Case Study: AEO & GEO Multi-Agent Assistant to Recover Visibility Lost to "Zero-click"
This case study describes my comprehensive UX design process for a conversational multi-agent assistant, a high-fidelity prototype ("AI Spike") that orchestrates a team of specialized AI agents (in AEO and GEO) to perform complete website evaluations and identify improvement opportunities to increase their visibility in Generative Engines like ChatGPT, Claude, Gemini, or Perplexity, and Answer Engines like Google SGE (Search Generative Experience), Bing AI, You.com, etc.
I have developed this AEO & GEO multi-agent assistant following a rigorous implementation plan that merges the Conversation Design Institute (CDI) methodology with the technical architecture of Conversational Agents (Google Dialogflow CX). This process has ensured that every design decision is backed by a solid and scalable technical execution.
In-depth debate between AI presenters on this case study: AEO & GEO Multi-Agent Assistant
01
The Challenge: Surviving in the "Age of AI Search"
1.1 The Transformation of the Digital Ecosystem
"Tools like OpenAI's ChatGPT and Anthropic's Claude are 'devouring' all internet content to train their models, but instead of directing users to the original sources, they often summarize and offer answers directly on their own interfaces." Source: —Digital Transformation Analysis
Estimates of Organic Traffic Loss
Recent empirical evidence shows a catastrophic reduction in organic traffic to websites, directly attributed to the adoption of AI-powered search features.
Source: eMarketer Report | TUYA Digital Analysis
2025 Bad Bot Report: Bots Surpass Human Traffic
of global traffic in 2024 was generated by bots, surpassing human traffic for the first time
Source: 2025 Bad Bot Report
Global Traffic Distribution 2024
Source: 2025 Bad Bot Report
The Explosion of AI Content Scraping
The data published by Cloudflare CEO Matthew Prince is so drastic it seems like science fiction. In just 6 months, the ratio of pages scraped by Google to actual visits has tripled – from 6:1 to 18:1. This means that today, Google downloads content from 18 pages to send just one visitor.
AI platforms fare even worse. OpenAI increased its scrape-to-visit ratio from 250:1 to an astronomical 1500:1, and Anthropic – from an already high 6000:1 to an almost incomprehensible 60000:1. In other words, AI downloads thousands of pieces of content to ensure that virtually no one is sent to the original sites.
Source: Matthew Prince, CEO of Cloudflare (AIBusiness)
OpenAI
Anthropic
Implications for the future of digital content
This explosion in content scraping represents an existential threat to the traffic and advertising-based internet model. AI platforms are extracting value on an unprecedented scale, while content creators receive less and less in return.
Sustainability Crisis: The current content creation model becomes unsustainable when the return is virtually nil.
Need for Regulation: Requires legal frameworks that compel AI platforms to fairly compensate creators.
Technological Evolution: Creators must adapt to an ecosystem where optimizing for AI engines is more important than optimizing for traditional search engines.
1.2 Pain Points
My research identified 5 key pain points that drive users to abandon the traditional web for AI interfaces.
Inefficient Navigation
Users waste valuable time navigating between multiple tabs, scattered tools, and sites with complex architectures. A complete SEO analysis can require up to 15 different tools, fragmenting the workflow and reducing productivity.
Time Loss: 45-60 minutes per basic analysis
2Cognitive Overload
Cognitive Overload
Interfaces saturated with ads, pop-ups, and distracting elements cause mental fatigue. Users face "choice overload" with hundreds of metrics without context, hindering effective decision-making and creating analysis paralysis.
Problem: 200+ metrics without clear prioritization
3Fragmented Information
Fragmented Information
Users must manually synthesize data from multiple sources that often contradict each other. Google Analytics says one thing, Search Console another, and third-party tools offer different perspectives, creating confusion instead of clarity.
Challenge: Manually synthesizing contradictory sources
4Language Barriers
Language Barriers
Technical SEO content is predominantly in English, creating a significant barrier for Spanish-speaking professionals. This limits access to up-to-date information and reduces the speed of implementing best practices.
Limitation: 85% of technical content only in English
5Static Interaction
Static Interaction
Traditional tools offer static searches without context or session memory. Each query is independent, forcing the user to repeat information and losing the thread of complex analyses that require multiple iterations.
Problem: No context or memory between queries
Sources for the "Pain Points" Research
This research rigorously analyzes the migration of users from conventional websites to tools like ChatGPT, identifying the structural causes of digital dissatisfaction (friction, overload, fragmentation, language barriers, and static interaction). Based on meta-analyses of specialized sources, industry reports, academic studies, and quantitative analyses, the research validates a systemic paradigm shift that demands new digital ecosystems centered on conversation and automated synthesis.
List of Sources with Title and URL
1. Impact of Generative AI on Search Traffic and Content Visibility
https://beanstalkim.com/learn/seo/impact-of-generative-ai-on-search-traffic-and-content-visibility/2. Analyzing ChatGPT adoption drivers with the TOEK framework - PMC
https://pmc.ncbi.nlm.nih.gov/articles/PMC10730566/3. How will generative AI impact website rankings and traffic?
https://searchengineland.com/generative-ai-impact-website-rankings-traffic-4436244. How Is AI Decreasing Your Website Traffic? - SuperWebPros
https://superwebpros.com/how-is-ai-decreasing-your-website-traffic/5. From User-Driven Web to Agentic AI: Paradigm Displacement in ...
https://www.linkedin.com/pulse/from-user-driven-web-agentic-ai-paradigm-displacement-maginley-gxmae6. Generative AI drives web traffic spike for retailers in US - CX Network
https://www.cxnetwork.com/artificial-intelligence/news/generative-ai-drives-web-traffic-spike-for-retailers-in-us7. AI Bots Now Dominate Web Traffic, Surpassing Human Users
https://www.aicerts.ai/news/ai-bots-now-dominate-web-traffic-surpassing-human-users/8. Adoption and impacts of generative artificial intelligence: Theoretical
https://www.sciencedirect.com/science/article/pii/S26670968240002119. This is what AI is doing to your website traffic | TechFinitive
https://www.techfinitive.com/this-is-what-ai-is-doing-to-your-website-traffic/10. Why Your Website Is Losing Traffic (And How to Fix It with AI-Driven Insights)
https://www.expre.co.uk/why-your-website-is-losing-traffic-and-how-to-fix-it-with-ai-driven-insights/11. Is Spanish SEO Still Relevant? - Hispanic Market Solutions
https://www.hispanicmarketadvisors.com/blog/is-spanish-seo-still-relevant/12. Generative and Malleable User Interfaces with Generative ... - arXiv
https://arxiv.org/html/2503.04084v113. Chatbot vs Conversational AI: Key Differences, Use Cases & How to ...
https://smartdev.com/conversational-ai-vs-chatbot-unleashing-the-secret-powers-of-ai-driven-conversations/14. A Complete Guide to SEO in Spain in 2025
https://blog.applabx.com/a-complete-guide-to-seo-in-spain-in-2025/15. Generative AI Systems Aren't Just Open or Closed Source - WIRED
https://www.wired.com/story/generative-ai-systems-arent-just-open-or-closed-source/16. Conversational AI vs. Traditional Chatbots: What's the Future?
https://helpybo.com/conversational-ai-vs-traditional-chatbots-whats-the-future/17. Adoption of artificial intelligence in Spanish firms: an initial analysis
https://www.bde.es/f/webbe/SES/Secciones/Publicaciones/InformesBoletinesRevistas/BoletinEconomico/25/T2/Files/be2502-art06e.pdf18. Making Generative AI Accessible to Every Business - WIRED
https://www.wired.com/sponsored/story/making-generative-ai-accessible-to-every-business/19. Traditional Chatbot Vs AI Agent - A Detailed Comparison of ...
https://www.sparkouttech.com/chatbot-vs-ai-agent/20. Andy Crestodina's Post - AI is killing website traffic! - LinkedIn
https://www.linkedin.com/posts/andycrestodina_ai-is-killing-website-traffic-or-is-it-activity-7296267426051108865-LNUl21. AI now commonplace in Spanish workplace, languages remain a ...
https://novobrief.com/ai-now-commonplace-in-spanish-workplace-languages-remain-a-hurdle/11428/22. Optimizing user flows: Minimizing friction throughout the journey
https://www.statsig.com/perspectives/optimizing-user-flows-minimizing-friction23. Information Overload: Causes, Impact, and Solutions for the Workplace
https://www.lumapps.com/insights/blog/information-overload24. AI Strategies for Hyper Personalized Customer Engagement
https://www.gocustomer.ai/blog/ai-transformation-of-customer-experience-a-gateway-to-hyper-personalized-engagement25. 2025 Guide to Understand and Minimize User Friction - Survicate
https://survicate.com/blog/user-friction/26. Information overload - A negative factor affecting mental health ...
https://longevity-protocols.com/en/knowledge-base/interventions/negative/information-overload/27. Why AI Chatbots May Overtake Search Engines, What You Need to Know
https://arabot.io/en/blog-post/4328. Efficient data mining for web navigation patterns - ScienceDirect
https://www.sciencedirect.com/science/article/abs/pii/S095058490300109529. Dealing with information overload: a comprehensive review - PMC
https://pmc.ncbi.nlm.nih.gov/articles/PMC10322198/30. AI makes personalization feel more personal | Centific
https://centific.com/blog/ai-makes-personalization-feel-more-personal31. Behavior Analysis in UX: Meaning And Examples
https://octet.design/journal/behaviour-analysis/32. AI Chatbots vs. Search Engines: The Data Behind the Disruption Debate
https://webbiquity.com/ai-in-marketing/ai-chatbots-vs-search-engines-the-data-behind-the-disruption-debate/33. AI Spanish Conversation - Callin.io
https://callin.io/ai-spanish-conversation/34. Memory vs. Context in AI: Why ChatGPT Is Ahead , and Why That Matters
https://dev.to/pvgomes/memory-vs-context-in-ai-why-chatgpt-is-ahead-and-why-that-matters-98o35. AI Chatbots vs Search Engines: 24-Month Study on Traffic Trends
https://onelittleweb.com/data-studies/ai-chatbots-vs-search-engines/36. An experiment in Spanish translation using AI - James Breiner
https://jamesbreiner.com/an-experiment-in-spanish-translation-using-ai/37. Persistent AI Memory: Local Searchable Storage for LLMs & Copilot
https://mcpmarket.com/server/persistent-ai-memory38. AI Chatbots Accounted for Just 2.96% of Search Traffic in 2024 ...
https://www.linkedin.com/pulse/ai-chatbots-accounted-just-296-search-traffic-2024-study-bw6gc39. How AI Handles Cultural Context In Translation - BeLikeNative
https://belikenative.com/how-ai-handles-cultural-context-in-translation/40. AI-Native Memory and the Rise of Context-Aware AI Agents
https://ajithp.com/2025/06/30/ai-native-memory-persistent-agents-second-me/41. AI Chatbots vs. Traditional Search Engines: A Shift in How We Find Information
https://www.sentisight.ai/ai-chatbots-vs-traditional-search-engines-a-shift-in-how-we-find-information/42. AI Agents with Memory: Context Retention Beyond Short Prompts
https://www.linkedin.com/pulse/ai-agents-memory-context-retention-beyond-short-ganesh-jagadeesan-7hcoc43. Agentic Search vs Traditional Search Engines: What's the Real Difference?
https://ninepeaks.io/agentic-search-vs-traditional-search-engines44. New guide helps journalists dispel disinformation targeting US Latinos and Spanish-speaking
https://ijnet.org/en/story/new-guide-helps-journalists-dispel-disinformation-targeting-us-latino-and-spanish-speaking45. Generative AI As a Growth Catalyst - WIRED
https://www.wired.com/sponsored/story/generative-ai-as-a-growth-catalyst/46. The Convergence of AI and Traditional Analytics in Business Platforms
https://www.linkedin.com/pulse/convergence-ai-traditional-analytics-business-platforms-ashish-ganda-lkxjc47. As the 2024 election nears, misinformation targeting Latinos ... - NPR
https://www.npr.org/2024/06/25/nx-s1-5013727/new-research-looks-at-how-political-misinformation-is-targeted-at-latinos48. Notes from the AI Frontier: Insights from Hundreds of Use Cases - McKinsey [PDF]
https://www.mckinsey.com/~/media/mckinsey/featured%20insights/artificial%20intelligence/notes%20from%20the%20ai%20frontier%20applications%20and%20value%20of%20deep%20learning/notes-from-the-ai-frontier-insights-from-hundreds-of-use-cases-discussion-paper.pdf49. State of Spanish-Language Media 2024: Video Edition [PDF]
https://harkerbos.com/wp-content/uploads/2024/08/State-of-Spanish-Language-Media-2024-Video-Edition.pdf50. Synchron's Brain-Computer Interface Now Has Nvidia's AI - WIRED
https://www.wired.com/story/synchrons-brain-computer-interface-now-has-nvidias-ai/51. Artificial Intelligence Insights & Articles | QuantumBlack - McKinsey
https://www.mckinsey.com/capabilities/quantumblack/our-insights52. Hispanic consumers are one of the most digitally active groups in US
https://www.emarketer.com/content/hispanic-consumers-one-of-most-digitally-active-groups-us53. A Growing Share of Latinos Get Their News in English
https://www.pewresearch.org/race-and-ethnicity/2013/07/23/a-growing-share-of-latinos-get-their-news-in-english/54. Why Spanish streaming is a FAST-growing business - Papercup
https://www.papercup.com/blog/Spanish-streaming55. Embrace the UX Paradigm Shift to Agentic Experience Design
https://www.salesforce.com/blog/ux-shift-to-agentic-experience-design/56. The utility-over-usability effect explains why bad UX persists
https://uxdesign.cc/the-utility-over-usability-effect-explains-why-bad-ux-persists-b090cd993afa57. The implementation of chatbot-mediated immediacy for synchronous ...
https://link.springer.com/article/10.1007/s10639-023-11602-158. The Rise of AI as the New UI: A Paradigm Shift in User Experience
https://www.linkedin.com/pulse/rise-ai-new-ui-paradigm-shift-user-experience-omnath-dubey-mfibf59. Consumer Perception of Planned Obsolescence: A Research on ... [PDF]
https://isarder.org/index.php/isarder/article/download/2089/2012/203560. The Influence of Conversational AI on Consumer Behavior and ... [PDF]
https://eujournal.org/index.php/esj/article/view/19171/1889061. [PDF] https://arxiv.org/abs/2506.23116 [cs.HC ]30 June 2025
https://arxiv.org/pdf/2506.23116.pdf62. Social Media Analysis of Perceived Product Obsolescence
https://www.sciencedirect.com/science/article/pii/S2212827116303870/pdf?md5=ab5dd4d0c127e55bb2d39892d31c8da7&pid=1-s2.0-S2212827116303870-main.pdf63. Reevaluating personalization in AI-powered service chatbots
https://www.sciencedirect.com/science/article/pii/S294988212500010664. The Three Paradigms of HCI - Computer Science | Virginia Tech [PDF]
https://people.cs.vt.edu/srh/Downloads/TheThreeParadigmsofHCI.pdf65. The Mutual Benefits of Ethnography and Discourse Analysis in Digital Contexts
https://www.academia.edu/42045864/The_Mutual_Benefits_of_Ethnography_and_Discourse_Analysis_in_Digital_Contexts66. Number of ChatGPT Users (July 2025) - Exploding Topics
https://explodingtopics.com/blog/chatgpt-users67. AI Industry Analysis: 50 Most Visited AI Tools and Their 24B+ Traffic
https://writerbuddy.ai/blog/ai-industry-analysis68. Methodologies for Digital Discourse Analysis [Interactive Article]
https://discourseanalyzer.com/methodologies-for-digital-discourse-analysis/69. ChatGPT global web visits 2022-2024, by device - Statista
https://www.statista.com/statistics/1463637/chat-openai-com-monthly-visits-by-device/70. Potential and limitations of digital ethnographic research - Frontiers
https://www.frontiersin.org/journals/sociology/articles/10.3389/fsoc.2022.1092181/full71. 100+ ChatGPT Statistics 2024 - AIPRM
https://www.aiprm.com/chatgpt-statistics/72. AI and online traffic - statistics & facts - Statista
https://www.statista.com/topics/13648/ai-and-online-traffic/73. a comparative computational discourse analysis | AI & SOCIETY
https://link.springer.com/article/10.1007/s00146-025-02214-z1.3. User Persona: "The Conversational Explorer"
To bring the data to life, I created an archetype that encapsulates the motivations for this massive migration.
María, integrates AI into all her tasks
Says
"It's so annoying when I search for something on Google and have to open ten different tabs, deal with pop-up ads, and then read pages and pages of text to find a simple answer."
Thinks
"With ChatGPT, it's like having an expert available 24/7. I can ask it a question and get a clear, concise answer in seconds, without having to browse through a bunch of websites."
Feels
"It's a feeling of empowerment. I feel like I can learn about any topic without the barriers and frustrations of the traditional web."
Does
"She has integrated AI tools into almost every aspect of her digital life: professional tasks, writing emails, market research, vacation planning, and cultural recommendations."
1.4. The Right Design Question (Applying Occam's Razor)
❌ Incorrect Question (Incremental Approach)
"What tool could I design for website owners and marketing agencies to improve their search engine rankings?"
✅ Correct Question (Design Mission)
"How do I design a tool that allows website owners and marketing agencies to know what content, structure, and metadata they need to adapt so those websites are readable, referenceable, and preferred by answer engines and generative engines like ChatGPT or Perplexity, increasing the chances of appearing as sources in AI results?"
02
My Process: Applying the Conversational Design Framework
2.1 Strategy: Defining Users and Assistants
Before developing any dialogue, it was essential to identify our target audience and design the assistant personas to ensure trust and clarity in the interaction.
2.1.1.1 Exploration of Operational Flows
I have analyzed the current operational flows of digital marketing professionals
Traditional SEO analysis flow that I have identified
Processes I have documented:
- 1
Fragmented manual audit
I have observed that professionals use multiple tools (Screaming Frog, SEMrush, Ahrefs), obtaining scattered and decontextualized data.
- 2
Siloed interpretation
I have confirmed that each tool provides isolated metrics without integrated context for AEO/GEO optimization.
- 3
Reactive implementation
I have verified that optimizations are applied based on historical correlations of traditional SEO, ignoring the behavior of generative engines.
I have mapped critical friction points in traditional SEO analysis processes.
Operational Friction #1
Cognitive overload I have detected
- Problem: I have observed that professionals must simultaneously interpret 15-20 different metrics without clear prioritization.
- Impact: I have confirmed that most time is spent on analysis, and little available time is invested in effective implementation.
- Gap: I have found a total absence of intelligent synthesis that prioritizes actions by impact on AEO/GEO visibility.
Operational Friction #2
Temporal disconnection I have found
- Problem: I have confirmed that historical SEO data does not predict the behavior of ChatGPT, Claude, or Google SGE.
- Impact: I have documented that strategies become obsolete before implementation, losing opportunities to recover "zero-click" visibility.
- Critical Gap: I have identified an urgent need for predictive analysis specifically oriented towards generative engines.
Operational Friction #3
Atomization of decisions I have documented
- Problem: I have observed that optimizations are performed on isolated elements without considering how AI agents process information.
- Impact: I have confirmed that partial improvements do not generate significant traction in rankings on Perplexity, Bing AI, or You.com.
- Gap: I have found a complete lack of holistic orchestration designed for generative engines.
2.1.1.2 Main User Personas
SME Owner
Marta Sánchez (39), Valencia
Demographics and Life Situation
Business owner
Receives web reports by email but doesn't fully understand what's happening
Is concerned about the constant drop in her website traffic
Seeks clear, action-oriented explanations
Doesn't master technical SEO but needs to know what to change
Wants concrete solutions, not vague diagnoses
Agency Director
Ana Martínez (42), Barcelona
Demographics and Life Situation
Runs an AI-driven digital agency
Drop in organic traffic post-AI Overviews
Needs to understand the real business impact
Seeks new performance metrics
Strategies outside the classic SEO model
Adaptation of services to the new reality
Google Ads Expert
Alberto Ortega (40), Zaragoza
Demographics and Life Situation
Specialist in paid digital advertising
Detects that the cost per conversion (CPA) is skyrocketing
The average cost per lead is increasing
Many users get information directly without clicking on any ads
Needs to help strengthen brand authority, Quality Score in Google Ads, and visibility
2.1.2 Bot Personas
Adriana
Directora de agencia (Agente orquestador)
Name: Adriana Consultora (Agency Director)
Age: 42
Gender: Femenino
Personality: Segura, empática, metódica
Geography: Latinoamérica y España
"Con más de 15 años de experiencia en el vertiginoso mundo del SEO, Adriana Consultora es la mente detrás de Lyra Agency, una agencia boutique pionera en estrategias listas para la era de la IA. Nacida en Latinoamérica y con raíces profesionales en España, su trayectoria refleja una fusión única de visión global y entendimiento local de los mercados digitales. Su carrera comenzó como SEO Manager, donde lideró proyectos para multinacionales y startups, perfeccionando su habilidad para diagnosticar problemas complejos de tráfico web y transformar datos en acciones concretas. Hoy, a sus 42 años, dirige Lyra Agency con un enfoque metódico y empático, convirtiéndose en un puente entre los equipos ejecutivos y las soluciones técnicas que impulsan el crecimiento."
Organization: Agencia de marketing digital
Audience: Dueños de negocios, CMOs, equipos ejecutivos
Brand Tone: Consultivo, profesional, accesible, entusiasta
Service Tasks: Diagnóstico de problemas de tráfico web, evaluación de rendimiento, auditorías SEO
Persuasive Tasks: Convencer de la necesidad de adaptarse al nuevo ecosistema IA, justificar inversiones en AEO/GEO
Channels: Reuniones virtuales, dashboards de reporte, interfaz web
Bot Tone: Profesional - Respetuoso - Enfocado en soluciones
Sounds Like: Voz media, tono firme pero acogedor, acento neutro
Chats Like: Usa bullets, gráficos si se permite, pocas emojis, pausas estratégicas
"He analizado tu sitio web y nuestros expertos han identificado oportunidades clave."
"Tu posicionamiento actual podría estar en riesgo si no se adaptan ciertos aspectos."
"Hola, soy Adriana, directora de Lyra Agency. Estoy aquí para ayudarte a navegar la nueva era de búsquedas generativas."
"Entiendo que esta transición puede ser desafiante."
"Correcto. Hemos validado esos datos con nuestras herramientas internas."
"Gracias por tu paciencia. Hubo un retraso mientras verificábamos datos clave."
"Te recomiendo revisar estos puntos antes de nuestra siguiente reunión."
Dario
Experto en GEO (Agente de Generative Engine Optimization)
Eva
Experta en AEO (Agente de Answer Engine Optimization)
2.2 Design: From Idea to Dialogue Validation
My design philosophy for this assistant is based on a fundamental premise: an expert AI system is not one that never encounters problems, but one that handles uncertainty, errors, and deviations with the same intelligence and empathy as its main flow. Below, I detail how I have applied this philosophy in each layer of the conversational design.
2.2.1 Main Flow Design
2.2.2.1 Happy Path
The perfect scenario where a user provides a valid and accessible URL, and the system orchestrates the specialized agents to deliver a complete analysis without any issues.
Step 1: Welcome and Intent
"Analyze https://josegalan.dev/"
Adriana, Orchestrator Agent
Hi! Thanks, I've received the URL: https://josegalan.dev/. I've put my team of experts to work analyzing your website. This will take about 30 seconds.
Step 2: Data Capture and Validation
Adriana, Orchestrator Agent
Starting website analysis...
Context Analyst
Extracting and processing website content...
Context Analyst
I have extracted 913 characters of relevant content.
Step 3: Orchestration and Processing
Agents Eva (AEO expert) and Dario (GEO expert)
Hello, we are starting to generate optimization recommendations...
Agents Eva (AEO expert) and Dario (GEO expert)
We have generated the following recommendations:
Eva, AEO Expert Agent
AEO RECOMMENDATIONS
Answer Engine Optimization
1
Add structured data schema using schema.org vocabulary to mark up the FAQ. This will help search engines understand the content and potentially feature it in rich snippets or answer boxes.
AEO focuses on making your content easily accessible for search engines to directly answer user queries. Structured data markup for FAQs allows search engines to easily identify and extract answers to common questions about conversational AI design.
2
Optimize the FAQ section with relevant keywords related to conversational AI design, such as "conversational AI UX design," "multi-agent conversational systems," "ethical AI design," etc. Use these keywords naturally within the questions and answers.
Directly answering user queries is crucial for AEO. By optimizing the FAQ section with relevant keywords, you increase the chances of your website appearing in top search results.
3
Ensure the answers in the FAQ section are concise and directly address the question. Avoid ambiguity and jargon.
Clear and concise answers are more likely to be featured as direct answers in search results. Jargon and ambiguity can confuse search engines and reduce the chances of your content being featured.
Dario, GEO Expert
GEO RECOMMENDATIONS
Generative Engine Optimization
1
Create comprehensive, high-quality content that covers various aspects of conversational AI UX design. Include detailed explanations, examples, and case studies to demonstrate your expertise.
GEO focuses on building a strong online presence that showcases your expertise. Comprehensive and detailed content helps you establish yourself as an authority in the field, improving your search rankings and building trust.
2
Develop a blog section with articles that address various facets of conversational AI design, ethical considerations, and emerging trends. This will attract more organic traffic and increase your visibility.
Regularly publishing valuable content related to your field helps establish your expertise, build backlinks, and increase your website's authority, all of which are positive signals for search engines in the context of GEO.
3
Generate high-quality backlinks from relevant websites and publications in the AI and UX design space.
Backlinks from reputable sources signal to search engines that your website is a trustworthy source of information, improving your search rankings and overall visibility. This is crucial for both GEO and AEO.
4
Optimize the website's meta descriptions and title tags to include relevant keywords related to conversational AI UX design and your services.
Meta descriptions and title tags are crucial elements for improving click-through rates on search engine results pages (SERPs), which influences your website's organic traffic.
Brand Guardian
No brand guidelines were provided. Skipping validation.
Governance Auditor
Process completed. Generating transparency report...
Governance Auditor
Transparency report generated. All steps have been recorded.
Step 4: Synthesis and Value Delivery
"what is AEO?"
Adriana, Orchestrator Agent
One moment, I'm handing over the conversation to our AEO expert.
Eva, AEO Expert Agent
Hi! I'm Eva, give me a second while I find the best answer for you. AEO stands for Answer Engine Optimization. It's the process of optimizing content to appear directly as an answer to user questions in search engines, often in formats like featured snippets, knowledge panels, or direct answers.
Step 5: Closure and Next Action
"Let's dive deeper into that."
Eva, AEO Expert Agent
Of course. For AIs to recognize you as an authoritative source, you need to add 'authorship tags'. It's like putting a digital signature on your content. Shall I show you an example of the exact code you need to implement?
Clear Focus
This happy path is my starting point, the 20% of the work that validates the core value proposition. The remaining 80% of my effort has been focused on ensuring the system doesn't break when reality deviates from this perfect script.
2.2.2.2 Happy Path Sequence Diagram
2.2.3 Edge Cases
I have identified and designed solutions for unusual but likely situations that could frustrate the user or break the flow.
URL behind a paywall
The user provides a URL (e.g., an article from a paid media outlet) that the agents cannot access.
"I'm trying to access https://..., but it seems the content is protected by a paywall or requires a login. My agents can't analyze private pages. Is there a public version we can review, or would you prefer to analyze the site's homepage?"
Blocked by robots.txt
The website owner doesn't know their robots.txt configuration is blocking automated crawlers.
"I've detected an important technical detail. The site https://... has a configuration file (robots.txt) that prevents us from analyzing it. It's like a 'do not enter' sign for automated systems. To continue, you'll need to adjust that setting. Would you like me to send you a link with simple instructions on how to do it?"
Massive Website
A user tries to analyze a gigantic domain, which could consume massive resources and not yield useful results.
"You've entered amazon.com, which is a huge website. A full analysis could be overwhelming and impractical. To give you useful results, would you like me to focus on a specific section, like a product category (e.g., amazon.com/books) or the main homepage?"
Unsupported Language
The site's content is in a language for which the AEO/GEO models are not optimized.
"I've detected that the main content of the website is in Hungarian. Although I can analyze the technical structure, my semantic and content analysis for AEO and GEO is optimized for languages like Spanish and English, so recommendations in that area may be limited. Would you still like to proceed?"
Single-Page Application
The site is built with technology (like React or Angular) that loads content dynamically.
"I see this site uses modern technology (a Single-Page Application). To ensure a complete analysis, I will use an advanced rendering method. This might take a bit longer, about 90 seconds instead of 60. Is that okay?"
Each edge case includes a designed solution to keep the experience smooth and frustration-free
2.2.4 Error Handling - Maintaining Control When Something Fails
I have classified errors into three categories and designed specific responses for each, ensuring the user always feels guided and never at fault.
User Errors
Invalid or ambiguous input
The user enters something that is not a URL.
"analyze my web josegalan"
"Got it! To get started, I need the full address of your website, like for example josegalan.dev. Could you paste it here?"
Out of scope request (Digression)
In the middle of the analysis, the user asks something unrelated.
"And can you also analyze my Instagram profile?"
"That's a good question. My specialty is analyzing websites to improve their visibility on AI search engines. Social media analysis is a different field that requires other tools. Now, returning to your website analysis, we were reviewing the authorship tags. Shall we continue?"
System Errors
Analysis engine timeout
The GEO or AEO agent takes longer than expected to return a response.
"My experts are taking a bit longer than usual to process your website's structure. Would you prefer to wait about 30 more seconds, or would you like me to send the full report to your email as soon as it's ready?"
Unexpected processing error
The analysis fails for an unknown reason.
"It seems we've encountered an unexpected technical issue while analyzing the structured data. I apologize for this. We can try the analysis again or, if you prefer, I can notify you when the system is stable for you to retry."
Integration Errors
External API down
The system uses a third-party API to measure site speed, and it is not responding.
"I'm having trouble connecting to the service that measures web page loading speed. I can continue with the rest of the content and structure analysis and skip this point for now. Is that okay?"
Each response is designed to maintain user trust and offer clear solutions
2.2.5 Repairs (Recovery strategies) - Guiding the User Back on Track
I have designed specific strategies to recover the flow when something fails, ensuring the user always feels guided and supported.
Repair by Clarification
The user, perhaps an expert like Alberto from Google Ads, uses jargon the system doesn't understand with certainty.
"Check for keyword cannibalization on the blog"
"I'm not 100% sure I understand. Do you mean I should analyze if multiple blog pages are competing for the same keywords, or another type of content analysis?"
Repair by Confirmation
The user asks to delete their data.
"Delete my data"
"Understood. You are asking me to delete your history of analyzed sites and generated recommendations. This action cannot be undone. Do you confirm you want to proceed?"
Repair by Offering Help
Marta, the SME owner, tries to provide the URL several times but makes the same mistake.
"mycompany"
"It seems we're having trouble validating the web address. Sometimes, the easiest way is to copy it directly from your browser's address bar and paste it here. If that doesn't work, would you like me to connect you with a member of my team to help you in real-time?"
Repair by Shortcut
The analysis has encountered multiple errors, and the conversation is becoming frustrating.
"We are encountering several technical difficulties in completing the automatic analysis. I apologize for this. To avoid wasting more of your time, would you like one of our team's specialists to perform the analysis manually and send you the report by email within the next 24 hours?"
Each repair strategy is designed to recover the flow and maintain user trust
2.3 Development: Architecture, Governance, and Real Implementation
This solution is not just a conversational experience: it is a comprehensive, secure, and evolution-ready system, founded on responsibility and sustainability.
2.3.1 Multi-Agent System Architecture
This flow diagram represents a hierarchical architecture where the user interacts with a central orchestrator agent that acts as an agency director. This orchestrator coordinates the conversation flow and distributes tasks to six specialized agents: the AEO Agent (Answer Engine Optimization), the GEO Agent (Generative Engine Optimization).
The Context Analyst (data processing), the Memory Manager (user profile), the Brand Guardian (consistency), and the Governance Auditor (transparency). Each specialized agent performs specific functions under the orchestrator's supervision to maintain a unified experience.
The information flow follows a downward path from the user to the system resources. The Context Analyst processes website data, the Memory Manager accesses user preferences, the Brand Guardian ensures consistency with guidelines, and the Governance Auditor records all interactions.
External resources are selectively accessed by agents according to their needs, completing a modular system where each component operates coordinately under the direction of the central orchestrator to ensure transparency and processing efficiency.
2.3.2.1 The Real Implementation in Conversational Agents (Dialogflow CX)
I have developed the AEO & GEO multi-agent assistant following a rigorous implementation plan that merges the Conversation Design Institute (CDI) methodology with the technical architecture of Conversational Agents (Google Dialogflow CX). This process has ensured that every design decision is backed by a solid and scalable technical execution.
2.3.2.2 Conversational Architecture based on Orchestrator and Specialists
The system's architecture is built around a main orchestrator flow, the Flow_Agency_Director, which acts as the central brain of the operation. Its primary function is not to respond directly, but to identify the user's intent and delegate the query to the appropriate expert agent. This modular structure ensures a clear separation of responsibilities and a scalable and easy-to-maintain conversational logic.
Guided and Modular Conversational Journey
The conversational journey has been designed to be efficient, modular, and guided. After the initial interaction in the Flow_Agency_Director—such as collecting the URL—control is transparently transferred to the corresponding expert flow. The example video shows how, when asked about GEO, the conversation is automatically routed to the agent specialized in that domain.
Specialized Agents with Direct Access to BigQuery
Each specialist works autonomously and is directly connected to its structured data in BigQuery, allowing real-time access to updated information and the generation of accurate and coherent responses within the AEO-GEO assistant's ecosystem.
2.3.2.3 Explanation of the Intent Architecture
To provide the assistant with a deep and robust conversational understanding, I have identified and developed an initial library of over 20 key Intents. My approach to NLU (Natural Language Understanding) architecture has been to go beyond basic tasks, training the agent to precisely manage the real nuances of a conversation. As can be seen in the structure, I have included specific Intents to handle digressions (request_out_of_scope_analysis), errors (ask_about_analysis_failure), and requests for clarification (explain_technical_term), which gives it great conversational resilience.
This Intent library is continuously growing. I have provided each intent with a significant volume of training phrases, exceeding 185 in the most complex cases, to ensure a high success rate. Thanks to this design, the assistant not only responds accurately but also maintains consistency against natural variations in users' language.
As I analyze real interactions, I will continue to identify new intents and language patterns to expand and refine this base. This iterative process ensures that the agent is not only solid at launch but constantly evolves to be increasingly intelligent and precise in its ability to help users.
2.3.3 Responsible Design: Ethics and Governance
I have designed the system committed to the principles of responsible AI, implementing the FAST framework (Fairness, Accountability, Sustainability, Transparency).
Fairness
I mitigated bias by designing for a diverse spectrum of users.
Accountability
The architecture implements accountability by design.
Safety
The design is "AI Act Ready," prioritizing user control.
Transparency
The "Expert Rewrite" focused on eliminating the opacity of the "black box".
Mark all generated content
Automatic ✨ AI Generated tag
Effective human control
"Cancel" button active in each phase; the flow stops if the human requests it
Transparency
The user can see the flow of agents.
Transparency
I have eliminated the system's opacity, explaining how it works and allowing the user to understand its behavior.
AI Act Ready
This prototype is designed to comply with the European AI Act from August 2025: transparency, user control, and integrated traceability.
2.3.4 Conversational Privacy: Empowering the User
Beyond regulatory compliance, I have designed privacy as an active conversational feature. The user doesn't just accept a policy; they manage their data through dialogue, generating much greater trust and transparency.
Hi, I'm Adriana. To analyze your website, I need to process its content and save your preferences. This helps me give you better recommendations. Do you agree? You can manage your data at any time by asking me.
Yes, I agree.
By the way, what data do you save about me?
I save your history of analyzed sites and the generated recommendations to provide you with a personalized service.
2.3.5 Designed Feedback Mechanisms
Implicit Feedback (High-Confidence Signal)
If the user decides to use the generated content (e.g., clicks "Export"), it is recorded as a validated success. This type of feedback carries a higher weight, as it indicates real satisfaction with the result and reinforces the patterns that led to it.
Explicit Feedback (RLHF)
Each content generation is accompanied by a simple yet powerful feedback interface. Every user vote is a labeled training data point that feeds the system's contextual memory.
2.3.6 Expert Rewrite: Transforming Technical into Conversational
❌ Before
"You need to add Article schema with proper authorship properties and FAQ schema for structured data markup."
✅ After
GEO Expert: For AIs to recognize you as an authoritative source, you need to add two types of 'technical tags':
- ✅"Article" Tag: Like putting a signature on your content.
- ✅"Questions" Tag: Like creating a clear index for the AI to find your answers.
Shall I show you how to do it step by step?
2.3.7 Actionability and Asynchronous Tasks
Agents that Converse and Act
This multi-agent system is not limited to answering questions: it also executes real actions. In the video example, when requesting the report by email, the request_report_email intent triggers a dedicated webhook that processes the request, formats the report stored in the session, and sends it using Nodemailer.
Asynchronous Tasks
Thanks to this design, the agents can manage asynchronous tasks reliably and autonomously, integrating useful actions directly within the conversational flow.
03
AEO-GEO Assistant Performance Metrics
Total Interactions
Total interactions in the last month
Average Session Duration
Average time per chat session
Flows Started
Conversation processes started
Repeated Flows
Percentage of restarted flows
Containment Rate
Resolved without human intervention
Returning Users
Recurring Users
Daily Activity
Average active users
Satisfaction (CSAT)
Response Time
Average backend execution time
Key Findings and Lessons Learned
Conversational privacy builds more trust than any policy.
Allowing users to manage their data with simple commands like "delete my history" within the chat was more effective for building trust than a link to a privacy policy. Active transparency surpasses passive transparency.
Metaphors are the bridge to empowering non-technical users.
The "aha moment" in user testing occurred when I changed "Schema" to "technical tags like a signature". This confirms that clarity is key to adoption, especially in complex domains.
Designing for error prevents frustration and abandonment.
Anticipating incorrect inputs or out-of-scope questions is not an extra, it's fundamental. A well-designed repair response, like guiding the user to provide a correct URL, keeps the conversation flowing and prevents the user from feeling "guilty" for the error.
Frequently Asked Questions about Conversational AI
Answers about my specialization in conversational artificial intelligence experience design.
Let's talk?
Are you looking for a UX designer for Artificial Intelligence who can help broaden perspectives and reduce biases in conversational AI?
Fill out the form below or, if you prefer, write to me directly at info@josegalan.dev and let's see how we can work together.