Why ChatGPT Recommends Your Competitor

You ask ChatGPT who's good at what you do, in your area, and it names a competitor. Not a vague "there are several options," a specific name, confidently stated. You check, and your business isn't a mystery to anyone: real website, real reviews, real history. So why them and not you?

If that's genuinely your situation, both of you clearly real, clearly legitimate, clearly in business, this isn't an entity-recognition problem. That diagnosis belongs to a different, more common case: a business ChatGPT can't confidently identify at all, vague, inconsistent, or unconfirmed anywhere independent. If you're not sure which situation you're in, our recognition guide walks through that self-audit first, ask ChatGPT what your business is and what it's known for, and see whether it hedges or answers specifically. This post picks up from the other side of that line: you both clear the recognition bar, so what actually decides which name comes out of the model?

It's Rarely One Competition

The instinct is to treat this like a single contest you lost: somehow your competitor is simply "better" at this than you. The more accurate picture, especially on systems that use query fan-out to answer a complex question, is that a single recommendation is often assembled from several smaller questions, each possibly checked against a different source. Your competitor doesn't need to beat you everywhere. They only need to win whichever specific branch the system happened to check that time, which is exactly why asking the same question twice can sometimes surface a different name.

That reframes the real question. It's not "why are they better than me," which is a hard, vague, demoralizing question. It's "which specific sources are currently backing them up, that aren't currently backing me up." That's answerable, and the data on what those sources actually look like is more specific, and more useful, than most advice on this topic lets on.

What's Actually Feeding the Model

Our own brand visibility guide already makes the general point that a lot of what engines cite comes from off your own site. What follows goes a layer deeper than that: a dedicated, much larger study of exactly which sources, and why one brand's mix looks nothing like another's.

Muck Rack, a media-monitoring company, ran the largest study of its kind on this exact question: what AI models actually cite, not what marketers assume they cite. Its December 2025 report analyzed one million links pulled from real responses across ChatGPT, Claude, Gemini and Perplexity, collected between July and December 2025. Four findings from it matter directly for the question of why a competitor gets named and you don't.

Each brand's citations come from its own small, specific set of sources, not a shared list. The report's own framing is direct: "the majority of work can be accomplished by targeting only 20 outlets. The catch is that there isn't a magic list of outlets for everyone. Each brand has its own mix of outlets driving their citations." Your competitor isn't winning because they cracked some universal formula. They're winning because a specific, small set of sources, probably ones you've never looked at, happen to mention them in a way the model trusts.

The sources that actually matter are often not the ones a PR effort targets. The report compared the journalists brands pitch against the journalists AI models actually cite for them, using its own platform's pitching data. The overlap was "only two percent, on average." That's a genuinely surprising gap, and it means the usual advice, "get more press," can completely miss the actual lever. If your competitor is being cited by a niche trade outlet, a well-regarded Reddit thread, or an industry directory nobody on a typical PR list would think to pitch, outworking them on conventional press coverage changes nothing.

"Earned media" gets used loosely, and the loose version overstates what it means. You'll see claims that 82 to 89% of AI citations come from "earned media." That's directionally real, Muck Rack's own December 2025 figure for non-paid citations overall is 94%, and 82% specifically falls into its earned-media bucket, but that bucket is deliberately broad: journalism, third-party blogs, government and nonprofit sites, social platforms, and reference sites like Wikipedia are all counted inside it. Journalism specifically, the thing most people picture when they hear "earned media," accounts for closer to a quarter of citations, 20 to 30% depending on the period. A PR Daily critique makes this point sharply: the single most-cited source on each major platform is often Wikipedia, PubMed, or Reddit, "none of which any PR team earns through media relations" in the traditional sense. If you've been told "earn media coverage" as the whole answer, it's incomplete. The real answer is closer to: find out which specific categories of independent source actually carry weight for your topic, which is frequently not journalism at all.

Niche, industry-specific outlets consistently outweigh generic authority. The report notes that AI models account for authority "in a more targeted way, like citing industry-specific sources for industry-specific queries," and that within most industries, the mix of outlets cited is distinct enough that a generic high-authority site doesn't substitute for the specific trade publication, community, or directory your category actually trusts. A moving company doesn't need a mention in a national business outlet nearly as much as it needs to show up somewhere people researching movers actually look, a local forum, a review aggregator specific to home services, a regional news story. Generic authority is a weaker lever here than it looks.

Freshness Is a Real, Measurable Tiebreaker

Here's a factor that's easy to miss because it has nothing to do with authority or relevance: how recently a source was published.

The same report found that half of all citations point to content published within the last 11 months, with the rest spread across a much longer tail. Recency is weighted even more heavily at the front end: roughly 4% of all citations in the study came from content published within the last seven days alone, rising to about 5% within two weeks and 8% within a month. Those aren't huge absolute shares, but they're a concentrated spike relative to how thin a single week normally is against an 11-month half-life, which is the report's own way of saying models reach for what's current when they can.

Practically, that means a page that's accurate but hasn't been touched in two years is at a real disadvantage next to a competitor's page updated last month, even if yours was more complete when it was written. This is a second explanation for "why them and not me" that has nothing to do with how good either business actually is.

Today's Answer Isn't a Permanent Verdict

Worth sitting with before anything else: the mix of sources models rely on keeps moving, sometimes within weeks. The same report tracked this directly across its two measurement periods and found real swings, not steady state. OpenAI measurably cut its reliance on Wikipedia between July and December 2025. Gemini briefly leaned heavily on YouTube for several weeks in November before reverting to its earlier pattern. Citations to Reuters climbed steadily across the same window. None of those shifts were announced, and none required anything to change about the businesses being described, only about which sources the models happened to be weighting more heavily that month.

That cuts both ways. A competitor's current edge isn't necessarily durable, the exact source backing them today may carry less weight in a few months regardless of anything either of you does. But it also means this isn't a one-time diagnosis you run once and file away. A gap you close this quarter can reopen later for reasons that have nothing to do with your own work slipping.

The Content Itself Still Has to Earn the Citation

Getting into the right sources and staying current both matter, but neither one substitutes for content a model actually needs to quote. Google's own guidance draws a useful distinction: generic content that dozens of sites already say the same way can be synthesized by a model without citing any particular source for it, while a specific, first-hand fact, result, or story has to be attributed to wherever it actually came from, because the model has no other way to produce it.

Apply that to the competitive question directly. If your "why choose us" page says roughly what your competitor's says, in roughly the same words a dozen other businesses in your category also use, a model doesn't need to cite either of you specifically to answer a general question about your category. It can just describe the category. The businesses that get named are usually the ones who said something a generic answer couldn't have included, a specific result, a specific method, a specific number, something that only exists because that business said it.

A Worked Example

Two independent moving companies operate in the same mid-sized city, both licensed, both with years in business, both with a normal spread of online reviews.

Company A's website is solid: clear pricing, clear service area, a few customer testimonials. Beyond its own site and a basic directory listing, there's nothing independent that mentions it anywhere. Ask an AI assistant who's reliable for a complicated move, and nothing in what the model can find distinguishes Company A from a dozen similar movers with similar sites.

Company B handled an unusual job eight months ago, moving a piano down four flights of a historic building with no elevator, and a local news outlet covered it as a human-interest story. That piece is specific, independently published, and impossible for a model to have synthesized from a generic "tips for moving" page. A thread on a local subreddit, started by an actual customer, mentions Company B by name when someone asks for a recommendation, with several replies confirming the experience. Neither of those things was something Company B's marketing team wrote. Both are exactly the kind of non-commodity, independently-sourced, reasonably fresh material the data above says actually gets pulled into an answer.

Neither company "beat" the other in some abstract sense. Company B simply has something specific, true, and findable backing it up, in a source Company A never thought to look for, let alone pursue.

What To Actually Do About It

Find out what's actually citing your competitor. Our brand visibility guide already covers the method, ask the same real customer questions, then look at what got cited alongside your competitor's name, not just whether you appeared. Run it with this post's finding in mind: you're hunting for a small, specific, probably unglamorous set of sources, not a generic sense that "they have more reviews."

Stop defaulting to traditional press pitching as the whole strategy. Given the two percent overlap finding above, a general PR push aimed at outlets nobody in your category is actually being cited from is effort spent on the wrong target. Go find out which kinds of sources are doing the work in your specific category first.

Create something a generic page can't be synthesized into replacing. A specific result, a specific story, a documented edge case, real first-hand detail. This is the one thing on this list that's entirely within your control and doesn't depend on any outside platform agreeing to mention you.

Keep your most important facts current, and let it show. A page that's visibly been maintained, updated pricing, a current year in a case study, a recent review quoted, competes better against the freshness weighting above than one that reads like it hasn't been touched since it was published.

Make sure the recognition layer underneath all of this is solid. None of the above does much good if the basic entity-clarity work isn't there yet, consistent naming, a clear description, real corroborating profiles. That's the foundation the guide above covers in full, and it's worth closing that gap before chasing anything on this list.

Common Mistakes

Assuming the competitor is simply bigger or better. Often they're not. They just happen to have one specific, fresh, independently-sourced thing backing them up that you don't.

Chasing generic press coverage as the fix. Given how rarely the outlets a PR effort targets match the outlets actually cited, this can be a lot of real effort aimed at the wrong list.

Treating "earned media" advice as if it only means journalism. It's a broader, looser bucket than that, Wikipedia, Reddit, directories, and nonprofit or government sites all count, and for many categories, those matter more than traditional press.

Publishing something once and leaving it. Given how heavily recency is weighted at the front end, a page that goes stale quietly loses ground to a fresher competitor, even one with a thinner history overall.

Skipping the recognition check. If a business genuinely isn't a clear, consistent entity yet, no amount of fresh, specific content fixes that first, more fundamental gap.

AI Schema Gen builds the entity-recognition foundation this all sits on top of, consistent identity, a connected profile, structured data that matches what's genuinely on your site. Run a free AI readiness check to see where that foundation stands before you start chasing anything else on this list.