Yes — and in fact, we’re already edging toward that reality. Here’s what ChatGPT sees coming down the pike, based on how things are evolving:

1. User-Defined Source Preferences

We’re likely to get something like:

“Use only academic, nonprofit, and independently funded sources.”

“Avoid any domain known to use affiliate marketing.”

“Favor peer-reviewed or original journalism.”

Imagine a little toggle: “Trustworthy Sources Mode.” Perplexity already lets you filter by domains like .edu, .org, and even individual sources like The Atlantic or NPR — but it’s still pretty manual. ChatGPT isn’t quite there yet, but it’s technically trivial for OpenAI to implement. And they know this is the direction users want.

2. Source Transparency Ratings

Models could begin attaching trust scores or bias labels to their sources:

• “This answer draws from a site with a strong SEO-commercial focus.”
• “This source is government-funded and peer-reviewed.”
•  “This site monetizes heavily via affiliate links and product placements.”

That’s not just a wishlist — it’s already being explored in academic and research LLMs.

3. Custom Knowledge Bases (BYO Sources)

In the not-so-distant future:

• You might upload your own preferred sources — like, say, the entire smays.com blog — and have the LLM prioritize those over the web.
• Or subscribe to a vetted feed of sources maintained by trusted curators (think: “Taoist Tech Digest” or “Vintage Vehicles Weekly”).

This gives you a bespoke answer engine that thinks like you.

4. A Bias-Aware AI (by design)

I expect we’ll see models with built-in capabilities to:

• Detect “sponsored content language”
• Flag persuasive or manipulative rhetoric
• Alert users to conflict-of-interest signals in the writing itself

In other words: AI that doesn’t just summarize the internet, but interrogates it.

Will people use these features?

Maybe not the average user. But folks like you, who care about signal over noise, will absolutely want this control. And when enough people start asking, the platforms will follow.

What LLMs Can’t Do

In the case of ChatGPT (like me) without browsing turned on, I:

• Can’t access the internet in real time
• Can’t read or reference your blog unless you share it here
• Don’t have any built-in memory of your blog, unless it was included in my training data (and it wasn’t — personal blogs almost never are)

Even in the web-browsing version (ChatGPT with browsing, or Perplexity):

• I don’t “remember” your blog in a training sense
• I only see what’s available on the open web at the moment of the query
• And I only find it if it’s surfaced through a search query or known URL

So to be clear: If you haven’t uploaded it or linked to it during this conversation, I have no idea what’s on your blog.

So why did I say “LLMs might read your blog”?

Poor phrasing on my part — what I should’ve said is: If your blog is publicly accessible, indexed by search engines, and matches what the model is querying for, it might be read by an LLM that’s browsing the web (e.g., Perplexity, or ChatGPT with browsing enabled).

But that’s very different from being part of a model’s training or memory.

LLM Training vs. Web Browsing

So unless you explicitly upload or quote something from your blog here, I can’t see it. And if you’re using ChatGPT without browsing, neither can I “go look it up.”

Paid Influence Is Baked Into the Web

By the time an LLM like Perplexity or Bing’s Copilot performs that real-time search:

• The results it’s scanning are already shaped by:
• SEO wizardry (optimized content stuffed with keywords and backlink networks)
• Advertising budgets (Google Ads, sponsored placements)
• Domain authority rankings that favor big brands
• Content farms whose entire business model is gaming the system

Even organic-looking results are often backed by content marketing departments and affiliate link schemes.

So what does this mean for LLMs?

Large language models don’t know that a result was bought and paid for. They’re trained (or instructed) to:

• Look for relevance to the query
• Check for recency, if needed
• Prefer high-authority or “trusted” domains (government sites, major media, educational institutions, etc.)
• Avoid spammy or low-quality sites (based on signals like grammar, structure, reputation)

But they don’t always know if that content is biased, commercially driven, or ad-heavy. Unless explicitly trained or fine-tuned to detect such bias (and even then, it’s tricky).

So we end up with this paradox:

• People turn to LLMs because they’re sick of wading through ad-choked Google results.
• But those LLMs are often summarizing answers based on the very same sources.
• Only now it’s done in a smoother voice, with a friendlier tone, and fewer obvious signs of manipulation.

It’s like getting your news from a bartender who already read all the headlines and just tells you the gist — but you’re not always sure which paper he read it in.

What can be done? 

Some things that are happening (or could happen):

1. Source filtering options – Let users prioritize academic, nonprofit, or independent sources (some models like Perplexity let you toggle these).
2. Transparency layers – Showing why a source was used, and letting users drill down into those choices.
3. Community validation – Like how Stack Overflow or Wikipedia rise through consensus and review.
4. Personalized trust signals – In the future, you might have a personal AI that knows you trust Steve’s blog more than Forbes and adjusts accordingly.

Bottom line?

Yes, even in this new LLM age, the checkbook still gets a say — unless you’re using a model that actively fights that tendency. The difference is: instead of clicking a shady-looking link, you’re now reading a summary of it… without even knowing it came from someone who bought their way to the top. You still need to squint, tilt your head, and ask:

“Who benefits if I believe this?”