You Are Writing the AI Answer That Recommends Your Competitor
Vendors author 77% of the content AI engines cite for software recommendations. But when AI cites your comparison page, it recommends your competitor 69% of the time. Here is what the data actually says and what founders should do about it.
Most B2B software companies are publishing "best of" listicles that rank themselves first. AI engines cite those pages as sources. Then they recommend a competitor from the same list. Broadcastwell's 2026 State of GEO study found that 77.4% of the most-cited content in AI search comes from vendors writing about themselves. The problem: that content is being used against them.
When AI Cites Your Page but Recommends Someone Else
BusinessTechWeekly analyzed 80 commercial queries that produced an AI Overview in Google. Self-ranked listicles were cited 323 times. In 224 of those cases, Google named the brand's own page as a source, then recommended a competitor listed inside that page.
That is a 69% backfire rate.
One example from the study: for "best LMS for selling courses," Google repeatedly cited Oasis LMS, a brand that ranks itself number one in its own comparison article. The AI Overview recommended Kajabi, Thinkific, LearnWorlds, and Teachable instead.
The pattern is consistent. You spend months producing a comparison page. You rank yourself at the top. You get the citation. Then the AI reads your page, extracts the competitor list you helpfully laid out, and uses it to build a shortlist where your competitors get the recommendation.
I call this the citation-recommendation gap. Getting cited is not the same as getting recommended.
Vendors Are Writing 77% of What AI Reads
The Broadcastwell State of GEO study ran 10 standardized buyer questions through AI search for each of 85 B2B software companies across 61 categories. That produced 860 scored answers and 5,160 traceable source citations.
When they classified the 100 most-cited domains, the breakdown was stark:
| Source Type | Share of Top-100 Citations |
|---|---|
| Vendor-authored content | 77.4% |
| Review platforms (G2, Capterra, TrustRadius) | 10.2% |
| Independent media | 6.5% |
| Analyst firms (Gartner led with 110 citations) | 5.9% |
| Community (Reddit, Quora, Stack Overflow) | 0.0% |
Read that again. More than three quarters of what AI cites when recommending software was written by the vendors themselves. Comparison posts, category guides, "best X software" listicles. Companies are writing the category answer for AI, and AI is using it.
But here is the part everyone misses: being cited as a source is not the same as being named as the answer. The Derivatex study found that a vendor's own website earned just 12% of the citations in AI Overview responses for software queries. Third-party "best of" lists earned 63%. They called it the 12% Problem.
Your content feeds the machine. It does not control what the machine says about you.
Why the AI Trusts Everyone Except You
Angelfish Marketing studied 1,052,053 AI citations. Across the answers that mention a brand, the brand's own site is cited as a source only 13% to 32% of the time. The rest comes from competitors, community platforms, review sites, and trade press.
Their head of growth put it plainly: "Marketers assume that if AI is talking about them, it's reading their website to do it. It's reading what everyone else has said about you."
This is the structural reality. AI engines treat your "best X" page as a data source, not as a recommendation engine. They extract the structured information, the feature lists, the competitor names, and then they apply their own ranking logic based on the signals they trust most: third-party review volume, editorial validation, and community consensus.
The G2 2026 Buyer Behavior Report confirms the demand side. Review sites at 38% just overtook AI chatbots at 37% as the top source shaping which vendors make a buyer's shortlist. And 69% of buyers chose a different vendor than they originally planned because of what an AI chatbot recommended.
So the buyer asks AI. AI reads your comparison page. AI recommends your competitor. The buyer trusts the recommendation because it came from what appears to be an independent source. But you wrote it.
What Founders Should Do Instead of Writing Self-Ranked Listicles
The data points to a clear hierarchy of what actually earns AI recommendations, not just citations.
Build third-party proof first. G2's Answer Economy data shows 85% of buyers view vendors more favorably when AI includes them. But inclusion is driven by third-party signals, not self-attestation. G2 category grid inclusion typically requires at least 10 reviews. Leader status takes 25 or more with a rating above 4.0. If your G2 profile has sparse or outdated reviews, you are not in the structured data AI reads for your category.
Stop writing "best X" content that lists your competitors. Every competitor you name in a comparison page becomes a candidate the AI can recommend instead of you. If you must publish comparison content, structure it around a specific buyer problem where your product has a documented advantage, not around a category grid that invites substitution.
Invest in the sources AI actually trusts for recommendations. The Gracker Trusted-Source Map breaks down each engine's preferences: ChatGPT favors Wikipedia (41.2% of citations), Reddit (34.7%), and G2. Perplexity emphasizes Reddit, LinkedIn, and G2 for B2B queries. Claude favors structured editorial and LinkedIn long-form. The G2 network (including Capterra, Software Advice, and GetApp) accounts for roughly 12.7% of citations in bottom-of-funnel B2B queries.
Run the test right now. Open ChatGPT, Perplexity, and Claude. Type your category query the way a buyer would: "best [your category] for [your ICP]." Note where you appear, what position, what the AI says about you, and what source it cites. If you are not in the top three, note who is. Then check whether the cited source is a page you wrote.
The Machine Relations Lens
This is what Machine Relations predicts. AI engines are corroboration machines. They do not trust what you say about yourself. They trust what independent sources say about you, weighted by the consistency and recency of the signal.
When you write a comparison page, you are creating raw material for the corroboration engine. But you are not the corroborator. You are the data supply. The recommendation goes to whoever has the strongest independent signal across the sources the AI trusts.
The winners in this system are the companies that invest in the signal layer: G2 reviews, editorial placements in trade publications, earned media, community presence. Not the ones producing the most vendor-authored content. The irony is that the vendor-authored content is feeding the machine that sends buyers elsewhere.
FAQ
Does publishing a "best X" listicle actually hurt my AI visibility?
It can. The BusinessTechWeekly study found a 69% rate of AI citing a vendor's listicle while recommending a competitor from that page. The listicle gives the AI a structured competitor grid to work with. If your third-party validation signals are weaker than a competitor listed on your own page, the AI will use your page as the citation and recommend them as the answer. The risk is not theoretical. It is measured.
What is the single most important thing I can do to improve AI recommendations for my brand?
Build review volume and recency on G2. The AmICited study found that g2.com leads all review platforms with 79 ChatGPT citing responses, ahead of Capterra at 55. G2's own research shows 45% of software buyers say a citation from a review site is the single most confidence-inspiring signal in an AI answer. Start there before producing more vendor-authored comparison content.
How do I know if my comparison content is being used to recommend competitors?
Run your top category queries through ChatGPT, Perplexity, Claude, and Google AI Mode. Check each response for two things: which sources are cited, and which vendors are recommended. If your domain appears as a cited source but your product is not in the recommendation, you have a citation-recommendation gap. The fix is not more content. The fix is stronger independent proof signals that outweigh the competitor signals on your own page.