Why ChatGPT Recommends Your Competitor as “The Alternative” (Not You)
Somewhere this week, a potential customer typed something like “best alternatives to [your biggest competitor]” into ChatGPT, Perplexity, or Gemini’s AI Overview. The answer named three or four brands.
If yours wasn’t one of them, this article is about why — and it isn’t the reason most agencies will tell you.
We didn’t guess at this. We ran the same category of query — “best alternatives to [incumbent]” — across five different e-commerce brands (Warby Parker, Glossier, Chewy, Bombas, Tiffany & Co.), logged every citation pattern, checked llms.txt presence, and cross-referenced against Google’s own AI Overview and Reddit consensus for the same queries. What we found breaks the two assumptions most SaaS teams are currently operating on.
Assumption 1: “If we just publish more comparison content, AI will start citing us.”
Assumption 2: “If we implement llms.txt, we’re covered.”
Both are incomplete. Here’s what actually determines whether you get named — and what it means if you’re a SaaS company building comparison pages right now.
There isn’t one “AI citation algorithm” — there are at least five different shapes
Every category we tested produced a structurally different answer format. Not a different ranking — a different shape of response. That matters more than most people realize, because it means the tactic that gets your competitor cited in one category can be irrelevant in yours.
Shape 1 — Single winner gets the deep-dive. In eyewear, one competitor (not the household name) got an entire expanded section with images and a founder-style narrative, while the more famous brand got a single line in a comparison table. The differentiator wasn’t domain authority or brand size — it was fresh, unambiguous, positively-framed third-party consensus. Recent “best of” roundups that ranked the winner clearly, without hedging. The bigger brand’s mentions existed too, but they were framed as debate, not endorsement. AI engines appear to read tone, not just presence.
Shape 2 — Comprehensive decision matrix, no single winner. In categories without a “prestige roundup” culture — where nobody publishes breathless “12 Best X Brands in 2026” listicles because the category is functional, not aspirational — the AI engine builds its own full comparison instead: every major competitor gets a profile, and the answer closes with “which one is right for you” segmented by use case.
This is the shape SaaS almost always falls into. Buyers don’t write “10 Most Prestigious CRMs” the way they write “10 Best Eyewear Brands.” SaaS comparison content is judged, discovered, decision-support content — which means your positioning has to win on segmentation and functional clarity, not vibes.
Shape 3 — Segmented by use case, sourced from Wikipedia-grade authority. In categories with decades of stable consensus, AI engines default to the most established, cross-validated sources — in one case, literally pulling brand bios from Wikipedia inside the chat interface. Newer entrants had almost no path into that shape regardless of content quality.
Shape 4 — Community-consensus callout, no imagery, no single winner. In lower-glamour categories with strong but unglamorous community discussion (think: durability, quality, “does this actually work” threads), the AI engine explicitly cited Reddit-style consensus by name, then moved on without crowning anyone.
The throughline: the shape of the answer is decided by the shape of the evidence available off-site. Your own website content is necessary but not sufficient. AI engines are pattern-matching against what the internet has already agreed on about your category.
The llms.txt finding that surprises almost everyone
We expected llms.txt to be the deciding factor in whether a brand’s own domain got cited cleanly, versus getting buried under third-party aggregators. It wasn’t — a finding that builds directly on our earlier teardown of the llms.txt myth (https://alneeko.com/llms-txt-myth-chatgpt-citation-teardown/).
What actually decided it: whether a clear, crawlable page existed that answered the exact query being asked.
One brand had a well-built llms.txt file — but it didn’t cover returns, so when the query was about returns, the AI engine cited third-party aggregator sites instead of the brand’s own domain, alongside a partial own-domain reference. Another brand’s llms.txt file did explicitly cover returns, refunds, exchanges, and international policy in granular detail — and got a single, fully clean, own-domain citation with zero third-party noise. Two brands had no llms.txt file at all and still got clean own-domain citations, because a single well-structured page on their site directly answered the question being asked.
The file isn’t the moat. Coverage is the moat.
What this means for your comparison pages specifically
- A comparison table up top, with entity names as clean, hyperlinked anchors.
- Segmented “best for” recommendations, not a single blanket winner.
- A page that fully answers the adjacent questions — pricing, switching process, data migration, support.
- Structured data that matches what’s actually on the page — Product, Review, FAQ, and Comparison schema.
And a note on third-party mentions: the strongest citation pattern we found wasn’t manufactured — it was existing, fresh, positively-framed consensus. We’ve written before about why Reddit specifically carries this much weight in AI citations (https://alneeko.com/reddit-citation-paradox-chatgpt-aeo/) — the short version is that a handful of forced, spammy mentions will do less for you than a handful of real ones.
The part most agencies can’t show you
That’s the piece we treat as non-negotiable: every comparison page, every schema implementation, every third-party mention gets checked against real prompts in Google SGE, Gemini, and Perplexity — not assumed, tested.
If you’re a SaaS founder or growth lead trying to make sure your platform — not your incumbent — is the one AI engines recommend, see how we approach AI visibility work (https://alneeko.com/pricing/) or get in touch to talk through what your category’s citation shape probably looks like.
