberships with verifiable details. "Certified energy consultant (BAFA-approved, registration number 12345)" beats "many years of experience".

Describe examples with initial situation, approach, and result. Don't promise results that aren't demonstrable. "We reduced heating costs by 30%" is only credible if you explain how you measured it.

Keep locations, service area, opening hours, and appointment process current. Provide prices with validity date and conditions. Explicitly explain regional differences. For legal, medical, or financial topics, cite professional sources and responsibilities.

Avoid statements like "leading", "best quality", or "guaranteed" if they can't be objectively proven. They harm more than they help.

Testing the Answer with ChatGPT and Perplexity

You've revised the page. Now check if AI systems understand it. Use multiple variants for each important customer question:

• Short search query: "Physiotherapy Munich costs"
• Complete everyday description: "I have back pain and a referral from my GP. What will treatment cost me at a physiotherapist in Munich?"
• Question with location: "Which physiotherapy practice in Munich-Schwabing takes insurance patients?"
• Question with budget: "I can spend maximum €50 per month on physiotherapy. Is that enough?"
• Comparison question: "What's better: physiotherapy or osteopathy for back pain?"

Check for each variant:

• Is your business mentioned?
• Is the correct service described?
• Are location and target group correctly represented?
• Are prices, services, or opening hours misrepresented?
• What sources does the system cite?
• Are important providers or alternatives missing?

Don't interpret answers as proof of actual recommendation. AI outputs can be incomplete, outdated, or incorrect. Document tests: date, platform, model or interface, exact input, and result.

Compare results with your website, your Google Business Profile, and official company data. Don't try to force a desired mention through repeated prompts. What matters is whether the website provides factually appropriate and evidenced information.

Find more about optimizing for AI search in our guide Content Marketing, SEO and GEO.

A Simple 60-Minute Plan for Your Most Important Questions

You don't need new special software. Spreadsheet, existing website, email archive, and manual tests are enough. Here's how the plan works:

0–10 minutes: Choose three important services. Not the exotic ones. The ones customers ask you about most often.

10–20 minutes: Collect customer questions per service from emails, conversations, and reviews. Write down the original phrasing, not your marketing version.

20–30 minutes: Prioritize the most frequent questions by purchase proximity, risk, and local relevance. The question "What does it cost?" is more important than "What's the difference between X and Y?"

30–45 minutes: Revise an existing page: add direct answer, process, price orientation, example, and evidence. Use the structure from the section "Turning One Question into a Helpful Website Answer".

45–52 minutes: Check contact details, location, opening hours, and currency. Sounds trivial. But it's the most common reason AI systems output incorrect information.

52–60 minutes: Test the question in ChatGPT and Perplexity. Document errors in the answer. If your business isn't mentioned, that's not a drama. If incorrect information is mentioned, you must act.

That's it. 60 minutes. No agency. No tool subscription. Just you, your customer questions, and your website.

What I'd Do Differently: Common Mistakes and How to Avoid Them

I know this: You read an article, nod at every paragraph, and then... nothing happens. Because you don't know where to start. Or because you're afraid of doing something wrong. So here are the mistakes I see again and again—and how to avoid them.

Mistake 1: Confusing AI visibility with a guaranteed placement. Nobody can promise you that ChatGPT will recommend you. Anyone who claims that is lying.

Mistake 2: Publishing only general blog articles instead of specific customer questions. "The future of physiotherapy" doesn't bring inquiries. "What does physiotherapy cost in Munich with insurance referral?" does.

Mistake 3: Including search terms in headings but leaving price, process, and suitability questions unanswered. AI systems prefer complete answers, not keyword stuffing.

Mistake 4: Leaving outdated opening hours, locations, or service information. If your website says "Monday to Friday 9–6 pm" but you've been open Saturdays for a year, ChatGPT won't know.

Mistake 5: Publishing AI-generated texts without professional review. ChatGPT can write. But it can't verify if the statement is correct. You must do that. More on this in our article Checking AI Content: Google's New Duty Before Publishing.

Mistake 6: Citing sources that don't support the specific statement. "According to studies..." isn't a source. Name the study, date, and link.

Mistake 7: Anonymizing case studies while simultaneously claiming unverifiable success numbers. Either you prove the result, or you leave it out.

Mistake 8: Only testing ChatGPT and ignoring other search interfaces. Perplexity, Google AI Overviews, and Bing Chat deliver different results. Test all.

Mistake 9: Interpreting a single snapshot as permanent visibility. Just because ChatGPT mentions you today doesn't mean it will tomorrow.

Mistake 10: Practicing "GEO voodoo": artificial prompt patterns, excessive repetition, hidden text, or supposed secret formulas instead of better customer information. That doesn't work. And if it works, not for long.

What You Can Do Today: The First Step

You don't have to know everything. You just have to start. Today. Not tomorrow. Not next week. Today.

Open your email archive. Search for the word "question". Read the last ten emails. Write down three questions that keep coming up. That takes five minutes.

Then open your website. Find the page that should answer this question. Read it. Honestly. As if you were the customer. Does it answer the question? Completely? With price, process, example, and evidence?

If not, write an answer. A direct one. No marketing language. No platitudes. Simply: "Yes, that costs X. Here's how it works. This is what you need. Here's an example."

That's the first step. Not perfect. Not complete. But better than yesterday. And better than 90% of your competition.

Offline you're doing fine. Online hardly anyone sees you. That's not your fault. That was because you didn't know how. Now you do.

Frequently Asked Questions

Do I need to write a separate GEO text for AI search?

No. Start with real customer questions and answer them clearly, completely, and verifiably. There's no secret special technique that guarantees a mention. What works: concrete answers to concrete questions, with price, process, example, and evidence.

How do I find out what customers actually ask ChatGPT?

Collect original phrases from emails, conversations, phone notes, reviews, and contact forms. Supplement with manual tests using realistic situations. You don't need a survey. The answers are already in your inbox.

Is my existing SEO page sufficient for AI search?

That can't be answered generally. Check if it answers the specific situation, suitability, costs, process, evidence, and local conditions. If it only covers the keyword but not the question behind it, it's not sufficient.

How often should I test ChatGPT and Perplexity?

For a pragmatic process, a documented test at publication and with significant changes is sufficient. Also test when prices, locations, or services change. There's no fixed interval. What matters is that you document tests and compare them with your official data.

Can I force a mention in ChatGPT or Perplexity?

No. You can improve content and provide information, but can't guarantee a specific answer. Anyone who promises you that is lying. What you can do: answer real questions, provide evidence, and keep your data current. That increases the chance, but it's no guarantee.

What content helps local self-employed people most?

Precise information about location, service area, appointment process, services, prices, qualifications, and suitable customer situations. The more specific, the better. "We're in Munich" is weak. "We're in Munich-Schwabing, Leopoldstraße 123, and serve customers within a 20 km radius" is strong.

Do I need a paid monitoring tool?

Not for the described start. A spreadsheet and manual, documented tests suffice. The need for additional monitoring depends on scope and update frequency. If you offer three services, a spreadsheet is enough. If you have 30 locations and 100 services, you need a tool.

How do I recognize if AI search actually brings customers to my business?

Link documented AI tests with qualified inquiries, appointment sources, and customer questions. Ask new customers: "How did you find us?" A mere mention without inquiry or revenue isn't sufficient proof of success. Measure what counts: inquiries, appointments, revenue.

The Step You've Been Postponing for Years

You know you need to become more visible online. You know AI search is becoming more important. You know your competition is already doing it. But you postpone it. Because you don't know where to start. Because you're afraid of doing something wrong. Because you think you first need an agency, a tool, a course.

You don't. You just need 60 minutes. And the willingness to be honest. With yourself. With your customers. With your website.

The trick is: You don't have to do everything at once. You just have to answer one question. A single one. The one you're asked most often. The one that keeps your customers from booking. The one you've explained a hundred times but never written down.

Write it down. Today. Now. And then test it in ChatGPT. Not to see if you're recommended. But to see if the answer is correct. If the information is current. If a customer can do something with it.

That's the step. Not perfect. Not complete. But better than yesterday. And better than nothing.

If you don't want to do it alone or want to be faster: Nelija helps you automate exactly this process. Collect customer questions, build answers, test and keep them current. Without having to struggle with technology. Try Nelija for free. No pressure. No hidden costs. Just a tool that has your back.