LLM authority is an evolving ecosystem. Build a better GEO strategy!

One of the biggest mistakes in GEO today is building strategy around the assumption that LLM authority sources will stay the same.

Nothing is permanent in AI discovery!

Between August 8th and August 14th, Reddit’s share of ChatGPT Search citations reportedly fell from 3.83% to 0.52%—an 86% relative decline in less than a week, according to Promptwatch data.

No one should be surprised by this decline. Reddit was always going to be an important source for LLMs, but it was never going to be a perpetual authority across every type of query.

As LLMs become better at evaluating the usefulness of the sources behind their responses, they will naturally shift toward sources that more directly, accurately, and authoritatively answer the real human’s underlying need. Reddit will remain valuable for opinions, experiences, comparisons, and community perspectives, but it is less effective when a prompt demands definitive product information, technical expertise, documentation, or trusted factual guidance.

The shift goes well beyond Reddit.

Qwairy, which monitors millions of citations, verified the drop against its own data and found that discussion platforms, directories and press sources lost share across the board, while official and authoritative domains gained it. 

In traditional SEO, we would simply call this an algorithm update.

But in GEO we are witnessing the continuously refactoring of LLMs and how they retrieve, weight, and prioritize sources as they get better at determining which information actually helps satisfy a user’s need.

That is an important distinction. The takeaway isn’t that Reddit suddenly became irrelevant. It’s that the models are becoming more selective about when Reddit, a directory, a publisher, or any other source is actually useful for a specific prompt.

What makes this even more interesting is that the decline isn’t isolated to ChatGPT. We’re seeing a similar directional shift in Google’s AI products, including AI Mode.

When multiple independent models begin reducing their reliance on similar categories of sources, that becomes a meaningful signal. It suggests these systems are independently arriving at similar conclusions about which source types are most useful for the prompts and tasks users, and increasingly agents, are asking them to complete.

Where does Reddit fit now?

Reddit still matters. It just matters in a much narrower and more specific way than the visibility it has enjoyed across search and LLMs over the last few years might suggest. In fact, we wrote about this a few years back as we notice the rise of Reddit within Google search and called out the lack of relevance during this surge.

Its real value is in moments where consumers are actively debating a decision and need context that is difficult to identify.

Think about fast-moving consumer trends, emerging products, breaking news, niche hobbies, highly specialized verticals, or situations where someone wants to understand how other people actually experienced something. In those cases, the community layer is the value.

Reddit offers an enormous archive of human conversation and experience, but its communities are also intentionally structured, moderated, and insulated. That makes them incredibly valuable in some situations while making them a poor representation of the broader real-world customer journey in others.

For GEO/SEO teams, the strategy should not simply be to “get onto Reddit.” It should be to understand which prompts, categories, and customer decisions genuinely benefit from community-driven context and invest accordingly. In some cases, that may mean not investing at all, particularly for brands that don’t naturally fit these types of discussions or specialized communities.

For GEO Strategy, this is incredibly useful.

These shifts give us a clearer way to prioritize where we invest. Instead of treating citation sources as static targets, we need to monitor how models are re-weighting the information ecosystem and adjust our strategies accordingly.

I believe this will become a major part of GEO optimization: tracking changes in source preference, understanding why certain categories are gaining or losing influence, and reallocating effort toward the sources that models increasingly trust to satisfy user intent.

In many ways, we are watching the emergence of a new kind of algorithm monitoring, only this time, it is happening across multiple LLMs at once. Giving us a faster view into the evolution of human search behavior.