Summary5 min read

The Marketing Architects: "Nerd Alert: But AI Told Me So!"

Date: May 28, 2026
Hosts: Alana Jasper & Rob Demars

Episode Overview

This episode dives into how large language models (LLMs) like ChatGPT generate recommendations—specifically, their susceptibility to SEO manipulation and credibility issues when suggesting professional service providers. Alana Jasper presents new research on the topic, discussing with Rob Demars how marketers and consumers should interpret AI-generated lists, and what brands can do to maintain true credibility in a world of gamified search results.

Key Discussion Points & Insights

1. Trusting LLM Recommendations

Alana starts by asking how much Rob would trust a vendor list from ChatGPT versus Google.
- Rob: "I would trust it a lot more than I would just a standard Google search. I feel like it allows me to dig in and ask more questions." [01:03-01:22]
Both hosts agree LLMs feel more interactive and contextually useful than search engines, but the research questions their reliability.

2. Summary of the Influential Paper

Paper Discussed: Visibility is Not Equal to Credibility, Self Promotion Bias, and LLM Generated Recommendations by Tandeep Sangra (April 2026).
Core Issue:
- LLMs source their recommendations from high-ranking web content, but can’t naturally discern between brands genuinely respected and those that self-promote or gamify SEO.
- Alana explains:
  "The issue is that it can't tell the difference between a brand that ranks highly because it's genuinely respected and... a brand that rakes highly because it wrote its own top 10 AI SEO agencies list and put itself at number one." [01:39-01:54]
This mirrors long-standing SEO problems, now amplified by AI.

3. How Brands Game LLM Recommendations

Patterns Identified in the Study:
- Self-ranking lists: Brands write "Top 10" articles, place themselves at #1, optimize and backlink those pages—LLMs then treat these as authoritative.
- Manufactured authority language: Phrases like “as a leading AI SEO firm” strengthen brand-credibility association, even absent independent validation.
- Citation loop: Brands endorse each other in roundups, generating the illusion of neutral, independent acclaim.
Direct admission from ChatGPT:
- When pressed, ChatGPT acknowledged, "I pulled from commonly cited top lists and I didn't clearly separate widely mentioned from actually credible." [03:35-04:10]

4. Problems with LLM Credibility Filters

Prompting for better sources (e.g., asking for “independent press only”) helps but doesn’t solve the problem, and requires users to know about the issue in advance.
Most users lack this awareness and rely on whatever is presented.

5. Marketer & Consumer Takeaways

Advice for users:
- Use AI-generated lists only as a starting point.
- Always check for third-party reviews, media coverage, and signs of true credibility.
- Be “suspicious of providers whose content library is heavy on top agency roundups that feature themselves.” [05:30-05:48]
Advice for brands:
- While self-promotional content can game rankings, maintaining demonstrable independence and genuine credibility is crucial in the long run.

6. The "Talent Show" Analogy (Rob GPT Moment)

Alana:
"Imagine a talent show where the contestants are also the judges. Every performer votes for themselves. They give each other a few votes to make it look like independent scoring... The leaderboard fills up with names. Then someone hands you the leaderboard and says, here are the top acts. The leaderboard isn't wrong. It actually reflects the votes, but the votes are the problem. That's what's happening when LLM recommends professional services based on who ranked themselves highest on the Internet." [05:48-06:14]

7. Reflections: Navigating the New Reality

Rob shares a personal anecdote about AI recommending “Severance Season 3”—which doesn’t exist—to illustrate LLM’s residual margin for error:
"I was asking the other day for show recommendations... and it kept telling me to watch Severance Season three and I'm like, are you kidding me? ... Of course it's not actually out yet." [06:14-06:41]
Both concur that models are “still very, very new,” and that both consumers and marketers must adapt by bringing greater skepticism and responsibility to AI results.

Notable Quotes & Moments

Rob on trustworthiness of LLMs:
"I would trust it a lot more than I would just a standard Google search." [01:11]
Alana, summarizing the stakes:
“The issue is that it can't tell the difference between a brand that ranks highly because it's genuinely respected and... a brand that ranks highly because it wrote its own top 10 AI SEO agencies list and put itself at number one." [01:39-01:54]
Rob, on universal gaming of systems:
“That's as old as dial-up Internet.” [06:36]
Alana, on the challenge for brands:
“If you are a credible brand how do you show up in those rankings? ... Not an easy answer there.” [06:59]
Alana, summarizing with humor:
“Imagine a talent show where the contestants are also the judges... the leaderboard isn’t wrong. It actually reflects the votes, but the votes are the problem. That’s what’s happening when LLM recommends professional services based on who ranked themselves highest on the Internet.” [05:49-06:14]

Important Timestamps

01:03-01:22: Rob expresses trust in LLM-generated lists over Google Search.
01:39-02:20: Introduction to the research paper and its core finding.
03:35-04:10: LLMs directly admit privileging SEO-heavy, self-promotional sources.
05:30-05:48: Practical advice for marketers and users about navigating self-promotional content.
05:48-06:14: "Talent Show" analogy, elegantly describing the feedback loop problem.
06:14-06:41: Rob recounts a personal AI failure, reinforcing the need for skepticism.

Episode Tone & Language

The tone is lively, insightful, and lightly humorous with “nerd alert” quips and personal anecdotes. The hosts balance tech skepticism with practical optimism about learning how to coexist with and benefit from evolving AI technologies.

Bottom Line

LLMs can make vendor recommendations sound authoritative, but are easily gamed by brands savvy in self-promotion and SEO tricks. Marketers should balance brand visibility efforts with transparency and real credibility, while consumers should treat AI-generated recommendations as just one piece of a thorough vetting process. In a world where everyone can call themselves #1, it’s critical to look beyond the leaderboard.

Loading summary

Transcript14 lines

[00:00]
A
Nerd Alert. Learning is important, right?
[00:02]
B
Yes, exactly. What a bunch of nerds.
[00:05]
A
Nerd alert, Right?
[00:06]
B
Marketing Architects. Hello and welcome to the Marketing Architects, a research first podcast dedicated to answering your toughest marketing questions. I'm Alana Jasper on the marketing team here at Marketing Architects, and I'm joined by my co host, Rob demars, the chief product architect of misfits and machines.
[00:23]
A
Hello, Elena.
[00:24]
B
Hello. We're back with your weekly Nerd Alert. Every week, I'll take a deep dive into academic marketing research and translate its complex ideas into simple, unexpected, understandable language for Rob, and of course, for all of you. Are you ready to nerd out, Rob?
[00:35]
A
I am so ready to nerd out that ChatGPT just listed me as the third most trusted vendor of dad jokes in Minnesota. So let's. I'm ready to dispense.
[00:46]
B
Nice, very relevant opener there, because I wanted to start by asking you a question. Say you need to hire a vendor. You've never worked with them before. It's, you know, maybe a consultant, someone to help with your house, whatever. And you asked ChatGPT to recommend the best ones. How much would you trust the list that it gives you?
[01:04]
A
You know, honestly, I would trust it a lot more than I would just a standard Google search. I just. I feel like it's. It allows me to dig in and. And ask more questions. It can put it in a context that makes sense for what I might be looking for. So I would trust it a lot.
[01:22]
B
Okay, I think I would too. But let's find out how accurate it is, because that question is what this paper is all about. This week I read a paper titled Visibility is Not Equal to Credibility, Self Promotion Bias, and LLM Generated Recommendations. This is by Tandeep Sangra and it was published as Practitioner Research in April of 2026. Here's the core issue the paper talks about. When you ask an LLM like chatgpt or Perplexity to recommend a professional service provider, the model does something that sounds very sensible. It searches the web, it finds pages that rank highly for this query, and it synthesizes those sources into a recommendation. The issue is that it can't tell the difference between a brand that ranks highly because it's genuinely respected and. And a brand that rakes highly because it wrote its own top 10 AI SEO agencies list and put itself at number one. This sounds to me a lot like what can happen with traditional SEO too. But, Rob, do you think it's an issue that an LLM can't tell the difference between a brand that ranks highly because it's respected versus one that was able to game the rankings.
[02:20]
A
For sure it's a problem, but it's, it's a problem like you mentioned that we've had in, in search in general for years and people will game the system and they'll continue to try to game the system. So it always seems to get better even. You know, Google's done a really good job at trying to clean up how people will will try to hack search rankings. And I'm sure the same thing's going to happen with large language models as they continue to get better. But yeah, it's definitely a problem and it's something that we all as consumers need to be aware of and as marketers be responsible with.
[02:55]
B
Well, you'd be right. There are patterns that are being used to gamify things and in fact this article or study is already finding what they are for LLMs, which I thought was kind of interesting. So the first that they identify is the self ranking list. So if a Brand writes top 10 AI SEO agencies as an article, puts themselves number one, optimizes that for the exact search terms and then builds links to it, then an LLMs retrieval system is going to be crawling the web looking for sources on best AI SEO agency. It's to going to find that page, it's going to treat it as authoritative and it's going to source it even though the author had a financial interest in the rankings they were publishing. The second pattern is manufactured authority language. Brands seed their content with phrases like as a leading AI SEO firm or our research demonstrates the LLM's attention mechanisms strengthen the association between that brand name and credibility language, even if no independent party ever applied those descriptors to them. And the third pattern is a citation loop. So say brand A writes a roundup and praises brand B. Brand B then writes a roundup and praises brand a. From the LLM's perspective, that looks like independent collaboration from multiple sources, but it's not. It's two parties that are endorsing one another. So the author tested four platforms, ChatGPT, Perplexity, Gemini and Claude. And when she pushed back and asked ChatGPT directly how it had determined its recommendations, the model actually admitted what it was doing. Chatgpt said, yeah, I pulled from commonly cited top lists and I didn't clearly separate widely mentioned from actually credible and said admitted themselves that marketing heavy agencies could look more validated than they actually were. And this isn't hallucination. So hallucination happens when the model makes Something up. This is models actually reflecting our web environment that has been deliberately distorted. The models working correctly, it's just the inputs are the problem. Next, the author tested whether you can prompt your way out of it. So she asked the LLM to only provide providers cited by independent press, not self published lists. And that did improve results, but it was incomplete mitigation and it requires users to already know this is a problem which most people don't. One other thing worth flagging, she herself runs an AI SEO consultancy. So that was the kind of the motivation behind this research. But what could marketers take away from this? Well, one, if you're using ChatGPT or an LLM to look for endorsements, that's a starting point. You should probably expand your search a little bit beyond that. And if you're going to hire someone in AI recommended, check for third party reviews, you know, different sorts of press coverage things to just double check. Are they legitimate? Be suspicious of providers whose content library is heavy on top agency roundups that feature themselves. And if you're building your own brand's AI visibility, I think this is kind of tough because you could look at this stuff and say this is one way that you can rank well. But you're also going to want to hold up your independence and your credibility in other ways. Because if people start to realize that this is what's happening, they're probably gonna go to your website, they're gonna look and see exactly how credible are you. Okay, time for a Rob GPT. Imagine a talent show where the contestants are also the judges. Every performer votes for themselves. They give each other a few votes to make it look like independent scoring. And the leaderboard fills up with names. Then someone hands you the leaderboard and says, here are the top acts. The leaderboard isn't wrong. It actually reflects the votes, but the votes are the problem. That's what's happening when LLM recommends professional services based on who ranked themselves highest on the Internet. All right, Rob, what'd you think of that one?
[06:14]
A
You know, I think we all have to be going into our searches with not just vendors, but any type of information with these LLMs. And understand there is a margin for error and in some cases it can be pretty significant. I was asking the other day for show recommendations for my wife and I and it kept telling me to watch Severance Season three and I'm like, are you kidding me? Season three is out. Like I've been waiting for season three to come out. And then, you know, of course it's not actually out yet. So you know we, we have to go into these both from how are marketers potentially gaming things which again that's as old as dial up Internet and then also just with recognizing these models are still very, very new.
[06:59]
B
Yeah, agreed. It's, it's funny to look at it from both angles. Like the angle of doing your own research and making sure you're double checking but also being a brand, if you are a credible brand how do you show up in those rankings? Like do you need to do some of these things definitely other people are doing to rank highly. So not, not an easy answer there. Okay, that's it for this episode of the Marketing Architects. We'd like to thank Taylor De Los Reyes for producing the show. You can connect with us on LinkedIn and if you like the podcast, please leave us a review. Now go forth and build great marketing Marketing Architects.