Why ChatGPT keeps citing your competitors and not you

The puzzle every audited SMB sees

A dental practice we walked through last month had everything search-marketers used to call "the right answers." A blog updated weekly. A services page with structured headers. FAQ blocks. Reasonable domain authority on Ahrefs. Pages indexed in Google, ranking on page one for a handful of local queries.

ChatGPT cited the practice zero times across a hundred customer-style queries we ran. Perplexity, twice. Claude, once. Gemini, never.

The competitor down the road, with thinner content and a slower-loading site, came up in eleven of the same hundred queries on ChatGPT alone.

The difference between the two sites had almost nothing to do with content quality. The competitor had been quoted in two trade-publication articles in the last eighteen months and featured in a regional business-news roundup. That was the gap.

What the data says about how AI assistants pick citations

Muck Rack analyzed 1.1 million AI citations across ChatGPT, Perplexity, Claude, and Gemini in early 2026. Their finding: 82 percent of citations come from earned media sources. News coverage, industry publications, syndicated articles, and third-party mentions account for the bulk of what AI systems point to when they answer a question. Ninety-four percent of citations come from non-paid sources of any kind.

A separate analysis from the AI Citation Economy report, published in April 2026, broke the citation pool down further. Brands themselves account for 52.5 percent of citations, but only when those brands have a public profile in news, reviews, or community discussions. News sites account for 20.3 percent of citations directly. Community forums add 5.9 percent. The remaining 21.3 percent is everything else combined.

For an SMB, the implication is direct. If your business is invisible in news, trade publications, and community forums, the LLM has no training signal connecting your brand to anything it can cite. The blog post on your own domain is, on its own, a thin signal.

Why this is structural, not editorial

LLMs are not making aesthetic judgments about content quality. They are pattern-matching against the corpora they were trained on. Those corpora are dominated by news sites, publishing platforms, and high-volume forums for one mechanical reason: those sources produce a lot of text per topic, and that text gets re-aggregated, syndicated, and quoted across the open web.

A blog post on a single SMB domain appears in the training data once. A trade-publication article that quotes the same SMB appears in the publication's archive, in dozens of news aggregators, in linked forum discussions, and sometimes in Wikipedia citations. By the time a model is trained on the open web, that single quote has been seen many times more than the blog post.

The Otterly AI research from early 2026 found that high-traffic domains earn 3x more AI citations than comparable low-traffic domains, after controlling for content relevance. Most of that 3x advantage tracks back to earned media history, not to anything you can replicate by writing more.

The 239 percent lift from distribution

The most actionable finding in the recent research comes from Authority Tech, who tracked content placed in third-party news outlets and trade publications versus the same content sitting on a brand domain. The median lift in AI search visibility was 239 percent. In some verticals it reached 325 percent.

Same content. Different distribution. The signal AI systems weight is the appearance of the brand name in places they were trained to trust.

For an SMB, this means the sequence operators have been running for a decade is upside down. Publishing more on your own domain is the slowest path to AI visibility. Getting one or two earned mentions in trade publications, a quote in a regional business article, or a feature on a vertical podcast typically lifts citation rates inside four to eight weeks. Quarterly publishing on the same brand domain produces a far slower curve.

What this looks like for the operator

The work involves two streams running in parallel:

1. Earned media outreach. Identify three to five trade journalists or podcasters who cover your vertical. Pitch a specific operator-side angle they would actually run. Land two mentions in the next ninety days. Most SMBs have not done this work and the bar is lower than it sounds. The format that lands best is a specific point of view backed by a small piece of original data from your operation. A dental practice can share what they observed about appointment-no-show patterns after a rate change. An HVAC operator can share what their service-call volume looked like during the last cold snap. Trade journalists need this kind of operator-side detail and most do not have a steady supply.

2. Forum and community presence. Active industry forums, regional business associations, and vertical Discord or Slack communities feed AI training data heavily. A few well-written contributions per quarter generate the kind of distributed mention pattern that LLMs weight. The contributions do not need to be self-promotional. Answering a question another operator asked, sharing a vendor recommendation with reasoning, or contributing to a discussion about regulatory changes all leave the kind of footprint that compounds over twelve months.

A third channel sometimes worth running in parallel: vertical podcast appearances. A single thirty-minute conversation on a podcast that covers your industry produces a transcript, show notes, episode pages, and clipped social posts. Each piece of that footprint is a separate mention in the AI training corpus. One podcast episode often outperforms six months of original blog posts on citation lift, by the same study that produced the 239 percent figure above.

The Singapore AI Search Visibility Index reported that brand visibility among tracked SMBs declined 35.9 percent in early 2026, with citation rate dropping 34.4 percent over the same period. The operators who reverse the trend are the ones who treat distribution as a primary channel, the way they treated SEO ten years ago. The shape of the work is similar. The difference is that the audience is now a model reading the open web, not a search algorithm reading a backlink graph.

The audit step

Before any of this, the diagnostic step matters. Knowing which AI assistants cite your business today, where those citations trace back, and which competitors are absorbing the queries you should be winning is the starting point for any earned-media plan.

A measurement layer that sits across ChatGPT, Perplexity, Claude, and Google AI Overviews tells you exactly where the gap is. Without it, every distribution effort is a guess.

Frequently asked

What counts as earned media for AI citations?

Mentions of your business outside your own website. Press coverage, trade publication features, podcast appearances, journalist quotes, and mentions in industry forums or community discussions all qualify. The common thread is that someone other than you decided to write about you.

Can a small SMB realistically get earned media coverage?

Yes, though the channels look different from enterprise press work. Local trade journalists, vertical podcasts, regional business associations, and active industry forums are all reachable for a single operator with a few hours a week. The bar is one or two mentions, not a national press hit.

How long until earned media starts moving AI citations?

Faster than backlinks moved Google. Most operators see citation lift in four to eight weeks once a few quality mentions land. The largest published study tracked a 239 percent median visibility lift across verticals after the distribution work landed. Background reading on the broader search-marketing playbook explains why the underlying mechanics differ from classic SEO link-building.