AI competitor analysis is the practice of tracking, for your actual buying questions, which specific competitors ChatGPT, Claude and Perplexity name instead of you — not just whether you show up, but who shows up when you don't. Classic competitive analysis compares websites, pricing and reviews. This adds a layer neither Google Analytics nor a rank tracker can see: the shortlist an AI engine hands a buyer before they ever visit a site — and the rivals it reaches for instead of you. Here's what to actually track, why the list moves, and how to read it without overreacting to noise.
AI competitor analysis tracks three things a rank tracker never had to: which named competitors appear for each buying question, in what order, and which sources the engine leaned on to pick them. That's a different object than a keyword ranking. Ask ChatGPT, Claude or Perplexity “what's the best [your category] for [your buyer]?” and you don't get ten blue links — you get a short, opinionated list, often three to five names, sometimes with reasons attached. Every name on that list that isn't you is a competitor winning the moment you wanted. This is the sharper, more specific cousin of the brand intelligence question “how does AI describe us” — here the question is “who does it describe instead.”
Bottom line: the unit to track isn't “are we mentioned” — it's the full named list, per question, per engine, logged over time.
Ask the same engine the same question twice and the list of named competitors can come back different — sometimes very different — even with nothing about the market having changed. That's not a bug in your tracking, it's a property of how these models generate answers: they sample from a distribution rather than reading off a fixed ranking. Research from SparkToro testing repeated identical prompts found that AI models rarely return the exact same list of brands twice — matching lists in less than 1 in 100 repeated runs, and almost never in the same order even when the set of plausible candidates is small.
Proof: SparkToro — New research: AIs are highly inconsistent when recommending brands or products.
Why it matters: one screenshot of “AI recommended our competitor” is a sample, not proof of a trend — you need repeated runs before you react to it.
ChatGPT, Claude and Perplexity don't just phrase things differently — they can name a different set of competitors for the same question, because each pulls from its own mix of training data, live search results and ranking logic. A competitor who dominates Perplexity's citations because their content is fresh and well-structured may barely register in Claude's answer, which leans more on how clearly a brand's identity and claims are stated. Treating “AI” as one channel hides this — treating it as three separate shortlists to track is what makes the share-of-voice comparison meaningful.
Do this: log competitor mentions per engine, not pooled together — a rival can be dominant on one and invisible on another.
Being named third with a hedge (“X is also worth considering for smaller teams”) is a different outcome than being named first outright — and a simple presence/absence count treats them the same. The context an engine wraps around a competitor's name is often more useful than the name itself: it tells you the exact claim or use case the model associates with them, which is usually the gap you need to close, not a mystery about “brand strength.”
Rule of thumb: record the sentence around each competitor mention, not just the mention — the framing is the actionable part.
A single competitor-analysis run tells you what an engine said once; only a fixed question set, repeated on a schedule, tells you whether a rival is actually gaining ground. The same discipline that makes a share-of-voice number trustworthy applies to competitor tracking: fix the questions, fix the engines, and compare the same list to itself over time rather than comparing one day's answers to a different day's different questions. Pair it with a cadence — see how often to re-check — or the “list” you're tracking is really just noise with a date on it.
Shortcut: Reflexa's Competitors view re-runs your fixed question set across ChatGPT, Claude and Perplexity and shows which rivals are gaining or losing ground, with the exact answer text attached.
Which competitors an engine reaches for isn't fixed by brand size — content and structure measurably shift it. Academic work benchmarking generative-engine responses found that applying concrete content strategies (clearer claims, added citations, better structure) could lift a source's visibility in generated answers by up to 40%, though the effect size varied a lot by topic and domain. That's the same lever behind a competitor's rise: not that they're a bigger brand, but that their content gives the model more to work with.
Proof: arXiv — GEO: Generative Engine Optimization (Aggarwal et al., ACM SIGKDD 2024).
Good news: a competitor's lead in AI answers is a content and evidence gap, not a fixed hierarchy — which means it's a gap you can close.
No competitor tracker turns a probabilistic system into a fixed leaderboard. Even a well-run, repeated question set is sampling a moving target, and a rival can appear once without it meaning anything. What repeated, per-engine tracking buys you isn't certainty about who's “winning” — it's enough signal, over enough runs, to tell a real shift from a single noisy answer.
Sources: SparkToro — AI recommendation inconsistency research · arXiv — GEO: Generative Engine Optimization · What is Brand Intelligence?
The free check reads your competitors across ChatGPT, Claude and Perplexity — 3 minutes, evidence included.