The visitors who used to come from search are dwindling — Ziff Davis said "Google traffic down 37%, AI citations at 10x clicks," People said "sessions fell 26% but revenue fell only 6%," and Amazon started listing outside stores in its own search for free

This week brought three sets of numbers that count "where the visitors come from." Ziff Davis (CNET, IGN, PCMag) said on its November 6 earnings call that organic Google traffic in Q3 fell about 37% from a year earlier, while its articles were cited about 2.7 billion times in Google AI Overviews and AI Mode over the same period, roughly 10 times its search clicks. People (PEOPLE, Allrecipes, Investopedia) wrote in its November 2 results that core sessions fell 26% from 2,170 million to 1,610 million, yet session-dependent revenue fell only 6%, and it named "AI-oriented search and content discovery experiences" as the cause. On November 4, Amazon opened a portal that lets stores that do not sell on Amazon appear in Amazon search and Alexa shopping for free just by connecting their catalog, and in the same week a survey that visited 200 UK and US stores with an automated browser found that one in seven blocked the way in, even though every one of those stores' robots.txt files allowed AI crawlers. We lay out how the three numbers are counted, and which columns a small team running apps and commerce should add to its own table.

This article was written with AI (Claude) and reviewed by an editor. Our company is a paying customer of both Anthropic, which makes Claude, and OpenAI, and we run apps, commerce, and AI channels ourselves. Ziff Davis, which appears in item 1 of today's article, is in copyright litigation against OpenAI, and People has content licensing agreements with OpenAI, Meta, and Microsoft. In the robots.txt section of the survey in item 3, Anthropic's crawler (Claude) appears as an "allowed crawler." Blend Commerce, which conducted that survey, is a Shopify-focused agency and disclosed its own interest. Figures announced by companies are marked "as announced by the company," and where something rests on a single source, that source is named alongside it. Quarterly results from three listed companies (Ziff Davis, People, Cloudflare) appear here, but this is not an investment opinion on any particular company and is not material for investment decisions.

In NO.38 we wrote that "there are now three front doors through which AI brings customers." The news was that ChatGPT takes search-style questions, that AI places orders in Shopify stores, and that Apple opened Siri to apps. This week's news is the other side of that. Two publishers put quarterly numbers on how many fewer visitors are coming through the original front door, search; Amazon began letting outside stores into the front door that is its own search; and on the store side, a survey found that the door through which AI shopping agents enter is closed not by policy but by default settings. We pick up where we read market share in NO.25, "what AI actually did" in NO.35, and a single line of robots.txt in NO.37. We separate facts, interpretation, and what we would do.

1. Two publishers' Q3 — traffic down, citations up (facts)

Ziff Davis released Q3 results on November 5 and held its earnings call on November 6. The figures CEO Vivek Shah gave on the call (as announced by the company) are as follows. Organic Google traffic to its sites fell about 37% year over year across CNET, IGN, PCMag and others, and since the decline in the same quarter a year earlier was about 23%, the two-year cumulative drop is more than half. In the same quarter its content was cited about 2.7 billion times in Google AI Overviews and AI Mode answers, which it said was about 10 times the clicks it received from Google Search. It said these figures came from Google Search Console and did not disclose the absolute number of clicks (PPC Land back-calculated roughly 270 million clicks from simple arithmetic, but we do not use that back-calculation). Revenue fell 5.6% to $289.3 million, and advertising and performance marketing revenue fell 7.7% to $186 million. Shah said "we are not planning for a search recovery," meaning the company operates on the assumption that traffic will keep declining; he said it is open to AI content licensing deals and that its lawsuit against OpenAI (filed April 2025) would help sort out this area. There were internal numbers as well. More than 90% of pull requests in the product and engineering teams are written by AI, and headcount is expected to fall 10% by year-end, though he added that "it is not all because of AI." We were unable to open the company's press release, filings, or call transcript; the figures above are only those on which three outlets that summarized the call (PPC Land, Stock Observer, Motley Fool) agree.

People (formerly IAC's Dotdash Meredith) released Q3 results on November 2 and held its call on November 3. These figures come from the company press release, which we opened directly. Core sessions 2,168 million → 1,613 million (−26%), total sessions 2,377 million → 1,794 million (−25%). The press release attributed the session decline "primarily to AI-oriented search and content discovery experiences," and said session-based revenue fell 6% "as that impact accelerated during the quarter." Page 5 of the investor presentation released the same day reads "core session decline accelerated to 26%, Google Search down 22%, AIO penetration 82%" (AIO is not spelled out, but reads as AI Overviews). On the call, CFO Tim Quinn said AI Overviews appear on more than 90% of the searches the company tracks, and CEO Neil Vogel said "we have lost close to 70% of our Google Search traffic" (PPC Land wrote "over the past few years," AdExchanger wrote "in two years," so the time frames differ). Revenue, however, moved the other way. Digital revenue rose 2% to $288.5 million, the twelfth consecutive quarter of growth; advertising fell to $161.9 million (−8%), but performance marketing rose to $73.9 million (+2%) and licensing and other rose to $52.7 million (+49%). The main drivers of licensing growth were syndication partners such as Apple News+ and the Meta content deal signed in Q4 2025, and the press release noted that this line includes "AI-related activity, including the use of large language models." Non-session revenue, which requires no sessions, reached 43% of digital revenue (39% a year earlier). One sentence from Vogel sums up the quarter: "Sessions were down 26%, but revenue was down only 6%." The company laid off 195 people (about 6%) on October 29, and Quinn said "we do not expect sessions to improve." The call also included Vogel saying the company had coined the phrase "Google Zero" internally (two outlets).

Other publishers' figures from the same week come from a single PPC Land roundup, so we do not use them in the body. We note only, as a single-source report, that the CEO of ad platform Teads said "AI summaries have cut publisher pageviews by 20–60%."

2. Amazon — lets outside stores in, keeps outside agents out (facts)

On November 4, Amazon opened the Shop Direct Merchant Workspace (Amazon News, company announcement). It is a free self-service portal that lets a store's products appear in Amazon search results and Alexa shopping once the store connects its catalog, even if it does not sell on Amazon. Registration takes three steps: create a seller account, verify your website, and share your catalog. The catalog can go through feed providers such as Feedonomics, Salsify, CEDCommerce, and VersaFeed, or directly via Amazon's Selling Partner API, and existing shopping feed formats are accepted as they are. Customers see a "Shop Direct" label on the product and either go to the store's site, or use Amazon's agentic purchasing feature Buy for Me to have Amazon's AI complete checkout on the store's site on their behalf (after the customer confirms the order details). Participation is free, the store chooses which products to include, it can leave at any time, and pricing, inventory, shipping, returns, and customer service stay with the store as before. The audience is all US customers (Amazon.com, the app, mobile web, and Alexa). The scale the company cited is "since launch last year, tens of millions of customers have discovered hundreds of millions of products from other stores" and "Shop Direct products appear in hundreds of millions of searches each month," but "tens of millions" and "hundreds of millions" are not defined, and no count of participating stores is given. Eligibility criteria, how Shop Direct products are mixed and ranked against paying sellers' products, and whether ads can be attached are absent from the announcement. Amanda Dorr, Amazon's vice president of core shopping, said it "makes it simple for stores to get their products discovered by Amazon customers."

There is a door on the other side too. In the same article, PPC Land noted that Amazon has restricted outside AI agents from entering its own store. It cited the lawsuit against Perplexity and the blocking of Meta's Muse agent in September as examples; we did not open the primary sources for either case this time (media summary). That Amazon said in March it would show "more than 100 million products from more than 400,000 stores" through Shop Direct is also recorded by PPC Land alone.

3. One store in seven blocks AI shopping agents "without knowing it" (facts)

We read the "AI Shopping Audit 2026" published on October 6 by Shopify conversion agency Blend Commerce (author: CEO Adam Pearce) directly. It visited 200 retailers, 100 in the UK and 100 in the US, 10 per industry across 20 industries (2 large, 2 mid-size, and 1 emerging per country), between September 8 and 15. The tools used were an unmodified Chromium browser and ordinary HTTP requests; it did not impersonate any particular AI company's agent, and requests were sent from residential connections in the store's own country. A store counted as a "refusal" only if it blocked twice (the second time on a fresh connection), and nothing was purchased (carts went only as far as the checkout screen). The audit has four layers, 17 items, and 100 points: Can it be reached (does the site load), can it be found (does robots.txt allow crawlers, is there a product sitemap), can it be understood (does the product schema include price, stock, GTIN, shipping, and returns), and can it transact (agent checkout endpoint, guest checkout, express checkout).

The results (as published by the survey's author). Average 65 points. By industry, baby products scored highest at 88 and fashion lowest at 49. 29 of 200 stores (about 1 in 7) refused the browser both times; those named are Home Depot, Best Buy, Macy's, Tesco, Argos, and Next. Refusal rates were 20% for large stores (16 of 80), 15% for mid-size (12 of 80), and 2.5% for emerging (1 of 40): the bigger the store, the more it blocked. Four stores, Saks Fifth Avenue, Bergdorf Goodman, Anthropologie, and JD Williams, refused everything including the home page and robots.txt requests and scored 0. The key sentence is this: of the 29 stores that refused, all 24 whose robots.txt could be read allowed the search crawlers of ChatGPT, Perplexity, and Claude. The report therefore attributed the refusals not to a policy of blocking AI but to bot protection software. In Pearce's words, "most of these stores have likely told the AI companies that their agents are welcome." Of the 171 stores whose product pages could be read, 82% had product schema, but only 22% had shipping information, 20% a returns policy, and 18% GTIN barcodes, and only 107 of 200 had a product sitemap. A live agent checkout endpoint existed at 38 of 200 stores, 37 of which were on Shopify. The 37 Shopify stores averaged 91 points against 59 for the rest, and the author noted that as a Shopify agency he has an interest in this result. He noted limitations too: 10 stores per industry is a small sample, the connection types differed between the UK and the US, and it is a snapshot of one week in September. We would add one more limitation. What this survey was blocked as was not an "AI agent" but an automated ordinary browser, so "blocked AI shopping agents" and "blocked automated browsers" cannot be distinguished in this survey.

In brief — Cloudflare: "Bots scan hundreds of sites and reward only one"

In its Q3 results on November 5 (8-K filing), Cloudflare reported revenue of $782.5 million (+39%), and its forward-looking statements mentioned an "agentic AI-first operating model" and workforce reductions. The filing says nothing about bots or crawlers. CEO Matthew Prince's remarks we saw in Techzine alone: bots scan hundreds of sites and reward only one of them with a mention in an AI answer or a product recommendation, so most sites get nothing from bot visits; since July 1, about 5% of the web has signed up for its content protection product (Meta is a customer); blocking AI bots is free, and Pay per Crawl, which charges crawlers for access, is in private beta. The article does not say where the remarks were made, so we leave them as a single outlet's report.

Reading the three numbers together (interpretation)

First, "exposure" and "visits" have become different numbers. Ziff Davis's 2.7 billion citations are 10 times its clicks. In NO.38 we counted three front doors through which AI brings customers; this quarter's numbers say the AI front door cites your articles but does not bring customers. What People showed is the next step. Sessions fell 26% but revenue fell only 6%, and licensing rose 49% so that non-session revenue reached 43%. It was a quarter carried not by the number of visits but by the value of a single visit and revenue that needs no visit (interpretation). "How many times our article was cited in AI answers" is no longer a number to boast about; it is a number for counting how many citations never turned into visits.

Second, Amazon has two doors. Into the front door that is its own search, it lets outside stores in for free; at the front door that is its own store, it keeps outside agents out. In NO.38 we wrote that "the platform holds the lock"; this time one more ring of keys is visible. What the store gets is exposure; what Amazon gets is the customer's search and purchase flow, and the checkout screen that Buy for Me completes. The price of "free" is not in the fee column but in the empty columns (ranking criteria, ad mixing, eligibility criteria) (interpretation). Those blanks are exactly the columns we should ask about whenever we hand our catalog to any platform.

Third, what blocks is not policy but defaults. In NO.37 we said that a line of robots.txt has to be read separately as "written" and "honored." This survey is the mirror image. The 24 stores whose robots.txt said AI crawlers were "allowed" were in fact blocking automated browsers. "Allowed" and "actually got in" also have to be read separately (interpretation). If Cloudflare's "5% of the web has signed up for protection" is accurate, the blocking default is spreading. The distribution, with large stores blocking 20% and emerging stores 2.5%, could mean either that smaller stores judged it better to leave the door open, or that they never bought the software to block with. Either way, for a team like ours the decision is not "block or not" but "whom do we let in, on what terms."

What we would do — three things this week for a team running apps, commerce, and AI channels (what we would do)

Split the traffic table into three rows: "search traffic / AI citation exposure / direct visits." Record quarterly Google click trends from Search Console for the homepage (apholdings.kr) and the commerce site, and check whether AI Overviews and AI Mode exposure show up separately, as they did for Ziff Davis (that company said it saw them in Search Console, but we have not yet checked how that breakdown appears in our own account). Then, following People's method, split our revenue into session-based (what requires a visit, like commerce revenue) and non-session-based (what comes without a visit, like app subscriptions, affiliates, and licensing) and keep one column for the ratio. This is for the same reason we proposed attaching the measurement method to market share figures in NO.25. This week we fill in only the last four quarters.
Check our own commerce site against Blend's four layers ourselves (it costs nothing). (1) Open our product pages with an automated ordinary Chromium browser and see whether bot protection blocks it. (2) Reopen robots.txt (together with the "three-bundle decision table" we deferred in NO.37). (3) Count whether the product schema has all five items: price, stock, GTIN, shipping, and returns (the three emptiest columns in the survey were shipping, returns, and GTIN). (4) See whether guest checkout and express checkout work from the product page. Record the result not as "open/blocked" but as "at which layer it is blocked." If it is blocked, the direction is not to turn off bot protection but to build an allow list. We are on the emerging-store side, so by the survey's distribution, leaving the door open is the right call.
Build a six-column table of "the terms on which a platform takes our catalog." Shop Direct is US-only and does not apply to us, but this is the table for when Shopify's "agent-ready by default" (NO.38) or a domestic platform asks for the same thing. The six columns are fee · can we leave · whether ranking criteria are disclosed · mixed with ads or not · whose customer data is it · on whose screen does checkout end. Put Amazon's announcement in this table and the first two columns fill in as "free · any time" while the last four stay empty; those blanks are the questions we would ask. Keeping two rows, "company announcement / what we verified," is the same as in NO.34.

What we did not pick this time

Next.ink reported that on the night of November 3–4, Google removed nine French AI-generated news sites from Discover by manual action ("large-scale spam and cloaking" email; the nine accounted for about 2% of French Discover articles). The only cross-source is an X post by monitoring firm 1492.vision and the back half of the article is paywalled, so we leave it as a future topic. Techzine's Kiro autonomous build, new controls for Google Cloud's security agent (Help Net Security, a follow-up candidate to NO.29 and 39), "99% of repository tokens never expire" (BlueFlag's own survey), and the problem of installing deprecated MCP servers (a follow-up candidate to NO.32) we did not open in full. The AI Weekly November 5 issue "AI conversations are evidence," the Bank of England's "AI is a fifth of UK growth," and Next.ink's "European law will ban chatbots from reproducing paywalled articles" and "NYT sues Brave" we saw only as headlines. OpenAI's procurement reporting (SiliconANGLE) and Palo Alto's acquisition of Clutch (Techzine) are investment-decision material and we do not use them. The quarterly figures for USA Today, News Corp, BuzzFeed, and Yelp come from a single PPC Land roundup and were left out. GPT-6 Cyber and the 948 Linux kernel reports (Techzine, Help Net Security) were covered in NO.41.

What we could not verify in this news

We were unable to open Ziff Davis's press release, filings, or call transcript. The 37%, 2.7 billion, "10 times clicks," 90% of PRs, and 10% headcount figures are those on which three outlets summarizing the call agree, and the company did not disclose absolute clicks. The 23% from a year earlier comes from two outlets. The Teads CEO's "20–60%" is from PPC Land alone.
We read People's press release and investor materials directly but not the call transcript. Quinn's "more than 90%" and Vogel's "close to 70%" were reported by two outlets, but the time frame for the 70% (several years / two years) differs. The expansion and measurement basis of "AIO 82%" are not in the materials, and the quarterly session decline rates (Q1 −17%, Q2 −22%) and the rise in revenue per session are PPC Land's compilation and estimates.
Amazon's announcement contains no eligibility criteria, ranking, or ad terms. The framing that "stores that do not sell on Amazon" are the target is PPC Land's summary; the official post directs existing sellers to a separate email. The Perplexity lawsuit, the Muse block, and the March figures (400,000 stores, 100 million products) are PPC Land's summary alone and we did not open the primary sources.
The Blend survey is a Shopify agency's own survey; robots.txt could not be read for 5 of the 29 refusing stores, so only 24 were verified; and the fact that what was blocked was an automated ordinary browser rather than an "AI agent" is a limitation of the survey design.
The venue of the Cloudflare CEO's remarks and the "5% of the web" figure come from Techzine alone. The filing does not contain them.

What we read alongside

We opened and checked sources of different kinds directly. We list them separately as company announcements and filings, survey reports, media coverage, and our own earlier articles.

Company announcements and filings:

Survey reports:

blendcommerce.comAI Shopping Audit 2026 (Blend Commerce, October 6)English, method, 17 items, results, conflict-of-interest disclosure, limitations

Media coverage:

Our earlier articles: