Publishers spent $113m in one month buying back their search traffic

There is a number now for what the collapse of organic search referral traffic costs the news business, and it is not a projection. In July 2026 the hundred largest media publishers spent an estimated $113 million on paid search, a figure Adweek reported on September 9 using Similarweb tracking data. That is 41% more than the same month a year earlier and 274% more than three years ago.

The traffic those budgets bought came to 23.7 million visits, up 39% year over year and 148% over three years. Spending is therefore rising faster than the traffic it produces, which is the ordinary signature of an auction getting more crowded.

Concentration is the first thing the data shows. Forbes accounted for $72.2 million of the July total on its own, roughly two thirds of everything the top hundred spent, a 34% year-on-year increase and about eight times its outlay three years ago. The New York Times spent $11.3 million in the month, more than double its figure a year earlier. The remaining ninety-eight publishers split what is left.

The second thing the data shows is why. Forbes organic search traffic fell 26.7% year over year. CNN fell 28.9%. USA Today fell 24.1%. Publishers are not buying incremental reach on top of a healthy base; they are replacing a base that is disappearing, and paying an auction operated by the same company whose result page changes removed it.

David Carr of Similarweb dated the acceleration precisely: "We do see a surge in pay-per-click spending recently, basically since April, that has ramped up." Shiv Gupta of U of Digital put the circularity in blunter terms, describing publishers as desperate and noting that the advertising revenue flows straight back into the systems reducing their referrals. In his phrasing, they are feeding the thing that is killing them. Scott Messer, a media consultant, took the opposite reading, pointing out that publishers are targeting high-yield commerce keywords where the arbitrage genuinely works and that a tactic which pays for itself deserves the budget behind it.

Both readings can hold. What neither disputes is the direction of the underlying supply.

That supply question got a sharp measurement the same day. Search Engine Roundtable reported on September 9 that the People Also Ask box, one of the last large organic units on the results page still routing clicks to third-party sites, has been almost entirely absorbed by AI Overviews. Two independent datasets converge on it. Saeed Khosravi of Allintitle recorded 100% of People Also Ask answers as AI-generated as of August 2026. Mark Williams-Cook of AlsoAsked, working from roughly 19.2 million English-language queries logged during 2026, put the figure at 86% in August and 97% by early September. "It seems highly likely we are on track for 100% shortly," he wrote.

Fourteen months ago the share was about 12%.

The mechanics matter here more than the headline percentage. A People Also Ask entry that expands into an extracted snippet carries a link, a domain and a click path. The same entry rendered as an AI Overview answer carries a synthesised paragraph and a citation cluster, which is a different unit with different behaviour. PPC Land has tracked the click consequences repeatedly: randomised research found AI Overviews cut publisher clicks by 39.8%, Google has disputed a Pew study putting the reduction at roughly half, and the company has pointed to billions of weekly clicks from AI surfaces while third-party click-through data fell 58%. Smaller sites have fared worse than large ones, with one analysis finding small publishers lost 60% of their search traffic.

So the two September 9 stories describe one machine from opposite ends. On the SERP, an organic unit that used to distribute clicks now resolves questions in place. In the ledgers of the hundred largest publishers, $113 million a month goes back into the ad auction on the same page. The AI Overviews documentation describes the first as an improvement in answer quality. The Similarweb data describes the second as a cost line growing at 41% a year.

There is a structural asymmetry worth naming. Paid search inventory has no equivalent of the organic slot it replaced: it is priced by auction, so as more publishers arrive to buy back their lost visitors, the clearing price rises for all of them. The 274% three-year increase in spend against a 148% increase in visits is that dynamic written down.

Newsrooms are being rebuilt around what search stopped sending

The second half of the story is organisational, and Digiday's Sara Guaglione set it out in a Media Briefing published September 10. Publishers are not only changing where the traffic comes from; they are changing who inside the building is responsible for it.

USA Today Co. is assembling a new audience and digital production team of between 23 and 30 people and hiring an executive editor to run it. Monica Richardson, a senior vice president at the company, described the reasoning without hedging: "Growing audience by producing more content isn't as effective as it once was. Search traffic is under pressure, and platforms are increasingly keeping user experiences to themselves instead of sending them back to us."

The financial context behind that restructuring is unforgiving. USA Today Co. reported second-quarter total revenues down 8.3% year over year, with digital advertising revenue down 9.2%.

The New York Times has moved Christine Liang into a role overseeing AI and off-platform discovery. Her framing was that the way people search the internet has changed, and so has the way they read, with news increasingly reaching audiences through AI-mediated pathways. The Washington Post created its first head of SEO and AI discovery in October 2025, appointing Kyle Sutton, and has since named Jeff D'Onofrio permanent chief executive and publisher. The Post's programmatic advertising revenue rose 53% year over year and it sold 300,000 individual subscriptions.

That last pairing is the interesting one, because it points in two directions at once. Programmatic revenue growing at 53% implies volume the open market still values, while a subscription push implies a bet on the opposite: fewer, deeper relationships that no intermediary can intercept.

Danielle Coffey, president and chief executive of the News/Media Alliance, described the shift as one toward deliberate scarcity, saying that for the first time she is seeing many in the industry create the scarcity needed to increase value with the audiences they actually intend to reach. Barry Adams of Polemic Digital put the same point less diplomatically, arguing that the loss of Google traffic has finally forced publishers to take their own audience seriously. Ameet Shah of Prohaska Consulting described the internal work as incorporating AI to improve analytics, workflows, response times and audience understanding while running leaner teams.

Leaner teams and deliberate scarcity are not the same strategy, and the briefing does not pretend they are. One is a cost response; the other is a pricing response. Most of the organisations named are attempting both simultaneously.

Underneath sits unresolved litigation. The Seattle Times and Newsday have filed copyright infringement claims against OpenAI and Microsoft, joining a docket where OpenAI and Microsoft have already asked a judge to dispose of the consolidated 10.8 million-article case and where the Department of Justice has weighed in supporting a fair-use defence. The New York Times has sued three AI companies and put the annual cost of its journalism at roughly $2 billion. Separately, Condé Nast disclosed a breach affecting 32.8 million user records, a reminder that the first-party data on which the scarcity strategy depends carries its own exposure.

The broader traffic picture has been visible for a while. US news website traffic hit a twelve-month low earlier in the year, publishers have threatened to withhold inventory from Google as ad supply fell 40%, and global digital subscriptions have grown while fragmenting across more titles. What is new in the September 10 briefing is the org chart. Head of AI discovery is now a job title with a reporting line, a budget and a headcount attached to it.

Google builds the reporting layer for AI shopping, and rewires local at the same time

While the referral economics reorder themselves, Google has been shipping the instrumentation for the surfaces that replaced the old ones. Two Search Engine Roundtable items dated September 9 describe that work from the merchant side.

The larger of the two concerns Merchant Center. Google's AI Performance Insights report, launched in July 2026 and expanded to further regions in early September, has gained three new sections. AI search intent shows how a merchant's products match customer AI searches. AI search terms surfaces terms appearing frequently in the AI searches that returned a product, with recommendations to work those terms into product descriptions. AI attributes flags missing structured product attributes that limit visibility in AI search results. Brodie Clark identified the change and posted side-by-side screenshots of the expanded report against the previous version, and Google updated the corresponding help documentation.

The design of the report is worth reading closely, because it is diagnostic rather than transactional. It does not report spend, impressions or bids on AI surfaces. It reports the fit between a product feed and the queries an AI system is resolving, and then prescribes feed edits. That places the optimisation work in the product data rather than in the campaign, which is consistent with the direction of Google's retail tooling generally. PPC Land covered the Merchant Center report tracking brands in AI Mode, the eight feed attributes AI Mode leans on, and the open-source Merch Intel dashboard built for the same purpose.

A note on continuity: Merchant Center organic reporting has already been restated once this year, when organic traffic figures dropped following an August 24 reporting change. New sections added to a report whose baseline moved recently are not directly comparable to anything a merchant recorded before them.

The second item covers local. Ginny Marvin, Google Ads liaison, announced two features ahead of the holiday period. Local Customer Optimization is a campaign-level toggle inside Performance Max for store goals that prioritises budget toward consumers on Google Maps, Waze and local Search, reaching them, in the announcement's phrasing, right when they are actively in-market nearby. It is available now. Store Sales in Data Manager, arriving in the coming weeks, lets an advertiser link a CRM system or a Google Sheet directly in Google Data Manager to feed in-store sales data into an account for measurement and optimisation.

Taken together the two describe a loop: navigation signals at the front end, offline revenue measurement at the back end, with Performance Max sitting between them. Waze inventory was added to Performance Max with expanded channel reporting earlier, and Data Manager has been absorbing store sales and Analytics events through successive API versions, with version 1.7 pulling the Google Marketing Platform into the same signal loop. The consolidation has had a cost elsewhere: Google blocked new offline conversion imports through the Ads API from June 15, pushing that traffic into Data Manager.

A smaller September 9 item completes the picture. Google rewrote its location asset requirements documentation, adding policy sections on tobacco, recreational drugs and restricted targeting in personalised advertising, removing sections on misleading content, unavailable offers, unclear relevance and trademarks, and replacing the legacy word extensions with assets throughout. The page now carries explicit troubleshooting steps for editorial violations involving capitalisation, punctuation and symbols, including the instruction to edit a business name into standard title case and then unlink and relink the Google Business Profile so the change syncs. Google stated the update "aims to provide clearer information to advertisers and does not change enforcement of this policy."

Documentation rewrites of that kind rarely announce a behavioural change, but they do record which policy surfaces the company expects to be argued about.

The obvious contextual audience is often the wrong one, and the home screen goes programmatic

Two AdExchanger pieces dated September 9 approach connected television from different ends, and read better together than apart.

The first, written by Victoria McNally, makes an argument that runs against the instinct behind most contextual buying. Zack Rosenberg, chief executive and founder of Qortex, described the job as aligning with the audience consuming the content rather than with the content's subject matter, and offered a counter-intuitive example from his platform's data: Nike advertising performs about four times better against cooking content than against basketball content. Qortex, which launched as CatapultX in late 2020 before pivoting to contextual video intelligence, says it has accumulated more than five billion consumer interactions.

The explanation is not mysterious once stated. Basketball content is saturated with athletic footwear advertising, so the marginal impression competes against many similar ones and reaches a viewer already addressed. Cooking content is not, so the same impression arrives uncontested. The obvious adjacency is the crowded one.

Coegi, a media agency working in programmatic, search and social, reported the same effect in a business-to-business campaign. Hannah Schatz, senior director of programmatic operations, said video completion rate rose from 14% to 40% on a YouTube campaign after the contextual segment was changed away from the client's assumption. Chris Kotyck, executive vice president of digital operations, described the conversation with clients as telling them plainly that the segment they believe is their audience is not the one performing.

The systematic version of that finding has been building for a while. Gracenote has argued that missing programme-level data costs connected television billions in linear budgets and has published research on the contextual targeting gap in CTV; Ogury has pushed persona-based targeting explicitly as a challenge to genre-based buying. Genre is a convenient proxy for audience, and it is a lossy one.

The second piece moves from the ad break to the screen a viewer sees first. Telly, the television manufacturer whose sets carry a secondary narrower display below the main panel, is pitching programmatic home screen advertising. Mike Shehan, Telly's chief revenue officer and formerly chief executive of SpotX, put the rationale in supply terms: "For home screen ads to stick, they need scale, which is where programmatic enablement comes in."

The technical arrangement is specific. Home screen placements run up to five minutes in length and are delivered through Magnite's SpringServe ad server, with demand arriving from The Trade Desk and Teads among others. Telly cited a streaming service, unnamed, that recorded five times more app opens from a five-minute home screen unit than from a two-minute one. Mike Treon, head of connected TV and video strategy at the independent agency PMG, described the home screen as a persistent front door that gives brands a standing presence rather than an interruption.

Telly is not first here. Nexxen has built a home screen network across multiple manufacturers, adding TCL and TiVo and standardising one format across three OEMs; Samsung opened its smart TV home screens to programmatic buying through The Trade Desk and DV360; ShowHeroes brought Whale TV home screen inventory to programmatic across nine markets. Magnite's SpringServe has meanwhile been extending its mediation footprint with partners including Disney.

What the category still lacks is agreed signalling. The IAB Tech Lab finalised its connected TV ad format specification work in July 2026 after a public comment period that closed in June, confirming six format signals following an earlier draft released for comment. Telly says it is working with the Tech Lab on standardised home screen specifications for the bidstream, which is the step that would let a buyer distinguish a home screen impression from an in-stream one without reading the deal name.

There is a data dimension the article records plainly. Telly's signup questionnaire collects household income, household size, brand preferences, insurance providers, vehicle types and lease status, and interest-based behaviours. A manufacturer that gathers declared attributes at device registration is assembling a first-party graph on hardware it sold, which is a different starting position from an app publisher inferring the same fields.

Insurers pull back from GLP-1 coverage, and the marketing changes shape

A category that has spent three years marketing on efficacy is being pushed toward marketing on price and trust, and the cause is employer benefit design rather than anything clinical. Adweek covered the shift on September 9.

The coverage numbers moved fast. The share of employers covering GLP-1 medications for weight management fell from 72% in 2025 to 60% in 2026, and a further 10% of those still covering them are unlikely to continue. Sixty percent of employers cover the drugs for diabetes only. Thirty-six percent cover both diabetes and weight loss. Cigna eliminated GLP-1 weight management coverage for its own employees during the summer of 2026.

Where coverage retreats, employers are redirecting people rather than replacing the benefit: 27% now encourage access through direct-to-consumer platforms, and 21% point employees toward flexible spending, health savings or health reimbursement accounts. Both routes move the transaction outside the insurance relationship and into a retail one.

Sam Brough, head of brand at the global brand research firm Tracksuit, described the consequence for competitive dynamics: "As major insurers roll back coverage of expensive weight-loss medications, more of the cost shifts from insurers onto consumers, which will change the way weight-loss brands compete." His argument is that when a prescription becomes an out-of-pocket purchase, the decision criteria shift toward whether a person trusts a brand and whether it feels made for them, and that the companies treating the moment as brand building rather than another awareness push will gain ground in what he called the out-of-pocket era.

The demographic detail attached to that is the over-65 cohort, which represents 12% of the US weight loss category. A Medicare GLP-1 Bridge Program covering Foundayo, Wegovy and Zepbound has created a distinct addressable group, and Eli Lilly's "A Life Covered" campaign, produced by Wieden and Kennedy Portland, targets Medicare beneficiaries directly.

The commercial layer around the category has been rearranging in parallel. Amazon Pharmacy has been delivering GLP-1 drugs to Medicare patients at $50 a month and added the Zepbound KwikPen at $299, pricing that only becomes a marketing variable once the payer stops absorbing it. The unregulated fringe has grown alongside the legitimate market, with advertising for unapproved peptides up 208% against 276% marketplace sales growth, and litigation has followed, including a $5 million class action over patches marketed as GLP-1 products that contained no GLP-1.

A benefits decision taken in a human resources department has therefore reorganised media strategy for an entire pharmaceutical category, moved a share of demand onto retail pharmacy price comparison, and expanded the market for products that imitate the originals. None of that appears in a media plan under its own name.

Also noted

September 10: Instacart launched Cart Assistant, a white-label AI shopping assistant retailers can run on their own sites and apps, alongside Clementine on its own marketplace; Food Bazaar, Heritage Grocers Group and Woodman's are live, with Aldi US, Harmon's, The Save Mart Companies and Stew Leonard's to follow, and advertising general manager Ali Miller said the company wants to learn what resonates at scale before integrating ads. Digiday
September 10: Agency contracts are lagging the AI systems now doing the work, with disclosure of which tools were used, ownership of AI-generated output, indemnity and human-oversight clauses all unsettled; Keri Bruce, partner and head of the advertising group at Reed Smith, summarised the client position as "Ultimately, people want disclosure. They want to know what tools are being used." Digiday
September 9: August cable news ratings put Fox News at 2.127 million primetime viewers against MS NOW at 940,000 and CNN at 624,000, with CNN posting the only substantial year-on-year growth at 41% in total primetime viewers, measured on Nielsen's Big Data plus Panel basis. Adweek
September 9: Paris-based Actionable raised $10 million led by Hi inov and Axeleo Capital to apply causal AI to customer service data, following $2.2 million in pre-seed funding in 2024; co-chief executive Nicolas Rieul argued that large language models without semantic layers and industry context "just don't understand" raw data. AdExchanger
September 9: Google Search Console briefly reported indexed pages as "Crawled - currently not indexed" overnight on September 8, a reporting fault rather than an indexing one; John Mueller said on Bluesky that he "double-checked with the team, and it was a short blip that was resolved quickly in the night." Search Engine Roundtable