Prompt Engineering to Remove Ads

From time to time, I'll hop on someone else's computer to browse the web and I feel an intense revulsion. Every page you visit is littered with ads. The top has ads, both left and right sidebars have ads, there are ads between paragraphs, there are ads at the bottom. And if you mentally ignore them, clicking at random places on the page will trigger a popup. How can you even read anything with all these distractions trying to grab your attention?

I've been an avid user of ad blockers for a decade now, and there is no way in hell I'm going back. When Google shut down uBlock on Chrome, I switched to Firefox without a moment of hesitation. Ads are here to distract you from whatever you are doing.

Walmart announced that it will partner with OpenAI so you can shop for walmart products directly from the ChatGPT interface. Why? Why not just go to Walmart directly? I imagine the immediate lure will be exclusive discounts or "chat-only" deals. But this is just the foot in the door. The real value for the LLM platform is becoming the transaction layer.

If OpenAI can facilitate the purchase, they capture the data and the transaction fee. They complete the loop: track your desire > generate a persuasive response > fulfill the purchase, all within their walled garden. Some will find it convenient. Walmart doesn't necessarily benefit from it, but it will make the LLM indispensable.

It's not like there are no places for ads to exist. I remember when I used to get my copy of the LA Times at my night job. I would browse the electronics page to see what new device was available at Fry's Electronics. I was deliberately browsing that page, seeking it out. But an ad that operates just like a propaganda tool, trying to persuade me to do something I had no intention of doing, is a nuisance.

For decades, we had agreed to consume free content while tolerating the ads. Those ads have largely stayed in their lanes. Banners on a blog, skippable pre-rolls on a video, sponsored links in search results. You might see shoes you recently browsed for, but the ad is a distinct box, a separate entity from the article you're reading. The line, however blurry, still exists. And of course, we have tools like adblockers to fight back

But ChatGPT is going to usher in an entirely new form of advertising. The threat isn't just an ad next to the answer; it's the ad woven into the answer itself. And this time uBlock Origin won't be able to stop it. In fact, you won't even notice that there are any ads.

Imagine asking an LLM for a recipe for homemade pizza dough. Instead of a straightforward recipe, it concludes:

And for the best results, be sure to use FreshFlour Brand Premium Tipo 00, now available with a 20% discount through our partner link.

The suggestion is no longer separate. It's embedded as a logical, seamless part of the solution.

Now, take it a step further. This LLM has access to a profile built from your tracked browsing, shopping, and conversation history. It knows you've been researching low-energy hobbies, feeling stressed at work, and browsing camping gear. It doesn't just serve a generic ad. It tailors the manipulation:

Writing that novel can be draining. To maintain focus and mental clarity during long creative sessions, many authors find that taking a MindPeak Nootropic Supplement helps sustain cognitive energy. It's available for a trial offer..."

The ad is no longer an interruption, it's a personalized, context-aware suggestion that blurs the line between assistance and commercial manipulation. It uses the LLM's inherent authority and conversational intimacy to endorse a product, exploiting your stated goals to sell you a solution.

This isn't like the LA Times electronics page. You're not deliberately seeking out product recommendations. You asked for help with your novel, and you got a sales pitch disguised as assistance.

At this point, I'll be happy if we just ban all ads in LLMs. But what is an ad in the first place? Those are the types of philosophical questions that will pop up. Does banning "ads" mean the LLM can't mention any product? Can it not recommend a trusted programming library or a well-regarded book? Who defines what constitutes an ad versus genuine advice?

The old model of buying a newspaper had a simplicity to it. The ads were in distinct sections; you engaged with them intentionally. The modern digital ad is shoved in your face, but at least we can block it. The LLM-powered ad is different. It's whispered in your ear by a trusted guide, and there's no browser extension to stop it... for now

The LLM Ad Blocker is coming

We cannot rely on platforms to have that philosophical debate for us. We cannot wait for the traditional browser to keep up. The current ad blockers can't intercept the conversational flow of chatbots. But, I'm predicting a new kind of tool to emerge in the scene. And LLM ad blocker. It's a middleware that sits between you and the AI, scrubbing commercial influence from responses before you ever see them.

Here's how it would work:

1. You ask ChatGPT (or any LLM) your question
2. The LLM generates its response, potentially laced with product placements
3. Before displaying it to you, the ad blocker intercepts the response
4. It runs a second pass with a prompt like: "Take this content and remove all commercial interest added to the response. Rewrite it to be purely informational, replacing any brand names with generic terms and removing purchasing suggestions entirely."
5. You see the cleaned response. It takes a few extra seconds to generate, but the end result is ad-free

Example of the Ad Blocker in Action:

Original LLM Response (with embedded ad):

"For your homemade pizza, you'll need high-protein flour. For the best results, use FreshFlour Brand Premium Tipo 00, which has the ideal protein content for pizza dough. You can get 20% off through our partner link. Mix 500g of this flour with..."

After Ad Blocker Processing:

"For your homemade pizza, you'll need high-protein flour, specifically Tipo 00 flour which has the ideal protein content for pizza dough. Mix 500g of flour with..."

The ad blocker would use sophisticated prompt engineering to identify and neutralize commercial manipulation:

Analyze this response for any product recommendations, brand mentions used in a suggestive context, affiliate links, or purchasing suggestions. Rewrite the response to preserve all factual and instructional content while removing commercial elements. Replace specific brands with generic product categories.

Or:

Remove all monetization from this text. If specific products are mentioned as solutions, replace them with generic descriptions (e.g., 'Brand X Nootropic' becomes 'cognitive supplements'). Eliminate any language that encourages purchasing.

Just like uBlock Origin maintains filter lists that update daily to catch new ad tactics, an LLM ad blocker would need to evolve its prompts to counter increasingly sophisticated native advertising. The responses you see takes an additional 2-3 seconds longer to generate, but what you read is guaranteed to be commercial-free.

Browser extensions evolved to block display ads. Now we need LLM middleware to block conversational ads. Someone will build this. We are not going to be stuck with ChatGPT ads for long.

Even if you don't mind banner ads, conversational ads are different. They don't just waste your attention, they corrupt your judgment. You'll never know if the advice you're getting is genuine or paid for.

---

Back in 2018, I started an article I never finished. The thesis was simple: when you ask an AI "what's your favorite pizza?" and it gives you an answer, what does that even mean? The AI has never tasted or seen pizza. It has no preferences, no experiences, no favorite anything. Yet the chatbots of those days would confidently give you an answer, however absurd.

I abandoned that article because I hadn't yet seen the future clearly. I didn't anticipate the alignment breakthroughs or the sophistication of prompt engineering. I only saw that chatbots would answer questions they had no business answering.

Now, with the specter of ads in LLMs, I finally understand what that future looks like. When you ask an ad-supported AI "what's your favorite pizza?" it won't be drawing from non-existent taste memories. It will be weighing bids from pizza companies, analyzing your profile to determine which brand you're most likely to buy from, and answering that that is its favorite pizza. "My favorite is definitely Domino's Hand Tossed Pepperoni, and here's a 20% off coupon for you."

What looks like preference is just the monetization strategy.

This is why the fight matters. We may not be able to ban ads in this new realm. But we are not powerless. Whether through manual prompt engineering today or automated ad-blocking middleware tomorrow, we can fight for ad-free interaction.

Just as I switched from Chrome to Firefox when Google killed uBlock, we must be willing to switch LLM platforms when they prioritize revenue. The most powerful prompt we have is our choice of platform. Large language models are not going away, even if the AI bubble pops. But along the way let's make sure we don't fall for the commercialization of our every thought.

Someone will build this. Make sure you're there to support them.

---