How Large Language Models Are Actually Trained

Language Learning Models (LLMs) like ChatGPT and Gemini are trained, almost entirely, on digital, publicly available content. Think: Wikipedia, digitized books and research papers, and billions of web pages.

Public content is rarely tested, controlled, or tied to validated outcomes. Your AI-generated marketing content reflects what’s visible, available, and free.

What LLMs Are Not Trained On

What these models are not preferentially trained on are A/B-tested fundraising copy, controlled campaigns, or performance-validated print and email communications.

LLMs do not automatically understand what constitutes high-converting copy, what suppresses response, or why certain messages outperform others.

Why AI Doesn’t Optimize for Results by Default

Unless prompted with performance data, AI won’t weigh the outcomes of using one piece of marketing copy over another.

Instead, LLMs identify patterns across billions of inputs — most of which are unverified, non-controlled, and disconnected from revenue performance.

The Predictable Result: Generic Copy at Scale

Over time, this dilutes the quality and performance of the marketing content it produces for you.

Your AI-generated copy will sound polished and read as technically competent… but it will have lost specificity, depth, and influence.

The result is predictable: AI generates generic copy because it is trained on generic copy. As AI-generated content increasingly floods the digital ecosystem, those same generic patterns are fed back into future training cycles.

Why This Is Especially Risky for Nonprofits

When nonprofits rely heavily on AI-generated copy, they place donor communications in the hands of a system trained on untested messaging, not on performance-validated appeals, segmented donor data, or long-term retention results.

The Practical Reality

Practically, this is truly no different than asking a stranger to write your donor communications based on what they’ve seen online or received in the mail.

Please don’t.

Bonus Insights: What Didn’t Make It Into the AI + Nonprofit Article (Video)

Work With Me at GrowBetter

At GrowBetter, I partner with nonprofits and NGOs of every size... from lean teams running a single annual campaign to established organizations who have scaled and are looking to optimize.

I help organizations like yours:

  • Fix underperforming appeals and email campaigns (you often have no idea how much y'ere underperforming until we dive in and begin getting you the lists you deserve...)
  • Increase response rates, average gift, and retention
  • Translate donor data into clear, persuasive messaging
  • Use AI... strategically

Whether you need:

  • deep dive into one critical campaign, or
  • longer-term strategy to grow revenue sustainably,

my work is always grounded in what actually lifts performance, not what merely sounds good.

If this article resonated, chances are something in your fundraising communications feels off.. flat response, declining engagement, or copy that sounds polished but isn’t converting.

Let’s look at it together. Book a free strategy call with me here!