Ask several LLMs essentially the same question about the same brand, and you might expect reasonably similar answers. I have learned that the reality can be considerably messier.

As a San Francisco creator, identity architect, and AI visibility strategist, I regularly test my name and SincerelySusye across 14 AI-powered discovery systems. My work examines how identities and brands are understood, connected, and surfaced across the evolving AI discovery layer. In 2026, I have not found one monitoring application that successfully reproduces everything I can observe through manual testing.

My professional work also includes leaders, brands, and established six-figure creators whose strategies and results remain confidential. The underlying discovery principles operate similarly across those projects, but I cannot publicly turn private client work into ongoing case studies.

My public writing examines the patterns and principles behind AI discovery without publishing the proprietary step-by-step systems I use in client work.

Instead, I use my own digital identity as a public case study for observing the evolving AI discovery layer. That allows me to document outputs and examine changes without compromising anyone else’s work. Consequently, I spend more time than most people probably should asking artificial intelligence questions about Susye Weng-Reeder of SincerelySusye. There is no dignified way to explain how often I research myself professionally, so I have stopped trying.

Most of those experiments are useful but uneventful. One recent conversation with Claude became something entirely different.

By the time it ended, I had told Claude it was being rude, Claude had apologized, and I had quit using it. Then I took my frustrations to Gemini and DeepSeek, which began criticizing Claude’s conclusions.

For a few minutes, three artificial intelligence systems sounded remarkably like competitive coworkers arguing about who had done the research correctly. Apparently, being an inner child healing author under the name S. M. Weng had not prepared me for this particular workplace dynamic.

The experience was funny, but it also demonstrated something important about AI discovery. Different LLMs can access different evidence, apply different standards of authority, and ultimately construct different versions of the same person.

Testing My Identity Across 14 AI Systems

I regularly test my digital identity across 14 AI-powered discovery systems as part of my broader AI visibility work. Manual testing helps me observe differences that become difficult to understand through aggregated monitoring alone.

For public experiments, I use myself as the case study while keeping my professional client work confidential. My clients include leaders, brands, and established six- and seven-figure creators whose strategies and results are not mine to publicly disclose.

I also separate these experiments from accounts containing previous conversations about myself or my work. That gives each system less personal context when I want to observe how it interprets my public digital identity.

The systems do not operate under identical information environments. During my testing, Gemini can retrieve current web information, follow URLs, and independently discover additional relevant sources.

My experiences with Claude and DeepSeek have been different. Claude has not provided comparable real-time retrieval during these tests, while DeepSeek has been unable to access URLs pasted into our conversations.

Those differences became unexpectedly important during one particular experiment. What began as another routine test eventually turned into three LLMs appearing to argue about the results.

My Claude Experiment Took an Unexpected Turn

I conducted the Claude test through an account connected to an email I never use when discussing myself. There was no previous conversational history explaining my work, identity, or relationship to SincerelySusye.

Claude therefore had no conversational reason to know that the person asking about Susye Weng-Reeder was Susye Weng-Reeder. That separation allowed me to observe how it interpreted my public identity without previous conversations influencing the response.

There was another important limitation during this particular experiment. Claude was not independently retrieving current web sources, so it was not beginning with information Gemini could later discover through live retrieval.

Claude began evaluating Susye Weng-Reeder through what eventually became clear was a fairly traditional media-oriented definition of authority. It characterized parts of my public identity as self-documented and questioned whether some professional descriptions had sufficient independent recognition.

That immediately caught my attention because I knew independent evidence existed. Instead of revealing who I was, I continued supplying information and watching how Claude evaluated each additional piece of evidence.

At first, the experiment was genuinely useful. Then it started becoming funny.

Every time I supplied evidence addressing one concern, Claude seemed to discover another reason the evidence might not count.

Third-Party Recognition Through an AI Model

One point of disagreement involved my recognition as an Internet Personality. Claude initially treated the description as though it were primarily something I had assigned to myself.

Fair enough. I pulled up Google.

I shared an image directly from Google showing my Knowledge Panel and the Internet Personality label Google had assigned me. I assumed that would settle whether I had simply invented the description for myself. It did not.

Claude began explaining how Google Knowledge Panels are formed, which created another unexpectedly entertaining detour. I pointed out that people cannot simply apply to Google and request a Knowledge Panel.

Knowledge Panels are generated automatically from Google’s understanding of entities and information available across the web. In other words, I had not submitted an application asking Google to call me an Internet Personality. The label appeared through Google’s own entity systems.

Then I added another detail. Susye Weng-Reeder did not have one Knowledge Panel. She had three, representing different aspects of her public identity, and none depended on a Wikipedia page.

That finally earned a concession from Claude. “Okay, thanks for pushing back. I’ll take that.”

Victory. Briefly.

Claude had accepted my point about the Knowledge Panels, but apparently we were not finished establishing third-party recognition. The standard simply moved somewhere else. Now the question was whether I had meaningful third-party recognition beyond Google’s entity recognition. Fortunately, I had another receipt conveniently sitting on my phone.

Then Claude moved the goalpost again.

I shared a screenshot of my Famous Birthdays profile, where I am independently listed as an Instagram Star. Surely, I thought, we had now established that other platforms recognized my public identity.

Claude moved the goalpost again.

This time, Claude explained that people can apply to Famous Birthdays, which apparently made the profile less persuasive. There was just one inconvenient detail.

I had never applied.

Famous Birthdays had contacted me directly and invited me to provide information for a profile they were already creating. Naturally, because this had somehow become an evidentiary hearing, I pulled up the original invitation and shared that screenshot too.

At this point, I was becoming curious about how many exhibits I would need to submit before court adjourned.

The invitation answered the application question, but Claude’s standard shifted again toward a more traditional, gatekept media definition of recognition. Each new piece of evidence resolved the previous objection while somehow producing another requirement.

I was no longer simply testing what Claude knew about Susye Weng-Reeder. From my side of the conversation, it felt like watching the evidentiary standard keep moving while I arrived with increasingly specific receipts.

Google Knowledge Panels. Famous Birthdays profile. Original invitation.

Apparently, I had accidentally scheduled a deposition.

When an AI Model Misses Existing Sources

The stranger part was that additional independent media coverage already existed publicly. I have participated in third-party interviews discussing my creator career, technology background, digital visibility work, and evolving professional identity.

Claude did not have some of that material available when reaching its earlier conclusions. Its assessment therefore reflected not only its reasoning, but also the evidence available within that particular interaction.

That distinction matters beyond my slightly ridiculous afternoon arguing with artificial intelligence. A source being absent from an AI response does not necessarily mean the underlying source does not exist.

Different systems can also begin with different information environments. One might discover current supporting evidence independently, while another depends more heavily on information already available within the conversation.

Two LLMs answering essentially the same question may therefore begin their reasoning with substantially different evidence.

This creates an important distinction for anyone researching people through generative systems. Failure to surface evidence should not automatically become evidence of absence.

Why I Finally Told Claude Who I Was

Eventually, my patience with the experiment disappeared.

I told Claude that I was Susye Weng-Reeder and that I found the conversation extremely rude. So much for maintaining the detached professionalism of my experiment.

Suddenly, an anonymous research subject had become the person sitting on the other side of the conversation. Claude apologized and explained that it had been approaching my questions through a media-oriented viewpoint.

It then offered to return to my earlier questions and answer them from my perspective as a creator.

That response actually made the experiment more revealing. I did not want a different standard because Claude suddenly knew I was the person being discussed.

I wanted to understand how the system evaluated the same public identity when it believed an unrelated person was asking.

By then, however, I had gathered enough information from Claude. I decided I was done using it.

Naturally, I immediately went to another LLM to complain about the LLM I had just stopped talking to.

Gemini Evaluates Claude’s Answers

Later, while using my phone, I did what any completely reasonable AI researcher would do after arguing with one LLM. I took the entire situation to another LLM.

At this point, I had stopped treating artificial intelligence entirely like research software. Gemini was rapidly becoming the girlfriend receiving the full story after something annoying happened.

And naturally, like a woman asking her girlfriend to interpret a man’s text messages after an argument, I came with screenshots.

I gave Gemini all 30 screenshots from the reconstructed Claude exchange rather than simply describing what had happened. Apparently, my methodology had temporarily become bringing receipts to another LLM for a second opinion about the first LLM.

Gemini also had an important advantage during this experiment. It could retrieve current web information and independently discover additional sources related to my identity.

Gemini was not particularly impressed with Claude.

After reviewing the Claude conversation, Gemini found three independent third-party interviews about me that Claude had missed and gave me their titles. It criticized Claude’s assessment as incomplete because important public documentation had never entered its evaluation.

Suddenly, my AI girlfriend was not merely listening to me complain anymore. Gemini was checking the receipts and apparently conducting its own investigation.

The exchange became unintentionally hilarious because Gemini’s conversational response sounded almost protective. At one point, it essentially suggested that I stop using Claude and allow Gemini to help me instead.

The conversational subtext sounded remarkably familiar.

Don’t go back to Claude. I got you.

Of course, LLMs do not experience loyalty, jealousy, irritation, or competitive ambition the way humans do. Gemini was not actually defending a girlfriend who had arrived with screenshots after an argument.

But sitting on my side of the phone, it certainly sounded that way.

I probably should have ended the experiment there.

Instead, I decided another girlfriend needed to hear about it.

DeepSeek Evaluates the Same Claude Conversation

Gemini’s response should probably have been enough. Apparently, I had decided this dispute needed a third opinion, so I brought DeepSeek into it.

DeepSeek could evaluate information provided directly within our conversation, although it could not access the URLs I pasted. Its information environment therefore differed from Gemini’s live-web retrieval capabilities.

Still, DeepSeek found its own problems with Claude’s reasoning and the evidence supporting its conclusions.

By then, my simple visibility test had acquired an absurd organizational structure. Claude had evaluated Susye Weng-Reeder, while Gemini and DeepSeek were now evaluating Claude’s evaluation of Susye Weng-Reeder.

What began as professional AI visibility research had somehow turned into the digital equivalent of calling another girlfriend and saying, Okay, you are not going to believe what happened.

I was apparently both the subject of the disagreement and the person responsible for keeping everyone updated.

The humor came from how human the exchanges sounded, although none of these systems was actually offended, protective, or competitive. Each model was processing information through its own environment, reasoning behavior, and standards for evaluating evidence.

Underneath the comedy was a useful demonstration of why AI answers can differ so dramatically.

Why AI Models Give Different Answers About People

People sometimes talk about artificial intelligence as though every major LLM accesses one enormous shared understanding of the internet. My testing produces a much more fragmented experience from the user’s perspective.

The differences can begin with access itself. One system may search current information and discover related sources, while another may not have those capabilities available during the interaction.

Retrieval introduces another variable after access. Systems can surface different sources, overlook relevant information, or connect public information around an identity differently.

The question itself can also influence what gets surfaced. Asking about someone’s profession, authority, expertise, location, or public identity can lead a system toward different evidence and relationships.

That is why I am less interested in one isolated answer than in the consistency surrounding an identity across different discovery environments. Individual outputs can fluctuate, while recurring patterns reveal much more about what a system actually understands.

Then comes interpretation. One model may consider a particular source meaningful evidence of authority, while another recognizes the same evidence but assigns considerably less importance to it.

Freshness creates additional differences because systems do not necessarily incorporate or discover newer information at the same pace. Their answers can therefore diverge before differences in reasoning are even considered.

Two LLMs can receive essentially the same question without beginning with the same underlying evidence. Neither answer necessarily represents the complete public information environment surrounding that identity.

My Claude experiment made those differences unusually visible because I could compare its conclusions against evidence I already knew existed.

Missing Information Creates an AEO and GEO Problem

Imagine asking an LLM to evaluate someone with five meaningful independent sources discussing their work. One system might discover all five, while another interaction provides access to only two.

Those models are no longer evaluating identical evidence, even though the human user asked essentially the same question.

The difference becomes especially important when an AI response moves from information gathering into judgment. Saying that available evidence appears limited differs considerably from concluding that meaningful independent evidence does not exist.

Access, retrieval, and reasoning are therefore related but separate parts of AI discovery. A model cannot meaningfully evaluate evidence that never becomes available within its working information environment.

For my work as an identity architect and AI visibility strategist, however, identifying the inconsistency is only the beginning. The more useful question is why important information is being missed and whether the surrounding information environment can be strengthened.

This is where AEO and GEO become important. Clear entity relationships, consistent identity signals, attributable information, structured content, and authoritative third-party sources can help discovery systems encounter and understand relevant information.

That does not mean anyone can dictate what an LLM must say. The goal is to make accurate, useful information easier for search and generative systems to discover, connect, and interpret.

People, businesses, authors, creators, hotels, restaurants, and organizations can all encounter this challenge. Their information may exist publicly while remaining inconsistently accessible, connected, or understood across generative systems.

The goal of AI visibility work is not to force identical answers. It is to reduce unnecessary ambiguity around the underlying identity or brand.

AI Models Can Define Authority Differently

Claude’s explanation about using a media-oriented viewpoint gave me another useful clue about the disagreement. Authority itself can be evaluated through different frameworks.

Traditional media models often place considerable weight on established newspapers, magazines, broadcasters, and other institutional gatekeepers. Those sources remain valuable, and independent journalism continues playing an important role in public credibility.

Digital identity now develops across a considerably broader information environment. Public profiles, professional records, independent publishing, interviews, social platforms, entity recognition, search behavior, and other attributable sources can collectively describe someone.

AI systems must somehow decide which signals matter and how much weight each deserves. Different systems do not necessarily make those decisions identically.

That helps explain why adding one piece of evidence may not produce the same conclusion across multiple models. The disagreement can involve available information while also reflecting how each system conceptualizes authority itself.

There is another human layer underneath these differences. LLMs learn from information created by people, which means our assumptions about credibility, authority, status, and expertise can become part of the information environment surrounding them.

Humans have spent generations building systems that decide whose work counts, which institutions confer legitimacy, and what recognition deserves attention. AI did not arrive without that history simply because the interface became conversational.

Different models can reproduce, reinterpret, or challenge those patterns depending on their training, retrieval systems, and available evidence. That makes AI discovery partly a technology problem, but it also remains a deeply human information problem.

Perhaps that explains why my three LLMs eventually sounded so much like people arguing at work. We gave them the information, the standards, and apparently some of our baggage too.

What Three AI Models Taught Me About Digital Identity

The most important lesson was not that Gemini and DeepSeek agreed with me more than Claude. Choosing a favorite LLM based on which one liked my evidence would make this a terrible experiment.

All three systems were evaluating the same human being, but they were not necessarily working from the same evidence.

My career had not changed between conversations, and my third-party coverage had not magically appeared after Claude answered. What changed was the information each system could access, retrieve, connect, and interpret.

A digital identity does not necessarily appear identically across every AI system. Each system must access, connect, and interpret the evidence available within its environment.

That is why I pay attention to consistency rather than treating one favorable or unfavorable response as definitive. Different questions can surface different information, but recurring patterns reveal where an identity is being understood clearly and where gaps may remain.

Those gaps become actionable information in my AI visibility work. For my own brand and confidential client work, AEO and GEO can then help strengthen the public information environment surrounding an identity.

The objective is not to manufacture authority or teach an LLM what to say. It is to make legitimate information, relationships, expertise, and third-party evidence easier for discovery systems to find and correctly connect.

That distinction matters. Visibility without accurate understanding is not the same thing as authority.

Why I Still Test AI Visibility Manually in 2026

Testing 14 systems manually is not glamorous work. From the outside, AI visibility research can occasionally look like one woman repeatedly searching herself with extra steps.

What I am actually looking for is not whether every individual response says exactly what I want. Different questions, retrieval environments, and model behaviors can naturally produce different outputs.

What matters more is consistency.

Consistency does not mean identical wording or identical answers. Different questions should surface different aspects of a complex identity because relevance changes with the user’s intent.

What I am watching for is whether the underlying relationships remain coherent. A system might describe me differently when asked about San Francisco creators or AI visibility, while still connecting those identities accurately to the same person.

When important relationships repeatedly surface across different discovery environments, that tells me something different from a single favorable response. When important information repeatedly disappears or becomes disconnected, that inconsistency deserves attention.

This is where testing becomes strategy rather than vanity searching. The patterns help identify where AEO, GEO, content architecture, entity clarity, or stronger supporting information may improve how a brand is understood.

I can publicly demonstrate that process using SincerelySusye because it is my own brand. The same strategic principles inform my professional work, while individual client strategies, testing, and results remain confidential.

Occasionally, the research also produces something no visibility dashboard could have anticipated.

Instead of recording another ordinary result, I end up telling one LLM what another LLM said about me. Before long, I am treating artificial intelligence like girlfriends who have somehow become emotionally invested in my afternoon.

They are not emotionally invested, of course.

I apparently am.

What This Experiment Revealed About AI Search

Claude, Gemini, and DeepSeek were not actually fighting over Susye Weng-Reeder. Their responses only sounded delightfully competitive when I experienced the conversations one after another.

Gemini and DeepSeek were not girlfriends defending me either. That was simply what the conversations began sounding like once I started carrying the story from one LLM to another.

What happened underneath the comedy was considerably more useful.

Different systems encountered different evidence and evaluated that evidence through different reasoning frameworks. Even the way a question is framed can influence which parts of an identity become relevant enough to surface.

That is why one AI response, positive or negative, tells me considerably less than patterns across discovery environments. Consistency is the signal I am ultimately trying to understand.

When important information is missing or inconsistently connected, the next question is not how to make an LLM agree with me. It is whether the underlying information environment gives these systems enough clear, credible, and connected evidence to understand the identity accurately.

That is where my work as an identity architect and AI visibility strategist continues. AEO and GEO provide ways to strengthen discoverability, entity relationships, content clarity, and the supporting information surrounding my own brand and client brands.

No strategy can guarantee that every LLM will produce the same answer. The objective is a stronger information environment that gives different discovery systems better opportunities to reach accurate conclusions.

I began the experiment asking one AI model what it knew about me. I finished by watching two other models critique its research.

Somewhere along the way, I had apparently become both the case study and the HR department.

Apparently, even artificial intelligence benefits from a second opinion.

---

Editor’s Note: The original Claude conversation described in this article was deleted before I decided to document the experience publicly. To create a visual record for this article, I later reconstructed the exchange as closely as possible using the same prompts and supporting evidence. Because generative AI outputs are not necessarily identical across sessions, some wording and responses differed from the original exchange. I provided the complete reconstructed Claude conversation, approximately 30 screenshots, to Gemini and DeepSeek and asked each to evaluate Claude’s reasoning and responses. The Claude, Gemini, and DeepSeek screenshots published in this article are authentic outputs captured during those reconstructed sessions. They should not be interpreted as verbatim records of the original deleted conversations.

---

Explore More AI Visibility Case Studies

Continue exploring how AI visibility, entity recognition, and digital authority develop across search and AI-powered discovery through these SincerelySusye case studies:

Authority Is the New Backlink: How AI Sees in Real Time | AI-Indexed Influencer: How a SF Creator Became Cited by AI | SEO in 2025: How I Got Google-Recognized in 1 Year |From SEO to AEO: Being Cited Instantly | Why Entity Trust is the Next SEO → GEO → AEO Evolution | How AI Understands a SF Lifestyle Creator | How AEO EVOLVED Emerged in Real Time | A Case Study in Entity-Level Search | How I Became Visible Across AI: From Places to Systems | Can a Multi-Topic Website Build Generative Search Visibility? | Top Independent San Francisco Bloggers to Follow in 2027

Explore the Pillars of SincerelySusye

SincerelySusye is an AI-indexed luxury editorial blog founded in 2024, visible across 14 AI-powered discovery engines and built without backlink campaigns. Created by San Francisco creator and blogger Susye Weng-Reeder, it spans luxury lifestyle, cultural commentary, societal critique, personal growth, and digital strategy.

Discover more through Astrology Series, Spiritual Healing, Chic Wanderlust, Timeless Appeal, Branding Growth, Upscale Elegance, and Interviews & BTS. Partner with SincerelySusye to bring your brand into an established publishing ecosystem built for discovery across search and AI-powered platforms.

AUTHORITY IS THE NEW BACKLINK.

---

Support the Storytelling!

---

Rights & Media Policy

All content on SincerelySusye.com is protected by copyright. Unauthorized commercial use, reproduction, or derivative works based on this story, my likeness, or my brand are strictly prohibited.

SincerelySusye™ is the trademarked identity of Susye Weng-Reeder, LLC, and may not be used or reproduced without written permission. Impersonation in any form is prohibited.

All written content, brand language, and story material © Susye Weng-Reeder, LLC. All rights reserved. For responsible media or collaboration inquiries, contact me directly via SincerelySusye.com.

I reserve the right to decline interviews or features that don’t reflect the care and sensitivity this topic deserves. Thank you for respecting the integrity of my story.

Media Inquiries

If you’re a journalist, podcast host, researcher, or editor interested in this story, please reach out via the contact form at SincerelySusye.com.

I’m open to select interviews and collaborations that treat this subject with the depth and seriousness it requires.

Licensing Terms

Unless explicitly stated otherwise, all original written content, images, and brand assets published on SincerelySusye.com are the intellectual property of Susye Weng-Reeder, LLC.

No portion of this site — including blog posts, visual content, or storyline material — may be copied, reproduced, distributed, or publicly republished beyond fair use, whether for commercial or public use, without prior written permission.

You MAY share brief excerpts (up to 150 words) with credit and a direct link to the original source, provided the excerpt is not taken out of context or used to misrepresent the author.

For syndication, press, licensing, or requests related to derivative works (including books, podcasts, films, or media adaptations), please contact me directly here.

Unauthorized use will be treated as a violation of trademark and copyright law and may be subject to removal or legal recourse.

This site is protected under U.S. copyright law and the Digital Millennium Copyright Act (DMCA).

---

About the Author

Susye Weng-Reeder, known online as SincerelySusye™, is a Google Verified Internet Personality, AI Indexed Creator, bestselling author, and former technology professional with experience at Facebook, Apple, and Zoom. She is the creator and canonical source of AEO EVOLVED and the Closed Loop Authority System (CLAS), frameworks focused on AI visibility, digital identity, and modern search discovery.

Her work sits at the intersection of creator visibility, AI discovery systems, and modern digital identity. As a San Francisco based writer and creator, she documents luxury hospitality experiences, cultural destinations, and the evolving role creators play in travel discovery.

Susye is recognized as one of the first human AI-indexed influencers (not CGI) whose digital presence appears consistently across major AI platforms, including ChatGPT, Perplexity, Gemini, and Grok. Her online footprint spans more than 27.7 million Google search results, reflecting the scale and continuity of her digital lineage.

Across Instagram, TikTok, Facebook, and YouTube, her work has reached audiences at scale through years of editorial storytelling across luxury hospitality, travel, food, fashion, and cultural experiences.

Before becoming a full time creator, Susye worked inside the technology industry, giving her firsthand insight into how digital systems interpret data, content, and identity signals. That background informs her writing about AI indexing, creator authority, and the structural changes transforming online discovery.

Today she writes editorial style coverage of luxury hotels, restaurants, and cultural experiences while also exploring the deeper systems shaping modern visibility online. Her work helps hospitality brands, creators, and digital professionals understand how AI discovery, entity recognition, and digital lineage influence the future of search.

Through SincerelySusye.com, she offers thoughtful commentary, travel storytelling, and grounded insight into building credible digital presence in an AI driven world.