The right AEO questions do not come from keyword research tools alone, and they should not come from AI alone either. Start with what customers actually ask, review the queries already generating impressions and clicks in Google Search Console, identify gaps in your existing content, and use AI systems for testing and ideation — not as proof of real demand. Keyword research and question research work best together.
A common mistake is to start with a list of questions generated by a tool and assume they are genuinely your customers' questions. They may be useful ideas, but they still need validation.
AEO does not replace keyword research. It adds another layer: you look not only at keywords and search volume, but also at the underlying need behind the search.
Questions and keywords
You need both.
The phrase:
"accountant Thessaloniki"
gives you a basic topic and location. A more detailed question can reveal more about the underlying need:
"I have a small e-shop and I'm looking for an accountant in Thessaloniki who knows e-commerce. What should I ask before I choose?"
The second phrasing is not "better" because it is longer. It is useful because it reveals more elements: type of business, specific need, location and uncertainty before the choice.
For a question like this, public information needs to exist that can be verified: whether the accounting firm states that it serves online stores, which relevant services it offers, in which area it operates and how someone can contact it. You should not rely on an AI system correctly inferring e-commerce expertise from the generic label "accounting firm" or knowing services that are not stated publicly.
For Google, core SEO best practices still apply across generative AI features in Search. You do not need a separate page for every possible phrasing of the same underlying need. The content must substantially cover the topic and be useful for the person reading it.
The general context is available in what AEO is.
Where do you find the real questions?
Four sources are especially useful:
-
Conversations with customers. What do they ask before they buy? What don't they understand? What makes them hesitate? Questions that recur in phone calls, meetings and messages are a strong indication of real need.
-
Sales, support and internal search. Recurring topics in emails, support requests, forms and on-site searches can reveal information gaps. The analysis should be done in aggregate and without transferring personal or sensitive data to tools that don't need it.
-
Search data. Google Search Console shows real queries for which your site has already appeared in Google. It does not include every query: some are hidden for privacy and there are limits on the rows shown. Even so, it provides a far more reliable basis than a list of questions generated by an AI tool alone.
-
Public content and AI testing. Review what your own website and competitors already cover around important customer needs. Then, you can test representative questions on ChatGPT, Gemini, Perplexity or other systems to see what kind of answer they return, which public information appears and, when sources are provided, which sources are linked to the answer. This is useful as a snapshot of the current landscape, not as a measure of how many users ask the question.
The first three sources help you find and confirm needs. The fourth mainly helps you see how these needs are addressed publicly today.
How do you tell which questions are worth covering?
Not every question needs to become an article.
Look mainly at four things:
- Relevance to the business: does it really concern a service, product or knowledge you have?
- Proof of need: does it appear in conversations, support, Search Console or another real source?
- Usefulness: can the answer reduce a substantial question or uncertainty?
- Ability to answer well: do you have real experience, evidence or specific information that adds something beyond a generic answer?
Frequency is useful, but it is not the only factor. A question that appears less frequently can have great value if it concerns an important decision, high cost, technical difficulty or something the customer struggles to confirm.
Once you've selected the questions, organize them into thematic groups. In the hub-and-spoke model we explain how related topics connect, while in the content calendar for AEO how they enter a realistic production order.
What do you do once you find them?
Do not mechanically create a new page for every different phrasing.
If many questions concern the same basic need, it usually makes more sense to cover them on the same comprehensive page or article. A separate page is needed when there is a genuinely different topic or a different user intent.
Questions can also be used as sections within an article or as FAQs when that genuinely helps the reader. The FAQ doesn't need to be written with the goal of "getting picked up by the AI," but to give short and accurate answers to real questions.
The question list should be reviewed as services, customers and search data change. If you'd like us to look at your own case, we start from the questions that already emerge from conversations with your customers and your business data.
No method of selecting questions guarantees a specific ranking, appearance, mention or recommendation in Google, ChatGPT, Gemini, Perplexity or any other AI system.