Skip to content
Fast Company Executive Board

The Fast Company Executive Board is a private, fee-based network of influential leaders, experts, executives, and entrepreneurs who share their insights with our audience.

An LLM window closes soon

Where will your brand be in AI search category rank?

[Photo: Getty Images]

Profile photo of Andrea Lechner-Becker

Chief Strategy Officer · GNW Consulting

The marketing industry has been told that a “window is closing” approximately every 18 months for the last 15 years. Social media in 2008. Mobile in 2012. Programmatic in 2015. Voice search in 2018. The metaverse in 2022. Each one came with a column much like this one, urging marketing leaders to move before the door shut.

Most of those doors didn’t shut. The category matured, the early movers gained a real but rarely existential advantage, and the rest of the market caught up at its own pace. So, when I tell you the window to establish your brand in AI search is closing, you have every reason to be skeptical.

Here’s why I think this one is different. Alongside Demand Metric, I co-authored The State of GEO in B2B Marketing. We surveyed 225 business-to-business (B2B) marketing and revenue leaders about their organizations’ GEO efforts.

Notably, 22% of respondents reported that AI-driven sources already account for more than 5% of their website traffic. Yet the widely cited industry benchmark is under 1%. My concerns are what help explain this gap.

How LLM citations shift the dynamics

A path-dependency dynamic is already producing visible traffic effects, not just theoretical ones.

In the previous era, Google search rankings were dynamic, as the algorithm continuously re-evaluates your page against your competitors’. A better backlink, a fresher article, a faster site—any of these can move your ranking on a timescale of days or weeks. Falling behind was recoverable.

Large language models (LLMs) don’t work this way. The associations they build between categories and brands form when the model reads a vast corpus of text and learns which brand names co-occur with which category descriptions.

Those associations are encoded in the model’s weights. They are then reinforced—or, in retrieval-augmented systems like Perplexity and Google’s AI Overviews, supplemented—by a retrieval layer that selects which current sources to include in any given answer.

Both layers favor incumbents. The pre-training corpus reflects which brands had the most coverage across the most credible sources during the training window. The retrieval layer prefers sources it has cited before, which tend to be those that have been cited frequently historically.

The ramifications of AI

Once your brand is in the cited LLM set, you stay in the cited set. Once it isn’t, getting in requires either waiting for the next model retrain (which is rare, expensive, and out of your control) or flooding in new corroborating sources at a pace that meaningfully rebalances retrieval. That second option is possible, but not cheap or fast.

There is also no consolation prize, as with Google search results. A Google page has 10 organic positions, plus ads, a featured snippet, People Also Ask, images, and video carousels. Even being on page two of Google captures some user traffic.

An LLM answer typically cites three to seven sources. If you are not in those three to seven, you are nowhere. The user does not scroll. There is no page two.

Our study demonstrated that in a meaningful share of B2B categories, brands that appear in AI-generated answers are receiving referred visitors. Brands that aren’t showing up are getting nothing. And, as noted, the absence from the retrieval set is the harder problem to solve later.

The companies investing in GEO right now are establishing the citation patterns that will serve as defaults for the next generation of buyers. Two years from now, when a user asks ChatGPT to recommend a vendor in your category, the model’s answer will be built from associations formed this year.

What we still don’t know

We don’t yet know exactly how sticky these associations are. Models get retrained, and new versions reshuffle some retrieval patterns.

GEO tactics are emerging that may lower the cost of displacing incumbents. This could include flooding Reddit threads with consistent positioning, building structured citation chains across YouTube and peer-review sites, or getting featured in the trade press that LLMs treat as authoritative. Successful companies that achieve displacement haven’t published their playbooks yet, and the practitioners pushing hardest in GEO disagree about which approaches matter most.

We do know that the cost of being a late entrant is rising faster than that of being an early entrant. Brands that establish themselves in the retrieval sets now will pay less to defend that position than late entrants will pay to enter it. That asymmetry is path dependency, and it is real even if the exact half-life of an LLM association remains uncertain.

What this means for marketing leaders

The implication is that the calculus for “should we invest in this now or wait until it’s clearer” has flipped.

For most emerging marketing channels, waiting is rational. You let early movers absorb the cost of experimentation, watch what works, and you enter once the tactics are validated. For AI search, waiting is the expensive option. Because the cost of entry rises with the category’s maturity, and the cost of being absent compounds in the background.

The brands that win the next five years of B2B discovery are being decided right now, in conversations happening inside an LLM model that read about your category last year and will read about it again next year. The brands the model remembers will be the ones doing the work to be remembered.

Andrea Lechner-Becker is the chief strategy officer at GNW Consulting.