When Chat Replaces Search: The Economics of AI-Mediated Information

Back in 2015, I was a software engineer building the invisible infrastructure that turned attention into money. Not the polished front end users saw, but the backend systems that bought, sold, and measured ads in milliseconds.

Here’s how that marketplace worked:

• Supply-Side Platforms (SSPs): Let publishers auction off their ad inventory programmatically, impression by impression.
• Demand-Side Platforms (DSPs): Allowed advertisers and agencies to decide in real time which impressions to buy, at what price, and with what targeting.
• Ad Exchanges: The clearinghouses where SSPs and DSPs met, running Vickrey (“second-price”) auctions in the blink of an eye.
• Data Management Platforms (DMPs): Packaged behavioral and demographic data into segments so buyers could bid smarter.
• Attribution & Measurement Providers: Tracked whether impressions converted into clicks, installs, or purchases.
• Verification & Safety Vendors: Flagged fraud, bots, and unsafe placements.
• Agency Trading Desks: Sat on top of DSPs, pooling buying power for major brands.
• Ad Networks: Legacy bundles of inventory with niche audiences or vertical targeting.
• Tag Managers & Analytics: The plumbing for pixels, tracking, and multi-touch attribution.

Every time you loaded a page, this system whirred into action. An impression was auctioned, data was applied, bids were placed, and the winning ad rendered, all before the page finished loading.

It was complex, often messy, but it monetized the open web. That system kept most of the internet free for decades. But LLMs are now reshaping how information is consumed and monetized.

The Great Unbundling

AI labs are now cutting deals directly with publishers and platforms: Axel Springer, Reddit, Stack Overflow. When people spend more time asking ChatGPT or Claude for answers than clicking through Google results, the traffic model publishers depend on begins to erode, fewer impressions and clicks despite consumers getting arguably more value than ever.

Google was about sifting through citations. AI chat products like ChatGPT, Perplexity, and Claude are about generating an abstract for just about any question you can think of, with the citations sometimes mentioned and other times embedded into model weights and unattributable.

Pressures on the Open Web

Several pressures are converging:

• Volunteer-driven communities may thin out. Why contribute if your expertise is scraped and reused elsewhere? Many are retreating into smaller, private spaces (e.g., Discord, Slack).
• High-quality content is moving behind paywalls. Outlets won’t let AI tools freely summarize their work forever. Lots of incentive to track robots.txt adherence and a litigation bonanza (e.g., Cloudflare’s proposed pay-per-crawl model).
• The remaining open web is noisier. Generative tools are accelerating the production of low-value content (aka AI slop).

The result is a more uneven landscape: premium content behind walls, mid-tier content absorbed into AI summaries, and a flood of synthetic filler competing for attention.

LLMs as the New PageRank

Google’s breakthrough was PageRank, links as votes that determined visibility.

Large language models are becoming the new PageRank. But they don’t just index; they synthesize and present information directly across sources, making it challenging to extract sources. The competition is no longer for search rankings, but for how models surface and represent content in conversation.

Publishers once optimized for Google’s crawlers. Now they’ll need to understand how they appear in Claude’s or ChatGPT’s responses if at all.

From SEO to AEO

The next frontier is Answer Engine Optimization (AEO). Just as SEO shaped a generation of publishing, AEO will shape the next. Brands are already experimenting: generating permutations of queries like “best corporate card” to test how tools surface their products. Some are licensing chat data from vertical AI applications to better understand what customers actually ask when making decisions.

Search intent was powerful. Chat intent is more precise. If search revealed what you wanted, chat reveals why you want it.

In-chat monetization will follow. Free AI tiers supported by ads. Sponsored responses. Branded suggestions woven into answers. It seems inevitable that free tiers will appear once advertising or data revenue can offset token costs.

Beyond Licensing: Microtransactions and Data Sharing

The blunt instrument today is licensing: a publisher gets a flat fee for its data, regardless of how often it’s used. That structure often benefits the buyer, who has better information about usage patterns, than the seller.

Microtransactions could change that balance. With MCP servers acting as intermediaries, vertical AI companies or general-purpose models like ChatGPT could pay per reference, per snippet, or per source. Imagine a company like Open Evidence supporting itself not just through broad licenses, but through arrangements where a journal like JAMA charges a few cents each time one of its articles is cited or summarized in an answer.

The tools already exist to price content this way. It would move the market from flat, uncertain licensing fees to usage-based pricing that better reflects actual value. And it might give brands and publishers stronger incentives to share their data rather than walling it off. I don’t know about you, but I would be willing to pay $1 to read some FT articles, but the current subscription pricing goes beyond my willingness to pay.

Options for Adapting

This isn’t about who “survives” but about what paths are open:

• Strike data partnerships: Early licensing deals with AI labs or vertical tools.
• Experiment with microtransactions: Price content per use rather than only through flat fees.
• Explore AEO strategies: Understand how products and brands are represented in model responses.
• Invest in direct relationships: Premium subscriptions, communities, and other ways of reaching audiences without depending on referral traffic.

No single option will replace the old adtech system. But together, these approaches point toward a more diversified set of models — some familiar, some new.

LLMs Redefining Discovery

The last era of the internet was built on PageRank, which decided what surfaced in search. The next will be built on LLMs and how they choose to represent and deliver information in response to a query.

The money will adapt, as it always does. The harder question is how publishers, communities, and creators adapt with licensing, microtransactions, new distribution models, or something else entirely.

The open web won’t disappear overnight, but its business model is being rewritten in real time, one prompt at a time.