Your content feeds AI answers that promote rivals
Being cited in an AI answer and being recommended by it are two different outcomes. Most enterprise content strategies are built for the wrong one.
Key takeaways
- AI systems regularly cite a brand's comparison content while recommending a rival named within it.
- Citation counts and recommendation rates are separate metrics; most enterprise content strategies only optimise for the first.
- Brands appearing consistently as named answers across many sources win AI recommendations, not brands that produce the most content.
- Comparison guides, vendor roundups, and 'best of' lists are structurally likely to boost competitors' AI visibility.
- Auditing which pages AI systems cite, and who those pages recommend, reveals the real cost of category content libraries.
Search Engine Journal's Lily Ray has identified a structural flaw in how most brands approach AI search: the content marketers write to demonstrate authority is routinely used to endorse their competitors.
The mechanism is specific and uncomfortable. Google's AI Overviews and similar LLM-powered answer engines frequently cite a brand's own "best of" listicle, product comparison page, or category roundup as a source, then use that content to recommend a rival. The citing brand provides the credibility. A competitor gets the recommendation. The content author funds the loss.
This is not an edge case. Ray's data shows it is a systematic pattern in AI-generated responses, particularly in categories where one brand has invested heavily in comparison and educational content. The more comprehensive your coverage of alternatives, the more useful your page is to a model assembling a balanced answer, and the more likely that answer names someone else.
The citation/recommendation split
SEO doctrine held, for a long time, that citations were the goal. Rank for the query, earn the click. That logic no longer holds in a world where the answer engine reads your page and composes its own response. A citation now means your content was consumed; it does not mean your brand was chosen.
The distinction matters most for enterprises that have spent years building what the industry called "thought leadership content": authoritative category pages, vendor comparison guides, buyer's checklists. This content was designed to attract organic traffic by ranking on informational queries. It is now, in many cases, training material for answers that end with a competitor's name.
For a multilateral institution or a large industrial group, the risk is somewhat different but structurally identical. An organisation like ISO or a major standards body publishes exhaustive, neutrally framed comparisons of approaches, frameworks, or member organisations. If those pages are structured as lists with named entities, AI systems will extract the entities and surface the most frequently co-cited ones as recommendations. The publishing institution becomes a data source, not an authority.
Financial services brands face a sharper version of the problem. A bank's mortgage comparison tool, a fund manager's ETF explainer, or an insurer's policy guide are all vulnerable: the AI reads the structured comparison, identifies the options, and recommends whichever one its training data or citation graph most strongly associates with positive outcomes. The brand that built the guide may not even appear in the final answer.
What the model actually rewards
The underlying issue is that generative AI systems optimise for answering the user's question, not for crediting the source. When a model reads a page that says "here are the top five providers in this category," it learns the five providers. It does not learn to prefer the publisher. The model's recommendation reflects the cumulative weight of citations across its entire training corpus, not the one page it just read.
This creates a counterintuitive content strategy problem. The brands winning in AI recommendations are not necessarily the brands producing the most content. They are the brands appearing most consistently as the named answer across many different sources. Frequency and cross-source consistency matter more than authorship.
The practical implication is structural. Brands that write "X is one of the best options" in their own content are adding to the citation count for X, regardless of whether X is themselves. Every piece of comparison content that names a rival, even to position that rival as second-best, reinforces that rival's presence in the model's answer pool.
Ray's analysis points toward a reorientation that most enterprise content teams have not yet made: optimising for recommendation rather than citation. That means fewer comprehensive roundups and more content designed to resolve a specific decision in a specific context, with the brand positioned as the answer to a precisely framed question, not as the impartial curator of all available answers.
For brands that have already built large content libraries organised around comparison and category coverage, the inventory itself is now a liability. The content is being read. The wrong names are being recommended. Auditing which pages are being cited by AI systems, then examining who those pages recommend, is not optional housekeeping. It tells you exactly how much of your content budget has been spent building someone else's AI visibility.