Mike Peralta, the chief revenue officer at publisher Future, can measure whether a client's brand appears in an AI answer. Connecting that appearance to a sale is harder. In Digiday's September 28 account of its publishing summit, Peralta acknowledged that the industry was still working out that connection. Publishers ought to keep his distinction in view when they are offered visibility as compensation for their work. A name in an answer may have value. The bills arrive in a less speculative form.
What a citation pays for
Consider what happens when an AI answer cites an independent product review. The reader can see where some of the information came from. There may be a link to inspect the testing, compare qualifications or find out why the reviewer reached a different conclusion from another source.
All of that is useful. Attribution helps a reader judge an answer, and it gives the person who did the work a chance to become known for it. Removing the credit would make the exchange worse.
But the reviewer may have paid for the product, spent days testing it and paid someone to edit the results. If the business earns money from advertising on its pages, the opportunity to collect that income generally comes with a visit. If it earns an affiliate commission, there is another action to complete. A subscription requires someone to decide the work is worth returning for.
A citation can lead to a visit, an affiliate sale or a subscription. The publisher still needs to earn enough from those actions to cover the cost of the work.
This sounds obvious until an industry starts treating the number of times a source appears in an answer as evidence that the source is being adequately rewarded. Visibility is easy to display. The less convenient question is whether enough people follow the path from recognition to support.
Pew Research Center's July 2025 analysis offers one reason to ask. Among the Google visits with an AI summary in its study, users clicked a link inside that summary just 1% of the time. The analysis drew on March 2025 browsing from 900 U.S. adults, with search results reconstructed in April. It was an observation of that period, not a forecast of every future interface or a measure of publisher revenue.
A high citation count could therefore come with few readers following the links.
The commercial meaning also depends on whose name appears. A furniture maker might benefit when an answer recommends its chair, even if the customer later buys from a local store. The review publisher has already supplied part of its own product when the answer repeats the testing conclusion. Both businesses may gain recognition, but they do not have the same opportunity to turn it into money.
Better visitors can still mean less income
Google's strongest response deserves more than a dismissive shrug. Liz Reid, who leads Google Search, argued in August 2025 that AI features were producing more useful visits. People could use a summary to get their bearings, then follow a link when they wanted to investigate or buy. Google described a quality click as one where the visitor did not quickly return to Search.
That is a reasonable signal of interest, though it does not establish a sale. A site full of visitors who immediately leave is hardly a thriving business. An answer that sends fewer people but helps the interested ones find exactly the right page could be a better source of customers.
For a publisher, however, the comparison has to reach the income statement. Suppose a site receives half as many visits. To preserve the same income from those visits, it needs to earn twice as much per visit, assuming everything else stays the same. That is hypothetical arithmetic, not a claim about what AI visitors spend. It shows what a promise of better traffic has to accomplish.
Readers might stay longer and see more advertising. They might return directly or subscribe. Some may arrive already persuaded that the publisher knows its subject. A business should take those possibilities seriously enough to measure them.
It should also be allowed to discover that they are insufficient.
A longer visit can still earn very little. A reader can trust a source without paying for it. Someone who sees a brand in an answer may buy months later, making the connection difficult to establish. A business needs to account for that doubt, whether the early signs look good or bad. An unmeasured benefit is possible, but its possibility does not settle a publisher's budget.
Google now provides separate AI Search visibility reports, so it would be wrong to say site owners can see nothing. Yet knowing a page appeared in an answer leaves the central business question open. A report of exposure and an account of earnings serve different purposes.
The same distinction should apply to claims of success from publishers themselves. A few valuable AI referrals are encouraging. They do not show that the new stream replaces what an individual site lost elsewhere. Comparing one expanding channel with its own small starting point can make growth look reassuring while the business as a whole earns less.
Payment is a separate bargain
There is another way to return value: pay for using the material.
Google has acknowledged that it is testing partnerships with websites whose content helps keep generated answers fresh and factual. Digiday reported on September 14 that publishers in an AI contribution pilot could see their monthly earnings. They had little detail on how the sum was worked out. Some participants wanted to learn by taking part. Others worried that accepting unclear terms would weaken their ability to seek better ones later.
It is early, and a pilot should have room to improve. Paying a source recognizes that it can contribute to an answer even when the reader never needs to visit it. Payment for that contribution is a transaction that can be considered separately from referral traffic.
The publisher then has something more definite to evaluate, provided the terms make sense:
- What use is being paid for?
- What affects the amount?
- Can the producer understand a change in payment well enough to plan the work that supplies the next answer?
Those are ordinary questions in a commercial relationship, even when the technology making use of the work is new.
Attribution should remain part of that relationship. A reader needs a route to the original, and a producer should receive credit. There is no reason to make publishers choose between recognition and compensation as though asking for both were an eccentric demand.
Peralta's admission leaves room for useful experiments. Some citations will bring business. Some partnerships will pay enough to support good work. Others will produce an impressive report and very little else. The producer needs to be able to tell which has happened before the next invoice comes due.
