BLOG | AEO and GEO with Personas

Answer Engine Optimization Using Data-Driven Personas

How Understanding Your Target Audience Improves Your Visibility in ChatGPT and Other AI Systems.

Table of contents

14 minutes reading time

Your brand appears in an AI response. That’s just an observation for now. Whether this can lead to a relevant connection depends on the question in which it appears, the needs of the person asking, and how the system categorizes your offering. A provider may be present in responses to general industry questions but absent when it comes to the specific selection criteria of its most important customers.

This is exactly where “ data-driven personas ” come into play. They describe the goals people pursue, the concerns that delay decisions, and the information they need. For Answer Engine Optimization (AEO), this provides a foundation for analyzing relevant questions and developing content in a targeted manner. Our approach at the Persona Institute combines target audience research with hands-on work on content and customer journeys.

Key Points at a Glance

AEO and GEO (“Generative Engine Optimization”) focus on visibility in direct and generated results, respectively. Technical SEO and helpful content remain relevant in this context.
data-driven personas help derive questions from decision-making situations. The prompts developed in this way are well-founded test scenarios, not automatically identified user queries.
The analysis becomes more meaningful through a breakdown by persona, decision-making phase, and AI system, as well as through repeated measurements.
Brand mentions, recommendations, source citations, and business results are different metrics. They should be analyzed separately.
The benefit lies in making specific decisions about what information is missing for which target audience and what evidence builds trust.

What AEO and GEO Have in Common with SEO

Answer Engine Optimization refers to the process of ensuring that content can be found in direct answers and used effectively. Generative Engine Optimization focuses on the answers provided by generative AI systems. In marketing, the two terms are often used interchangeably. There is no universally accepted definition that clearly distinguishes between them.

Term	Focus
SEO	Visibility in search engines and high-quality traffic from search results
AEO	Identify content for direct answers and examine how they are used
GEO	Examine and improve the visibility and presentation of content and brands in generated responses

For businesses, the practical question is crucial: For which queries is their own offering taken into account, what sources does the response draw upon, and is the presentation factually accurate? Google continues to classify the optimization of its generative search features as SEO. The existing quality and search principles remain the foundation. ^[1]

Research also calls for careful consideration. The seminal GEO study by Aggarwal and colleagues examines how changes to content affect its visibility in generated responses. The results vary depending on factors such as subject area and the specific measure taken. Such experiments provide insights; however, their findings cannot be generalized as a blanket guarantee of success for every brand and every current system—especially since these are evolving rapidly. ^[2]

From Search Term to Decision-Making Situation

“CRM for Small and Medium-Sized Businesses” describes a topic. The question “We are implementing a CRM system for a machinery manufacturer with 120 employees, but we don’t have our own IT department. What solutions can be operated with minimal support?” also describes a challenge and its limitations. This gives rise to different information needs: the implementation effort, training, ongoing support, and costs must be clearly understood.

Not every AI query is that detailed. People also ask short questions and only add details as the conversation progresses. For your analysis, you therefore need both brief introductory questions and well-reasoned contextual variations and follow-up questions. Keyword data remains useful, but it only partially reflects these conversation flows.

Added to this is the way the systems work. Google describes a possible “query fan-out” for AI Overviews and AI Mode: A query is branched out into several related search queries. ^[3] For content planning, this implies that subtopics can also be relevant. In the context of CRM, these might include data migration, rights management, or support. A single, general product description can hardly adequately cover these diverse tasks.

Why " data-driven personas " Are Relevant to AEO

Decision-making logic makes questions easy to understand

A " data-driven persona " distills empirically observable patterns of a target audience. When working with AI-powered search, we are particularly interested in purchase triggers, desired outcomes, information-seeking behavior, decision-making criteria, and barriers. At the Persona Institute, we therefore view personas as a working foundation for concrete communication decisions. ^[4]

Age or job title alone rarely explain what kind of answer someone needs. It is more informative, for example, to know that a decision-maker must justify new software internally, wants to avoid a previous implementation error, and has limited personnel resources. This raises questions about effort, documentation, and risks. The data must show which of these are actually important.

Different requirements apply within the buying center

In B2B, the Ideal Customer Profile helps identify which companies are a good fit for the offering. Personas describe the perspectives of the people involved. ^[4] Business units, procurement, and IT may all evaluate the same solution while asking different questions: Does it solve our problem? Are the total costs acceptable? Does it integrate with our systems?

A single visibility metric for the entire company can mask such differences. Therefore, check whether your content also addresses the needs of those who can approve or block a decision. The job title serves as a starting point for this analysis, not an automatic explanation for specific motives.

Tailoring language to the target audience helps with topic planning

Interviews and documented customer questions show how people describe a problem. “We’re losing too many customers after the first year” leads to a different starting point than “We need a churn model.” Both questions may be related, but they require different levels of prior knowledge. Content should take this level of knowledge into account and explain technical terms when the target audience needs them.

What data set enables reliable inferences

The starting point is a clearly defined target audience. In-house surveys and appropriate external data sets can reveal how widespread certain needs and attitudes are. Qualitative interviews shed light on motivations and language. CRM, support, and usage data round out the picture by providing documented contacts and behavior. The key factor is whether the respective source can actually answer the specific question. ^[5]

At the Persona Institute, we combine existing company data with additional target audience data and translate the insights into persona profiles and recommendations for action. ^[6] For AEO, it must remain clear which finding a question was derived from. A large database alone does not prove that a narrowly defined subgroup is adequately represented. Check the reference group, sample size, survey date, and the relevance of the characteristics surveyed.

DIN SPEC 33462 on data-driven personas provides a methodological reference point. The Persona Institute spearheaded this initiative. The specification describes a systematic approach to creating and using data-driven personas. It serves as a framework for persona work; however, it does not guarantee visibility in AI responses. ^[7]^[8]

When adapting test questions, we recommend distinguishing between three types of statements:

Observed: A specific question came up in an interview or a documented customer interaction. This confirms that it exists, but not yet how common it is in the overall market.
Derived: A documented need suggests a question. The wording is a test scenario whose relevance you can further evaluate.
Unverified: An idea originated from an internal brainstorming session or was supplemented by AI. It remains a hypothesis until supporting evidence is available.

Caution is also needed when combining characteristics. If a survey frequently reveals cost consciousness and safety concerns, this does not necessarily prove that the same individuals associate the two to that extent. For such combinations, you need appropriate analyses or supplementary qualitative evidence. Otherwise, you end up with an artificial persona for whom there is no reliable basis for decision-making.

Five Steps to Developing an AEO Strategy Using Personas

Step 1: Translate relevant findings into test questions

Start with a specific business objective: Do you want to be among the first options considered, be included in provider comparisons, or identify misconceptions about your services? Then select the relevant personas and decision-making scenarios.

For each test question, document the underlying finding. The following table provides an example of how a finding can be transformed into a verifiable content requirement. The findings are provided for illustrative purposes only and do not come from a study evaluated here.

Findings from the Persona	Test question derived from this	Content Requirement to Be Reviewed
Not much time for the introduction	How involved is the process of switching from Excel to a CRM?	Steps, Responsibilities, and Realistic Requirements
Concerns About Ongoing Additional Costs	What costs are incurred after implementing a CRM system?	Licensing Model, Support, and Possible Additional Services
Need for internal safeguards	What documentation do I need to get approval from IT?	Technical Documentation and Verifiable Safety Information
Limited experience with this category	How can I tell if we even need a CRM?	Decision Criteria and Limitations of the Solution

Use only the context that the findings support. A prompt that already lists all the unique features of your offering can skew the results in your favor. In that case, it’s hardly suitable for testing whether the brand also appears when the need is described in open-ended terms.

Step 2: Organize questions along the journey

A person may first want to understand a problem and then evaluate a provider later on. Therefore, organize the prompt inventory according to decision-making phases. These phases help with planning; actual research paths may jump back and forth.

Phase	Example of a question	Purpose of the Content
Understanding the Problem	Why does our team hardly ever use the existing personas?	Explain the causes and prerequisites
Explore possible solutions	How can personas be validated using data?	Explain the methods and their limitations
Compare Alternatives	When are interviews helpful, and when are quantitative surveys?	Show Differences and Selection Criteria
Select a provider	Who can help us with data-driven B2B personas?	Provide evidence of the proposal, the process, and relevant references
Back Up Your Decision	How transparent is the data from Provider X?	Ensure Transparency Regarding Data Sources, Methodology, and Limitations

Distinguish between queries that do not mention brand names and those that do. The former assess discoverability within the appropriate context of user intent. The latter assess what a system says about a brand that is already known. Both are useful, but they address different research questions.

A possible pilot design includes three personas, five phases, and six phrasings per combination—for a total of 90 different prompts. With two systems and three iterations, this results in 540 responses. This is a sample calculation for planning purposes, not a quality standard. More important than a fixed number is ensuring that the questions relevant to decision-making are covered and that similar variations are not inadvertently overrepresented.

Step 3: Take repeatable measurements and analyze the results separately

For each run, record at least the prompt, date, platform, available model names, search mode, language, region, and personalization settings. Save the complete response and the sources cited. Starting a new chat and disabling personalization reduce the influence of previous conversations; however, they do not create a completely neutral or unchanging test environment.

Conduct repeat tests using the same protocol. Test individual questions in separate conversations. When analyzing conversation histories, keep the order and follow-up questions consistent and evaluate them separately. Measurements taken via an API should not be equated with results from the public user interface without verification.

Define in advance what counts as a mention, recommendation, or quote. The following metrics, for example, are suitable for a transparent evaluation:

Key figure	Calculation using the specified test set	Statement
Mention rate	Answers with a rating divided by the total number of valid answers	How common the brand is
Citation Rate	Answers with a source link to the user's own domain, divided by the total number of evaluable answers	How often does one's own content appear explicitly as a source?
Recommendation Rate	Responses with a positive recommendation, divided by the total number of responses to multiple-choice questions	How often the offer is taken into account during the selection process
Fact-Error Rate	Responses containing at least one verifiable error regarding the brand, divided by the total number of responses containing brand statements	How reliable is the brand's portrayal?

When reviewing source links, also verify whether the linked page actually supports the corresponding statement. A mention may also appear in a warning or as an unsuitable alternative. Similarly, citing a source does not necessarily imply a recommendation. Terms such as “Citation Share” may also use different denominators than the citation rate defined here. Document your calculations before comparing tools or time periods.

For Google queries without AI Overview, the display rate should also remain visible: Otherwise, a high rate among a small number of AI responses could mask a low actual presence. Report results by system, persona, and phase before calculating an overall figure. If you apply weighting, justify the weights based on target audience shares or business priorities.

The metric describes your test set, not your brand’s market share across all AI conversations. Repeated tests demonstrate the stability of the responses but do not replace a representative survey of user behavior. Therefore, supplement the tests with available platform data. For example, Microsoft provides reports on citations, topics, and query intent in Bing Webmaster Tools. These reports also pertain to the specific services tracked. ^[9]

Step 4: Improve content based on information needs

Derive a specific editorial task from each relevant gap. If the persona needs to justify an investment internally, clear cost breakdowns, a detailed implementation plan, and verifiable references will be helpful. Another general guide to the product category will only provide limited answers to this task.

State the key answer early on, and then explain the assumptions and limitations. Use headings that help readers find the information they’re looking for. Numbers need a source, a point of reference, and a time frame. Your own studies become more meaningful when the target audience, methodology, and number of cases are clearly identified. In case studies, the initial situation, the action taken, and the result should remain distinct.

Technical accessibility remains a prerequisite. For linked sources in Google's AI search features, pages must be indexed and suitable for displaying a snippet. ^[3] Also, check whether internal links lead to important content and whether any conflicting product or company information needs to be corrected.

Google's recommendations do not specify a required answer length of 40 to 60 words or a specific AI markup. Even structured data does not guarantee that it will be taken into account. ^[1] Use them in a way that aligns with the visible content and the specific search function. An FAQ section is useful when it answers actual unanswered questions. Creating a separate page for each prompt variation is not a compelling content strategy.

Step 5: Consider External Sources and Business Impact

Analyze which external sources are cited in the responses and what they say about your company. These may include professional articles, directories, comparative offers, or customer reviews. From this analysis, you can identify specific actions to take: have outdated information corrected, publicly explain your methodology, or contribute to a professional debate with a well-documented study.

However, a mention on a frequently cited platform does not guarantee a subsequent recommendation. Prioritize sources that are appropriate in terms of subject matter and are actually relevant to your target audience. Also, check whether the AI query actually makes sense when directed at a specific provider or a neutral explanation. Not every answer requires a brand name.

Finally, link the visibility analysis to qualified inquiries, applications, or sales. Sessions with a recognizable AI referrer reveal only part of the potential contribution. A website visited directly afterward often does not reveal the previous AI contact. Surveys regarding the origin of a contact can provide additional insights. Furthermore, changes following a redesign are not definitive proof of effectiveness: models, sources, and the competition also change. An unchanged comparison set helps to better contextualize such influences.

Case Study from a Buying Center

A fictional provider of analytics software examines the perspectives of marketing and IT. The following example illustrates the analysis; it is not a customer case study and does not include any measured performance metrics.

For the marketing manager, the main question is how to consolidate campaign results without having to create time-consuming manual reports. For the IT manager, the focus is on what data will be processed, what access rights will be granted, and how the system can be integrated. In a real-world project, these requirements would first need to be evaluated through research.

Suppose the tests reveal frequent mentions in response to reporting questions and few in response to questions regarding technical approval. In that case, a general increase in the amount of text would be of little use. It would be necessary to determine whether comprehensible integration documentation is missing, whether existing evidence is publicly available, or whether the proposal actually fails to meet the required criteria.

This distinction is crucial: A visibility gap may indicate an information gap, an accessibility issue, or a lack of product suitability. The analysis must determine the underlying cause. Content can demonstrate existing suitability, but it cannot compensate for a lack of performance.

How Candidate Personas Help in Recruiting

The same approach can be applied to employer-related questions. An employee working shifts may need specific information to help them plan their work schedule. A skilled worker open to a career change might be evaluating career development opportunities, leadership style, or the actual proportion of remote work. Candidate personas help identify such questions based on the motivations and barriers that have been identified.

Review both open questions regarding potential employers and questions about your own company. In the latter case, it is particularly important to ensure that the description is accurate and specific to the location. General cultural promises cannot replace concrete information about work schedules, onboarding, or professional development. Similarly, promised working conditions must actually apply, and any differences between roles or locations must be explained accordingly.

Five Mistakes to Avoid When Working with Personas and AEO

Treat generated questions as actual demand. Linguistic plausibility does not indicate how common a problem is or whether a target audience uses AI for that purpose.
Include the desired provider in the question. Product features that are too specific can skew an open visibility check.
Equate mention with recommendation. The context in which the brand appears and whether the statements are accurate are also crucial.
Evaluate overall figures without breaking them down. Differences between personas and decision-making phases may be masked by the average.
Treat every omission as part of the content assignment. First, assess the information needs, accessibility, and actual suitability of the offering.

Start the implementation process using existing personas

Start with a key persona and a clear decision-making scenario. Compile the questions and objections that have been raised, use them to develop a manageable test set, and document the answers. The first outcome should be a well-reasoned priority: What missing information would most help this target group in making their decision?

The Persona Agent from the Persona Institute can apply existing persona knowledge when reviewing texts, landing pages, and messages. ^[10] For AEO, this brainstorming process can be used to identify unclear answers and unaddressed objections. The feedback generated is derived from the knowledge base; it does not result in any additional market research data.

Would you like to use your persona data for this purpose? Talk to us about your target audiences and the decisions where your content is intended to provide guidance. Together, we can determine which insights are already reliable and where additional research would be helpful.

Frequently Asked Questions About AEO with Personas

Is AEO replacing traditional search engine optimization?

No. Search engine accessibility and relevant content remain important. AEO expands this analysis to include whether content and brands appear in search results and how they are presented there.

Can you derive real prompts from a persona?

Only if such questions were actually asked. Otherwise, plausible test scenarios can be derived from documented needs. These should be compared with real-world questions and available usage data.

How many prompts do I need per persona?

There is no universal figure for this. The scope depends on the decisions being examined, the subgroups, and the desired coverage. A small, well-justified set is more useful for getting started than many variants that are barely distinguishable from one another.

What is a good visibility rate?

A blanket threshold is not appropriate. The ratio depends on the test set, the system, the phase, and the calculation. Evaluate comparable measurements over time and verify that the relevant information is being conveyed.

Can Persona Agents handle the research?

You can organize existing findings, propose questions, and evaluate content from a documented perspective. The suitability of the source data and the relevance of new hypotheses must still be reviewed by experts.