This article originally appeared on Forbes.

‍

There's a familiar refrain on LinkedIn right now: Trust is the new battleground in AI-driven shopping, so logically brands should seed more content into Reddit threads.

It's not wrong, exactly, but it's aimed at the wrong target based on a deeper misunderstanding of how large language models decide who to trust.

AI models aren't loyal to Reddit, but they are starved of alternatives.

When ChatGPT or Perplexity leans on Reddit for a recommendation query, it's because Reddit is one of the few large, easily parsed repositories of unfiltered, first-person "I actually used this" content available. People have long appended "Reddit" to their searches to find real opinions instead of SEO-optimized content. AI systems trained on that behavior show the same pattern when a question calls for opinion rather than fact.

An analysis of roughly 30 million cited sources by AI search analytics tool Peec AI found Reddit, YouTube and LinkedIn dominate citations precisely because they carry real user discussion, not brand messaging.

The model isn't asking "What does Reddit think?" Instead, it's asking: "Where can I find a real person describing what actually happened?" Reddit is just the largest, messiest answer at internet scale, a proxy for trust rather than the source of it.

Brands chasing Reddit placement are often optimizing for the symptom instead of the cause.

Why Trust Became The Whole Game

Unlike traditional search, where search engines rank pages, generative search engines synthesize an answer and put their own credibility behind it.

A traditional search engine could show 10 blue links, and let readers decide.

A chat-based answer engine has to pick, and every brand claim it repeats without independent backing is a claim it's personally vouching for. A product page calling a jacket "the warmest on the market" has an obvious incentive problem. A hiking forum thread describing a Colorado winter in it does not.

Models are increasingly built to weigh the second kind of content more heavily.

The Framework Marketers Cite Without Understanding

SEO specialists have discussed Google's experience, expertise, authoritativeness and trustworthiness (E-E-A-T) framework for years, but it's taken on new relevance for AI because it's also the clearest public blueprint for what "trustworthy" means to a machine.

E-E-A-T comes from Google's Search Quality Rater Guidelines, and Google has said the guidelines illuminate the standard its automated systems are tuned toward.

"Experience" matters most: Raters weigh whether a reviewer actually used the product they're describing—a hiking boot review benefits from someone who's hiked in the boots.

Trust overrides the rest, since content can show experience, expertise and authority and still fail if it isn't accurate. Generative AI wasn't trained on E-E-A-T as a rule book, but it's absorbed the same logic. The question of who to believe sits under both search and AI answers.

The Ghost Citation Problem

Analytics firm Seer Interactive coined "ghost citation" after studying more than half a million LLM responses across 20 brands: cases where a brand's own content is cited as the source, but a competitor gets named as the recommendation.

Their tests point to a structural explanation: The model decides which brand to recommend first, drawing on what it already "knows" from training, then goes looking for a source URL.

Meanwhile, a parallel study by Semrush of nearly 4,000 domain appearances found 62% of AI citations carry no brand mention at all—weak "entity-level" presence in training data made this common, while a strong public identity approached a near-zero rate in some categories.

The implication of these studies of AI search: A brand can do everything right on its own site—clean data, sharp copy, perfect schema—and still be invisible in the moment that matters, because citation and recommendation aren't the same system.

Why AI Hedges

This is the mechanism behind a pattern anyone using AI shopping assistants has noticed: the hedging. Ask an AI engine for a recommendation and it often qualifies, caveats and offers several options rather than commits—a rational response to an evidence problem, not a stylistic tic.

When a model can't find independent accounts to back a claim, refusing to commit is the safer output; confidence tracks the density of verification behind it. Brands supplying only their own marketing copy hand the model exactly the self-interested claim it's trained to discount.

The Content Brands Already Have

Here's the encouraging part: Most brands aren't short on the raw material that builds machine trust. They're sitting on years of reviews, unboxing videos, community threads and creator content.

This is exactly the "experience" layer E-E-A-T asks for, and exactly what generative engines are built to weigh over brand copy. Nielsen research has long shown people trust word-of-mouth over brand advertising, and that dynamic now earns machine trust for structural, not sentimental, reasons.

The throughline across all of this is to focus on trust, not visibility. Being found and being cited are necessary, but neither one gets a brand recommended, because a generative engine will not put its own credibility behind a claim it cannot independently verify.

Making the claim, however polished, is no longer the hard part. Earning the kind of corroborated, third-party evidence a model is willing to stake its answer on is. Brands should treat trust as something to be demonstrated by others, not asserted by themselves.

‍