The Platform Trap, Again
Reddit accounts for 46.7% of Perplexity's commercial query citations. Most brands have no idea what that means for their AI positioning.
Tim de Rosen | CEO, AIVO, Inc. | May 2026
In 2014, Facebook changed its algorithm. Organic reach for brand pages collapsed from around 16% to under 2%. Brands that had spent years building Facebook communities discovered, in a matter of weeks, that the asset was never theirs. They had been building on rented land.
The AI industry is making the same mistake.
A new AIVO working paper published today on Zenodo documents the scale of Reddit's concentration in LLM citation pools, and what that concentration means for brands managing AI purchase recommendation exposure. The findings are not marginal. For the query category that matters most to commercial outcomes, conversational and recommendation prompts, Reddit accounts for 46.7% of Perplexity's top-cited sources. Its nearest competitor is YouTube at 13.9%. Reddit's share is more than three times that of its next most-cited source.
This is not a visibility problem. It is a resilience problem. And the measurement frameworks most brands are currently using cannot see it.
What the data shows
The 46.7% figure comes from Profound's analysis of 680 million citations across ChatGPT, Google AI Overviews, and Perplexity, conducted between August 2024 and June 2025. It is independently confirmed by Discovered Labs' citation pattern research published in December 2025. Two separate research bodies, two separate methodologies, the same number.
The overall citation share tells a different story. Across all query types globally, Reddit's share of Perplexity's total citations sits at 6.6%. That figure is accurate and it is also misleading for brand strategists. The queries that drive AI-assisted purchase decisions are not evenly distributed across all query types. They are concentrated in the conversational, opinion-seeking, and recommendation category. In that category, Reddit is not one source among many. It is the dominant source by a factor of three.
For brands operating in consumer goods, beauty, nutrition, financial services, and technology, where Perplexity is a meaningful purchase discovery channel, this concentration is a material strategic fact.
Three ways this goes wrong
The first risk is legal and commercial friction. Reddit has pursued aggressive data licensing positions with AI companies. Disputes over content access and scraping agreements can produce rapid, non-signalled changes to how LLM platforms index and weight Reddit content. A brand whose positive AI citation profile is substantially Reddit-derived can experience citation degradation overnight, with no change in its own content or strategy, and no advance warning from standard monitoring data.
The second risk is automated moderation vulnerability. Reddit's moderation infrastructure operates on automated systems that respond to reporting patterns. Coordinated reporting campaigns, whether from competitors or from organised communities hostile to a brand or category, can trigger bans on subreddits and accounts without human review. The content quality is irrelevant. The reporting pattern is what triggers the action. A brand that has built a carefully constructed Reddit presence can lose its citation foundation through a process that has nothing to do with its own behaviour.
The third risk is the one that sits underneath both of the above: invisibility. Standard AEO and GEO measurement frameworks track citation volume, mention frequency, and share of voice. A brand with a high Reddit-driven citation share looks identical to a brand with a diversified citation base on every standard metric, right up until the moment of disruption. The metrics provide no leading indicator of concentration risk. You do not know the problem exists until the signal is already gone.
Citation is not selection
There is a deeper issue that the concentration problem sits inside.
The dominant approach to brand AI performance measurement treats citation volume as a proxy for brand health. The logic is intuitive: if an LLM cites your brand more often, your brand is better positioned. This logic is wrong.
Citation is a necessary but not sufficient condition for brand selection. A brand that is cited in an LLM response may still be eliminated at the final selection turn, the moment when a consumer asks an AI system to make a recommendation. AIVO's measurement of Conversational Survival Rate, the proportion of structured purchase recommendation probes in which a brand survives to final selection, consistently shows brands with high citation volume and poor selection outcomes. They are visible. They are not chosen.
The Reddit concentration problem compounds this. A brand with a Reddit-dependent citation foundation is not just exposed to platform volatility. It is also building visibility on a source whose contribution to final selection is not measured by any standard framework. Being cited in a Reddit thread and surviving to a purchase recommendation are different things, and the distance between them is where most brand AI investment is currently disappearing.
What resilience looks like
AIVO Meridian measures brand AI performance across five PSOS dimensions: breadth, depth, resilience, sentiment, and decay. Resilience specifically measures the stability of a brand's citation foundation across model outputs, query formulations, and source diversity. Weak resilience is a leading indicator of Conversational Survival Rate degradation. It appears in the data before it appears in citation volume metrics. It is the signal that a brand's AI positioning is exposed before the exposure becomes visible.
A brand with strong PSOS resilience has a citation foundation distributed across source types that are not correlated in their volatility. Peer-reviewed research deposits, brand-owned evidence pages, editorial third-party coverage, and press release infrastructure each operate on different mechanisms and are not simultaneously vulnerable to the same disruption event. This is not a theoretical architecture. It is the structure that AIVO uses to build citation foundations for clients across consumer and enterprise categories.
The appropriate response to Reddit concentration is not to leave Reddit. Reddit is a significant signal source for LLM platforms and will remain so. The appropriate response is to measure concentration, understand where it sits relative to category peers, build diversified citation foundations, and track resilience as a leading indicator rather than waiting for volume metrics to show the problem after it has already occurred.
The asset that matters
The Facebook pages collapse was painful for the brands that lived through it. What made it painful was not the algorithm change itself. It was the discovery that something they had invested heavily in, something that appeared on dashboards as a growing asset, was not actually theirs to keep.
The AI layer has the same structural risk, at higher stakes, in a less visible location. The selection moment inside an LLM recommendation sequence is where purchase decisions are increasingly being made. It is not on a feed anyone can see. It is not measured by any standard tool. And it is contingent, in ways that brands do not yet fully appreciate, on citation foundations that can disappear.
The platforms will keep changing. The selection moment is the only asset worth protecting.
AIVO Working Paper WP-2026-14: "Reddit, Citation Instability, and the AI Purchase Recommendation Layer" is published open access on Zenodo. DOI: 10.5281/zenodo.20269711
Tim de Rosen is CEO and Co-Founder of AIVO, Inc. AIVO Meridian measures Conversational Survival Rate across AI purchase recommendation sequences. aivomeridian.com