Imagine writing a detailed repair guide, then seeing its advice appear in a search answer. Your page is credited. The person asking the question gets what they need and leaves. Was the guide successfully discovered?
For the reader, perhaps. For the publisher, the answer depends on what discovery was supposed to achieve. A citation can acknowledge a source without bringing anyone onto its website. That gap is the useful starting point for understanding the video’s concern about the internet’s business model.
Follow the route from answer to reader
Consider a fictional site that explains how to maintain old bicycles. It publishes original measurements, photographs and instructions. A conventional search journey might bring a visitor to the guide, where they can inspect the photographs, explore another repair or decide to support the site.
A generated answer can shorten that journey. It might describe the adjustment in the search interface and offer supporting links. The user may be satisfied before opening the original guide.
This example is a mechanism, not a forecast of what happens to every publisher. It suggests three separate questions: Did the answer identify the source? Did anyone visit it? Did that visit help sustain the work? A report answering one question cannot stand in for all three.
What the click evidence actually measured
Pew Research Center’s July 2025 analysis used browsing records from 900 U.S. adults during March 2025. Researchers collected corresponding Google result pages in April. Traditional result links were clicked on 8% of visits with an AI summary, compared with 15% without one; links in summaries were clicked on 1% of visits.
Those are observed differences between search visits. They are not the percentage of revenue every publisher lost. The study did not randomly assign identical queries to an answer or no-answer interface. Query types differed, and reconstructed results could change over time. It covers a particular Google sample and period, rather than all search products.
The finding still matters: the presence of source links did not ensure people used them. It supports concern about the route back to a source, while leaving the size of any particular publisher’s loss unresolved.
How the platform describes the product
Google’s documentation says AI Overviews and AI Mode may run multiple related searches to assemble responses and supporting links. It also says ordinary search eligibility and SEO practices remain relevant; no special AI markup guarantees inclusion.
Google reports AI-feature traffic within Search Console’s overall Web search reporting. That makes an ordinary traffic chart insufficient to isolate the effect of summaries. Its description explains its own product, rather than independently proving a benefit for every publisher.
For our bicycle guide, the practical implication is to keep a record of the actual observation. “This page received fewer search visits” is evidence. “AI took all those visits” is an explanation that needs further support.
Make a small worksheet before changing the site
Choose a handful of pages that matter to the site’s purpose. For each, write down the question it answers, the useful action a visitor can take and the evidence available about discovery.
| Question | A useful record |
|---|---|
| Was the source visible? | A dated example of the result or citation |
| Did people arrive? | Visits or search clicks for the relevant page |
| Did visits help? | A relevant outcome, with its measurement limits |
| What else changed? | Page edits, availability and other plausible explanations |
Use comparable periods where possible. If the guide was rewritten, moved or unavailable during one period, note that before drawing conclusions. A single screenshot establishes what appeared once; it does not measure how often it appeared or what users did afterward.
The point is to make a claim small enough to check. You can investigate a particular page’s discovery problem much more usefully than trying to settle whether “the internet is dying.”
Fewer visits and better visits are separate possibilities
Consider two invented periods, used only to show the arithmetic:
| Period | Visits | Readers taking the valued action | Action rate |
|---|---|---|---|
| Earlier | 1,000 | 20 | 2% |
| Later | 600 | 18 | 3% |
The rate improved by one percentage point, while the number of actions fell by two. Calling the later visits “better” describes the rate; it does not recover the lost actions.
Neither number cancels the other. A publisher selling a service may care particularly about qualified enquiries. A public-interest guide may care about how many people can inspect the full instructions. Define that purpose before deciding which measure is reassuring.
Also distinguish helping a reader from funding the source. A concise answer may genuinely help someone who never visits. Whether that interaction pays for the next round of reporting, testing or maintenance is another question. The same event can produce convenience for one party and an unresolved funding problem for another.
Give the reader a reason to inspect the original
For the bicycle guide, a useful editorial response might be to make the original measurements easy to find, show which model was tested and explain where the procedure stops applying. These are choices about reader value, not promises of search ranking.
A reader checking an AI answer can use the same approach. Open the cited guide and locate the passage supporting the instruction. Check whether the answer kept its conditions: the relevant bicycle model, tool, date or exception. The RAG explanation follows how retrieved evidence reaches an answer; a citation still leaves the reader to check that it supports the particular instruction.
Avoid treating a more polished summary as stronger evidence. If an answer smooths away the difference between two incompatible measurements, readability has improved while usefulness has declined. The original page gives you a chance to recover that distinction.
The video’s broad concern becomes concrete here: an answer service depends on information someone had a reason to produce. Following the path from citation to visit to useful outcome shows where that relationship is working, and where it still needs a better arrangement.
