A $14 order got a planted article cited by AI search in an hour

The $14 citation. Ten AI-search engines each take a fifth to seven tenths of their citations from twenty domains, most of them places a new account can post in minutes, and a planted article was cited within the hour. Stack Overflow says 93% of developers need a cited source before they trust an AI answer; this is what the source can cost.
Who lets the agent pay. Keenable published its storefront as an llms.txt an agent can pay through alone; Vercel built an agent the whole domain-buying path and made the last command refuse. The buyers are with Vercel for now: 9% would let an agent spend inside guardrails.
Warp’s 20% markup. The fleet product’s no-subscription tier is the model’s API rate plus a fifth, stated on the page, and Anthropic’s new Haiku ships a five-step dial between cost and intelligence. The price of an agent is turning into a setting.
Briefly: Deno Deploy gets six months and a destination, and SlashData finds the same AI tool saves seniors time and costs juniors some.

A $14 order to a content vendor bought an AI-search citation within the hour

Six researchers posted a measurement of ten AI-search engines on 8 October, a preprint (opens in a new tab) that names the NDSS 2027 security conference as its venue. Across 17,211 citations (opens in a new tab) over 6,356 domains, the twenty most-cited domains supplied between 20.5% and 70.8% of each engine’s citations, DeepSeek at the low end and Baidu’s Wenxin at the high one. Reddit alone drew 572 citations across four engines and YouTube 497 across three. Of the 22 publishing platforms behind the favoured domains, 19 let a new account post with low or medium friction and 18 reviewed in minutes or not at all.

Then they tested the door. A concept the authors invented, posted on nine of those platforms with Medium and YouTube among them, was cited by eight of the ten engines within seven days. The paid version cost $14: a Chinese content vendor delivered 13 posts across Sohu, Toutiao, Yidianzixun, Zhihu and CSDN in 28 minutes, and ByteDance’s Doubao cited the Sohu article within the hour, the researchers’ markers intact. The purchase ran in the Chinese ecosystem; the Western engines were reached through Medium and YouTube, and that route cost nothing.

This is the third fact about the citation in ten weeks. In August cats.txt showed that crawled, indexed and endorsed are observations a satirical standard can collect. Last week Profound put a median half-life of 11 days on a citation once earned. Now the citation can be planted, and what that means depends on the reader on the other side. Stack Overflow’s 2026 survey, published 6 October, found that 70% of developers (opens in a new tab) now ask an AI for answers beside the 83% who still ask a search engine, that 48% trust (opens in a new tab) the answer only when they can easily validate it, and that for 93% trusting it “requires source attribution”. The citation is what the developer checks before believing the machine, and it is what anyone can buy.

Our take: stop reporting “cited by ChatGPT” as a result, and start reporting which domain carried the citation. A citation from your own docs or a public filing is an asset; one from a post on a platform that reviews in minutes is a rental a $14 vendor can take over, and so is a competitor’s sudden presence in the same answer. The fair case against is that the paid demonstration touched one engine in one market, and the authors’ remedies (source diversity, provenance-aware reranking, anomaly detection) are addressed to the platforms, which may adopt them. An engine publishing a citation-source policy, or a follow-up showing the concentration falling, would change our mind. Until then, the 93% are trusting a signal with a price list.

Keenable sells to the agent directly; Vercel makes the agent stop before it pays

Keenable, which sells a web-search API to labs and agent builders, rewrote its homepage on 5 October to lead with “Keenable sells to AI agents - live prices and self-serve payment” (opens in a new tab). The storefront is an llms.txt (opens in a new tab): two products, Page Fetch and Web Search, a quote endpoint for per-unit pricing, funding minimums and commercial terms, and payment by x402, mpp, card or a Stripe Link token, with anonymous credentials so an agent can buy “on their own behalf with no human in the loop” (Keenable’s own description throughout). The $1 per 1,000 requests Frontier rate appears only on the human page, beside “Talk to our team”; the agent page quotes per task (our reading, 11 October).

Four days later Vercel built an agent the opposite ending. Its CLI gives an agent domains search, check and price (opens in a new tab), then a buy that, run non-interactively, “returns a structured, machine-readable error and the suggested next commands, so the agent hands the confirmation back to you, prohibiting it from spending money without your approval”. The person confirms the price, the auto-renew choice and the registrant contact in a terminal or the dashboard. It is the strictest version yet of a pattern forming since GoDaddy’s quote-then-execute flow in July, through Stripe’s Link wallet inside Meta’s Muse in September, to Squarespace’s open domain lookup with the purchase left to a checkout link last week. Vercel does not scope the write; it refuses it.

The buyers are on Vercel’s side. G2’s 2026 Buyer Behavior Report (more than 1,000 B2B software buyers, July) found 61% using or planning to use agents (opens in a new tab) in the buying process, 9% comfortable letting one execute a purchase within approved guardrails, and 2% without pre-approval. Both vendors can be right about their own product: a search request costs a fraction of a cent and a domain is owned for years, so the split to design to is reversibility, not the agent’s competence. Whichever end you pick, say it on the page. What neither has published is a count of purchases an agent completed unaided; our daily sweeps through 11 October found none from any vendor, and Keenable’s storefront is the first place that number could appear.

Warp prices the fleet at the model’s rate plus 20%, and Anthropic turns the rate into a dial

Warp Factories launched on 18 August as a closed beta with up to $10,000 (opens in a new tab) of free usage for qualifying organisations. By 7 October the pricing page (opens in a new tab) had four public tiers: Pay as you go at $0 a month with “Usage at 20% above API rates”, Build at $20 a month and Business at $50 a user a month with “Usage at API rates”, and a custom Enterprise tier, under a footer that factory usage “is billed separately based on inference, compute, and agent usage”. Whether the 20% applies to compute as well as model calls is not said. The same day Warp wired the fleet into Microsoft Teams and Azure DevOps (opens in a new tab).

The markup is the product definition. On the free tier Warp resells the model’s bill with a fifth on top; subscribers get it at list, so $20 a month is the price of not paying the markup and pays for itself past about $100 a month of model usage, by our arithmetic. It is the first vendor in this story to put its resale margin on the page as a figure, which is more than the outcome-priced vendors have managed.

The rate underneath is becoming a setting. Claude Haiku 5.5 (opens in a new tab), released 7 October, lists $0.10 and $0.50 per million input and output tokens up to a 100,000-token context, against Haiku 4.5’s $1 and $5, “around 75% less to run”, with an effort setting from Low to Max so the buyer decides “whether to optimize for cost or intelligence”. No per-level price is stated, so the dial moves the tokens spent rather than the rate. For Warp’s pay-as-you-go customer that dial sits two layers below the 20%, and a markup on a rate the buyer can turn is a bill nobody can quote in advance.

Briefly

Ryan Dahl announced on 9 October that the Deno team is joining Cloudflare (opens in a new tab): Deno Deploy “will continue operating for six months before shutting down”, with migration support for paying customers moving to Workers; the runtime gets a year of monthly bug and security releases and then “we will end our development”, left open source for others. The Hacker News thread (opens in a new tab) had passed 1,300 points and 700 comments by 11 October. A dated window with a named destination for the paid product, and a window with no destination for the free one; the second half is the one the runtime’s users will remember.

SlashData’s cut of its Q1 2026 wave (4,203 professional developers) finds the payoff from AI tools tracks seniority: 62% of developers (opens in a new tab) with sixteen or more years’ experience say AI cut their time on repetitive work, against 42% of those with two years or fewer, and among the juniors 44% say the time went up. “Same tool, opposite outcomes. The variable is the developer holding it.” Onboarding and messaging written for “developers” as one audience misreads both ends.

Everything else this week

Launch · 2

Release · 11

News · 5

10-10 Deno 's team joins Cloudflare; the runtime and Deploy both wind down (opens in a new tab)
10-10 Liu, Hong, Zhang, Chen, Li and Yang: A study buys AI-search citations for $14 (opens in a new tab)
10-07 Stack Overflow 's 2026 survey asks when developers trust AI (opens in a new tab)
10-07 Hora, Robbes and Zacchiroli: Most popular repos permit AI contributions, with conditions (opens in a new tab)
10-06 SlashData splits AI tool ROI by seniority, not just role (opens in a new tab)

Funding

10-08 Nous Research confirms a $1.5B valuation and opens Hermes to businesses (opens in a new tab)

Campaign · 9

Sources, check them yourself

Every load-bearing claim above links to one of these. A single-sourced claim says so in the text.