Open your analytics dashboard on a slow Tuesday and you will see it before you see anything else: impressions climbing, clicks flat, the two lines pulling apart like a couple that stopped talking months ago and only now filed the paperwork. We are a small, independent editorial shop. We do not have a data science team, a lobbyist, or a seat at any table where Google explains itself to anyone. What we do have is the same dashboard everyone else has, and lately it has been telling an ugly story.

The number everyone quotes

Start with the number everyone already knows: 58 percent. It comes from Ahrefs, which compared aggregated Search Console data from December 2023 against December 2025 across 300,000 keywords and found that pages in the number one spot lost 58 percent of their click-through rate once an AI Overview appeared above them. That figure keeps surfacing alongside Penske Media's antitrust complaint against Google, filed September 12, 2025 in the U.S. District Court for the District of Columbia, and it deserves precision. It is Ahrefs' math, not a Google admission, and it travels through a legal filing and the coverage around it, not a peer reviewed journal. Penske, which owns Rolling Stone, Variety, and Deadline, alleges AI Overviews now appear on roughly one in five Google searches that would otherwise send a reader to its sites, and that its affiliate revenue had fallen by more than a third from its peak by the end of 2024. The number is real, in that Ahrefs really calculated it. Whether it describes your site is a different question.

What Pew actually measured

Zoom out from the lawsuit and the picture gets calmer, not louder. Pew Research Center tracked the browsing activity of roughly 900 U.S. adults for the month of March 2025, across 68,879 unique Google searches, 12,593 of which returned an AI summary. Published that July, the finding was blunt: users clicked a traditional search result in 8 percent of visits when an AI summary appeared, and in 15 percent of visits when it did not. A mere 1 percent of visits produced a click on a link inside the summary itself. An AI summary roughly halves your odds of getting clicked, according to the one study in this fight built on observed behavior instead of an internal dashboard or a court exhibit.

Digital Content Next, the trade group for premium publishers, surveyed 19 member companies, a mix of news and non-news brands, over eight weeks in May and June 2025, and published its results that August. The median finding: Google referral traffic down 10 percent year over year, a 7 percent decline for news brands and 14 percent for non-news ones. No 58, no free fall, just a slow leak that adds up if nobody bails the boat. Google says the boat is fine. In an August 2025 post, Liz Reid, who runs Google Search, wrote that total organic click volume to websites has been "relatively stable year over year" and that the clicks Google sends are higher quality than a year earlier, and waved off contrary reports as flawed or outdated. Three sources, three numbers, one platform insisting nothing is wrong. Pick whichever you trust. Our own dashboard reads like the DCN number with a limp: not a cliff, a slope, and a steep one on exactly the pages an AI summary can swallow whole.

Why size makes it worse

Here is the part platforms do not dwell on: the damage is not evenly distributed. Axios reported in March 2026 on Chartbeat data from thousands of client sites, and the split by size is stark. Over the two years ending in December 2025, referral traffic from search fell 22 percent for large publishers, 47 percent for medium ones, and 60 percent for small publishers, which is the category we are in. A newsroom the size of a small city can absorb that with subscriptions, live events, or a licensing deal with an AI company that wants the association with real journalism. A handful of people cannot. We do not get a seat at Google's negotiating table, and we are not suing anybody. What we do get to decide is what we publish and how, and that turns out to matter more than it used to.

The playbook

So here is what we actually do, plus what we have watched work for outlets our size elsewhere, filed as instruction rather than hope.

  • Publish the primary material, not a summary of it. Original data, documents, on-the-record interviews, and firsthand reporting are the raw ingredient an AI system has to cite, because there is no substitute to pull from instead. We publish our methodology and the underlying figures on anything we call research, so there is something specific to cite.
  • Build things that will not compress. A ranked list survives being summarized in one paragraph. A calculator, a price tracker, or a searchable dataset does not, because its value sits in the using, not the reading. Our own network runs calculators and lookup tools for exactly this reason. A chatbot can describe what they do, but it cannot hand you your own result.
  • Own an audience Google cannot stand between you and. Email is the obvious channel. So is an app, a paid membership, or anything else a reader opted into once and does not have to search their way back to. We lean on direct readers and licensing partners rather than a newsletter, but the principle is the same: own the relationship, do not rent it.
  • Make yourself citable on purpose. Name your authors. Date your claims. Publish a plain methodology note on anything you call a study, even a small one. The systems doing the summarizing are pattern matching on trust signals same as anyone else, and an unsigned, undated page is not one.
  • Take licensing money when it is real, and read the terms first. Perplexity's Comet Plus subscription, launched in August 2025, shares 80 percent of its revenue with participating publishers, backed by a $42.5 million pool for early partners. That will not replace a search referral business, but it is not nothing. A revenue share on a product with ten users, on the other hand, is a rounding error with a press release attached.
  • Say the quiet part about what is gone. Commodity how-to queries, the ones with a single correct answer and no ambiguity, are not returning to your site. What time zone is Colorado in. How many tablespoons in a cup. An AI summary answers those better than most articles ever did, and pretending otherwise burns energy you need for the work only you can do. We do not write that kind of piece, not because we saw this coming, but because it was never good work.

None of this reverses the numbers above, the 8 percent, the 10 percent, the 58 percent depending on whose lawyer or dashboard you trust, and none of it arrives as a rescue plan handed down by a platform that changed the deal without asking first. What it is, is a bet: that a small publisher who reports something nobody else has, builds something a summary cannot fake, keeps its receipts in public, and gets read by people who chose to be there, can still make a living inside a search engine that increasingly prefers to answer for itself. We are making that bet with our own site. Ask us again next year whether it worked.

Analysis reflects Encore Editorial's own experience, our own traffic data, and the sources named above. Questions and corrections go through our contact page.