The old bargain was simple enough to manage. A site let crawlers in, search engines indexed its pages, users clicked the blue links, and the publisher collected whatever the visit was worth, whether that was attention, advertising, leads, subscriptions, sales or reputation. It was never an equal deal, though operators knew its terms and could plan around them.

AI search changes what is being exchanged. A system can read a page, answer the user from it, cite it or skip it, blend it with several other sources, train a model on it or use it to trigger a transaction, and the reader may never form a direct relationship with the author. The visitors who do click through may be fewer and more intent on buying. Any business that publishes useful material is exposed to this. Consulting frameworks, software documentation, analyst research, law-firm primers and manufacturer manuals can all be consumed by AI systems without sending anyone back to the source.

Ranking therefore stops being the only practical question. For each right, an operator has to decide whether to grant it, price it or withhold it, and then measure how much direct relationship survives once an AI system answers on the operator's behalf.

Counting lost traffic misses what has changed: the rights in source material have come apart, and each one needs its own rule.

Evidence From The Market

Pew Research Center's 2025 browsing study found that Google users who saw an AI summary clicked traditional search results at a lower rate than users who did not, and almost never clicked the source links inside the summary. Google argues that AI features increase search usage and can send valuable traffic. Some publishers are weighing tougher controls, licensing demands, registration walls, bot filtering or litigation.

Cloudflare has turned the argument into mechanism. Its Pay Per Crawl experiment lets site owners set prices for AI crawler access, and its newer Pay Per Use experiments point toward compensation when content appears in AI search results. OpenAI now gives publishers and developers explicit crawler guidance for ChatGPT search. Google is exploring more granular controls for Search AI features. Perplexity has published updated guidance on robots.txt and says it has disabled an earlier web-address summary behavior that drew publisher objections.

No single answer has emerged. Controls, licensing deals, lawsuits, crawler policies, usage analytics and commercial experiments are all running at once, as they tend to in a market whose structure has not yet settled.

Six Rights That Can Be Granted Separately

Choosing between blocking AI and allowing it is too crude a decision. A company should separate at least six rights.

Right Question Failure mode Minimum control
Access Right Which crawlers, agents, and search systems may fetch which pages? The company treats every bot as either friend or thief, then loses useful distribution or gives away too much. A crawler policy by page class, bot class, freshness need, and server-cost burden.
Citation Right When should the company want to be named, linked, previewed, or excluded? The source appears in answers without enough context to earn trust, or disappears from answers where authority matters. A citation standard for flagship explainers, proprietary frameworks, documentation, and commodity pages.
Answer Right Which content can be summarized into an answer, and which content should require a visit? The company publishes its best decision logic in a form that can be consumed without relationship, consent, or conversion. A line between public doctrine, teaser evidence, operational templates, worked examples, and paid artifacts.
Training Right May the content improve a model, or only appear in search-like retrieval? The company grants training value by accident while believing it only granted discovery value. Separate crawl, index, training, and retrieval policies wherever tools permit the distinction.
Payment Right Which uses deserve free access, paid crawl, paid answer appearance, licensing, or refusal? The company negotiates as if all pages have the same value, then prices scarce source material like commodity inventory. A value tier for evergreen authority pages, timely analysis, proprietary datasets, and conversion assets.
Relationship Right What direct relationship remains after the AI system answers the reader? The company wins citations but loses email capture, subscriber habit, buyer identity, and feedback loops. A first-party path from public answer to owned channel, paid note, consultation, product trial, or review request.

Expert Businesses Face The Same Logic

Media companies feel the pain first because traffic sits on their profit-and-loss statement. Expert businesses meet the same problem more quietly. A consulting firm that sells no advertising still publishes to attract clients. Software companies that care little about pageviews rely on documentation, comparison pages, tutorials and case studies to carry buyer intent, and a founder with no need for mass traffic still needs enough direct trust to start a paid relationship.

What counts in an AI answer is how the source is used. Material used only as background supplies intelligence to someone else's product; a citation without a visit lends authority, and an anonymous visit gives discovery. Relationship value arrives only with a subscription, a purchase, a reply or a forward, so each of these outcomes needs its own rule.

Two Bad Defaults

Panic blocking is a poor default. Some pages should be blocked, others licensed, and others left open where reach and authority are the aim. A flagship page can stay visible while deliberately incomplete, with the full version behind a relationship. Another page can be rewritten so that an AI answer carries the thesis while the operating method stays behind a visit or a paid layer.

Passive optimism is the other. Google's claim that AI search raises query volume and sends higher-quality clicks may hold in some categories, but an average gives an operator nothing to act on. Several outcomes are live at the same time: fewer clicks that convert better, more mentions that convert worse, crawler load with no audience, and citations that buyers never remember. Industry averages cannot say what to do with a given page, so the rules have to be set page by page.

How This Publication Splits Its Own Material

This publication follows the same approach on a small scale. I want the public memos to travel, which means they should be discoverable, quotable and useful before anyone pays, because that is how trust forms. Reusable worksheets, scored examples, partner dossiers and operating templates belong to a different class: they are work product and should not be free inventory for every downstream use.

In practice the arguments are published openly, with a clear path to the owned relationship, and templates and detailed implementation material are reserved for the paid layer or direct review. The useful measurement is which public pages earn citations, replies, subscribers, partner feedback and inbound questions. When AI systems consume the public layer and produce no relationship or licensing value, the rules for those pages should change.

Start With Your Top Twenty Pages

Take the twenty public pages or assets that matter most to the business. Classify each as a reach asset, authority asset, conversion asset, protected work product or licensed source asset, and then set the access, citation, answer, training, payment and relationship rights for each class.

Perfect control is unavailable, since the tools are immature, the standards are contested and platform behavior will keep changing. What an operator can do is stop granting every right by default. A page can be public and still closed to training; a memo can be discoverable while the worksheet behind it is not. Being cited does not require giving up the customer relationship, and a crawler allowed on one class of page can be kept off the rest.

The AI-search fight will run through courtrooms and platform policy rooms, and operators cannot wait for it to finish. Treat source material as a bundle of rights, price or withhold the ones whose use destroys strategic value, and after each change check how many citations, replies and subscribers the affected pages still produce once an AI system has answered from them.

Source Notes