How ChatGPT Really Decides Which Brands to Recommend

Most of the advice floating around on how to get recommended by ChatGPT, Claude and Gemini focuses on technical tactics: how to structure your content, how to format your FAQs, how to signal freshness to crawlers. That’s not nothing. But it’s the wrong place to start, and it’s where most brands are wasting their energy.

A few years ago, I wrote a book called Trust Signals: Brand Building in a Post-Truth World, mapping out how careful buyers evaluate brands. They look for media coverage. They check review sites. They notice how a website presents itself. Each signal answers the same question: Can I trust this brand? I thought I was writing about human behavior. Turns out I was also describing what LLMs do.

When someone asks ChatGPT or Perplexity which company to hire, which software to buy or which agency to consider, those systems draw on a training corpus built from the web and, in a compressed and probabilistic way, run the same trust evaluation a diligent buyer runs manually. The brands that show up favorably are the ones with the strongest trust footprint across three distinct categories.

Website trust signals: Your digital first impression

The first category is what your own properties communicate. Website trust signals are everything visitors encounter when they land on your site — design quality, clarity of messaging, customer logos, testimonials, case studies, awards, certifications, team pages and the dozens of other elements that tell a visitor within seconds whether you’re the real thing.

LLMs are trained on web content, so the content and structure of your site contribute to what they learn about you. A thin, generic site gives them almost nothing. Your website should function less like a brochure and more like a trust portfolio — real client logos, specific outcomes in case studies, leadership pages that demonstrate genuine expertise.

Inbound trust signals: The breadcrumb trail

The second category — and the one that matters most for AI visibility — is what the rest of the internet says about you. Inbound trust signals are all the external sources that guide buyers toward you: media coverage, customer reviews, influencer mentions, guest bylines, awards, directory listings, Wikipedia entries, analyst citations, social proof of every kind.

This is where AI recommendation algorithms lean most heavily. LLM training datasets are curated to favor authoritative, frequently-cited sources, so high-authority content is more likely to surface in outputs. Media coverage, analyst citations, verified reviews on G2 or TrustRadius, original research picked up by trade outlets — all of it becomes part of your inbound trust record, the accumulated external evidence that your brand is real and worth recommending.

The brands that appear in AI-generated answers didn’t hack anything. They spent years building this kind of third-party validation — because what other voices say about you carries more weight, with both human buyers and the models trained on their behavior, than anything you say about yourself.

Think about the breadcrumb trail a careful buyer follows: the Wikipedia article, the Better Business Bureau profile, the analyst mention, the award that signals others have vetted you. That trail teaches an AI system which brands deserve to be in the answer.

This is also where most brands are weakest. Publishing more blog posts contributes almost nothing. What compounds is a feature in a trade publication with real domain authority, a survey your company commissioned that journalists covered, a byline that other sites cite.

These signals build human trust and, because authoritative coverage is more likely to be crawled and indexed, increase the chance your brand appears in the training data LLMs learn from. You can’t content-market your way to AI visibility. What works is the patient pursuit of external validation from sources that carry genuine authority.

SEO trust signals: What only Google can see

The third category belongs specifically to Google. SEO trust signals are signals that only Google can access — through Google Search Console, Google Analytics and its position at the center of web search. That gives Google data nobody else can see: how long visitors stay on your site, how often your brand is searched by name, how queries combine brand names with keywords, how long domains have been established, whether sites are mobile-friendly and technically sound.

The most telling is branded search volume — how often people search for your company by name. Google is widely understood to treat this as a proxy for genuine reputation. These signals matter for AI visibility indirectly: They influence where Google ranks your content, which shapes what gets crawled, indexed and ultimately included in the training data LLMs learn from.

What this means in practice

The path to AI visibility is a disciplined, long-term investment in all three categories. Build a website that functions as a trust portfolio. Invest in earned media — not volume, but authority. Pursue the reviews and citations that build your inbound record.

You can’t shortcut your way into an AI recommendation. The technical tricks have their place, but they only amplify a trust footprint that already exists. Build that first. The rest follows.

Most of the advice floating around on how to get recommended by ChatGPT, Claude and Gemini focuses on technical tactics: how to structure your content, how to format your FAQs, how to signal freshness to crawlers. That’s not nothing. But it’s the wrong place to start, and it’s where most brands are wasting their energy.

A few years ago, I wrote a book called Trust Signals: Brand Building in a Post-Truth World, mapping out how careful buyers evaluate brands. They look for media coverage. They check review sites. They notice how a website presents itself. Each signal answers the same question: Can I trust this brand? I thought I was writing about human behavior. Turns out I was also describing what LLMs do.

When someone asks ChatGPT or Perplexity which company to hire, which software to buy or which agency to consider, those systems draw on a training corpus built from the web and, in a compressed and probabilistic way, run the same trust evaluation a diligent buyer runs manually. The brands that show up favorably are the ones with the strongest trust footprint across three distinct categories.