Research

Research

Published research on AI visibility and LLM recommendation systems.

TL;DR — SummaryTandeep Sangra's paper, "Visibility ≠ Credibility: Self-Promotion Bias in LLM-Generated Recommendations," investigates a structural bias in how LLMs surface self-promoted brands as apparently credible, and proposes a five-point framework buyers can use to verify AI-generated recommendations. Published April 2026, available on SSRN and ResearchGate.

Visibility ≠ Credibility: Self-Promotion Bias in LLM-Generated Recommendations

Author: Tandeep Sangra, Independent
Date: April 18, 2026
Available at: SSRN →  |  ResearchGate →

Abstract

Large language models such as ChatGPT and Perplexity AI are increasingly used as recommendation engines for professional services, including marketing consultants and digital agencies. This paper investigates a structural bias in LLM recommendation outputs: the tendency to surface self-promoted brands as apparently credible due to co-occurrence patterns in retrieval data.

Through direct query testing across four major LLM platforms and systematic source analysis — including a documented, on-record acknowledgement by ChatGPT of its own bias — the research establishes that visibility in AI-generated recommendations does not equate to independently validated credibility.

What the paper covers

Independent citation

This research has been cited independently in industry commentary. A UK-based professional advisory newsletter for agencies and communications teams wrote: "Tandeep Sangra has documented self-promotion bias in LLM recommendations. Brands publishing self-referential top ten content are disproportionately surfaced when users ask AI for service recommendations."

Frequently Asked Questions

The paper is published on SSRN and ResearchGate as a working paper. As with any early-stage research on these platforms, it has not yet completed formal peer review.
If an LLM is more likely to surface a brand because that brand has published extensive self-referential "best of" content about itself, rather than because independent sources validate its quality, a buyer relying on that recommendation may be evaluating manufactured visibility rather than earned credibility. The paper's five-point framework gives buyers a way to check the difference.