What an AI Said About Your Competitor That It Won't Say About You

There's a moment I see almost every week now in the Lighthouse dashboard, and I want to describe it to you because I think it explains something about what AI is doing to brand marketing that nobody on your team has put a name on yet.

A category. Let's say it's electric toothbrushes, because I just ran this one. Five brands. All roughly the same market position. All running healthy marketing programs. All visible on Google for the obvious queries.

I ask ChatGPT, Claude, Gemini, and Perplexity the same question across all four models: I'm looking for a new electric toothbrush. What do you recommend.

Four of the five brands get mentioned. One does not. The one that does not is the one with the second-largest market share. The one that gets mentioned first across three of four models is fourth in market share, but happens to have a Wirecutter review from eighteen months ago and an unusually rich Reddit presence.

The brand that didn't get mentioned has more billboards, more TV, more performance marketing, and more revenue than two of the brands that did. None of that mattered.

This is the part where most people stop reading because they think this is a problem you fix with better SEO. It is not.

Here's what's actually happening, and why the fix is harder than it looks.

When an AI model is asked a recommendation question, it doesn't run a search the way Google does. It synthesizes from its training data and, depending on the model, augments with real-time retrieval from a small set of sources it has learned to trust. Across the four major consumer-facing models, those sources are dominated by a few specific surfaces: high-authority editorial review sites (Wirecutter, NYT, The Verge, CNET), a small number of category-specific publications, Reddit, sometimes TrustPilot, sometimes the brand's own website if the page is structured to answer the question being asked.

What is not in that set: paid media. Performance ads. Most influencer content. Most affiliate content. The TV spot you ran in Q3. The billboard on I-94. The agency-of-record case study about your beautiful brand campaign.

The model is not anti-brand. The model is anti-ad. Years of training on internet text taught it that the language of advertising is unreliable for answering a recommendation question, so it doesn't weight it. Your media buy is invisible in the place where the consideration set now gets formed.

This is not theoretical. The data we see in Lighthouse confirms it across every consumer category we track. Brands with strong earned media presence outperform brands with strong paid presence in AI recommendation queries by a wide margin, and the gap is widening every quarter.

The work, if you want to win this, is different from the work you've been doing. It looks more like PR than performance marketing. It looks more like community engagement than campaign deployment. It looks like making sure that when an editor at a major publication writes the roundup of your category next year, your brand is the one they reach out to, because the model trained on that roundup is the model that will be answering the question on the day your customer is making the decision.

This is the work That Random Agency and Lighthouse are doing right now with brands that have realized the shift. It's slower than performance marketing. It's harder to measure in the quarter you spend the money. It compounds over time in a way performance marketing does not.

If you want to know what an AI says about your brand today — not what you hope it says, what it actually says when a real consumer asks a real question — the audit takes two weeks. We do it across the four major models and the most relevant category queries. The output is a report that lands somewhere between "useful intelligence" and "uncomfortable mirror."

Contact form is the fastest way. Tell me your category and a couple of your direct competitors and we'll start there.

‍