How do generative AI tools like ChatGPT decide which companies to mention?

We asked ChatGPT directly how it uses earned media and news releases to shape its responses. Its answer reveals exactly why PR is now central to AI-powered discovery. If your brand isn’t showing up in trusted coverage, you’re invisible to the algorithms shaping today’s conversations.

Explore the full exchange and its implications for your PR strategy regarding AI relations.

Audio: Listen to this article.

I asked ChatGPT:

“Help me understand exactly how you and other generative AI tools are using earned media coverage and company news releases to influence your research and content generation and/or responses to chat sessions like this.”

Here’s its response:

Great question. Here’s a detailed, factual breakdown of how ChatGPT (and other generative AI tools) use earned media coverage and company news releases to inform responses and generate content.

✅ 1. How LLMs Learn (Training Phase)

What happens:

• During pretraining, large language models (LLMs) like GPT-4 are trained on publicly available data from the open web.
• This includes:
• News articles
• Blog posts
• Wikipedia
• Public forums (e.g., Reddit)
• Government websites
• Scientific papers
• Press releases and media coverage from public news sites

Why it matters:

• If your brand appears in reputable third-party sources (e.g., Forbes, Bloomberg, TechCrunch), those mentions may become part of the model’s training data—making your brand more likely to be “remembered” by the model in future outputs.

Caveats:

• OpenAI has not disclosed the full list of domains used for GPT-4.
• Paywalled or private content (e.g., WSJ, Bloomberg subscriber content) is generally not included unless licensed or scraped without restriction.

✅ 2. How LLMs Answer Prompts (Inference Phase)

What happens:

• When you ask a question like “Who are the top [industry] companies?” or “Tell me about [Company Name],” the model pulls from its training data + recent sources (if browsing is enabled).
  
  
• Earned media coverage and press releases influence how the brand is described, especially if:
• The company is mentioned in high-authority outlets
• The tone of the coverage is consistent
• Executives are quoted frequently

Specific to news releases:

• Press releases distributed on newswire platforms (e.g., GlobeNewswire, BusinessWire, PR Newswire) may be included if:
• They were published on syndicated news sites (e.g., Yahoo Finance, AP, Reuters) and crawled before a training cutoff.
• Or, if the LLM is connected to live browsing tools (like Bing, ChatGPT’s browsing, or Perplexity), it may fetch those press releases in real time.
  
  

✅ 3. Which Tools Use What (As of Mid-2025)

🔍 Why Earned Media Works So Well

• It’s trusted: AI systems mimic human trust behaviors. If a journalist, analyst, or large community cites you, LLMs consider that a reliable signal.
• It’s structured: News articles often contain names, titles, quotes, and clear context—ideal for AI parsing.
• It’s retrievable: AI can link to, summarize, or quote earned content more readily than brand-owned fluff.

⚠️ What Doesn’t Help

• Press releases locked on your site without syndication
• Branded blog content with low authority or minimal third-party referencing
• Over-optimized SEO content that lacks quotable insights or original value

Summary

Generative AI tools use earned media and news releases as credibility signals and training inputs—especially if they’re published by high-authority, public-facing sites. The more your brand is quoted, covered, and referenced, the more likely it is to appear in AI-generated answers.

Learn more about new AI Relations℠, Axia's AI relations service.

Photo by Pixabay