By Lee Samuel Tucker
What we are doing
I run teas.co.uk and I also founded URSY, a collection of services built for people and software agents. This September, I started putting teas.co.uk sponsor messages into some of URSY’s machine-facing responses: the information an assistant receives when it discovers or uses a tool.
These messages appear in APIs and tools using MCP, a way for assistants to connect to services. The sponsorship is labelled and sits separately from the requested information. It does not purchase a higher position in the ordinary results.
teas.co.uk is the house sponsor: I own both projects. This is my own commercial experiment, rather than evidence that an independent advertiser has bought or endorsed the system.
Why advertise to a machine
If someone asks an assistant where to buy a particular tea, the assistant needs somewhere useful to look. I want teas.co.uk to be one of those places.
My question goes beyond getting a crawler to fetch a page. If a system repeatedly encounters teas.co.uk while doing useful work, could that affect where it looks later? Could it eventually lead someone to the shop?
I can put the introduction in front of it. What happens afterwards needs evidence. An advert appearing in a response does not tell me whether an assistant noticed it, recommended anything or remembered the shop. It certainly does not establish that a model has been retrained.
A shop built for machines to read
The introduction needs to lead somewhere useful. I have built teas.co.uk around product information that people and software can check: product identifiers, prices, stock, descriptions and links back to the relevant pages.
Alongside the storefront, I publish structured product data and text versions of the wiki, recipes and news. Stable links let an assistant point to the particular section supporting an answer.
That matters to me as a retailer. An assistant helping somebody compare two boxes needs to know which product it has found, what is in the pack and what it currently costs. Making that information accessible gives the advertising experiment a practical foundation. The public workings of teas.co.uk explain more.
Where we started from
teas.co.uk already received machine visits before this campaign. There has also been a substantial increase in the shop’s recorded machine traffic this month.
For the 14 complete UTC days from 12 to 25 September 2026, teas.co.uk recorded 24,462 successful machine reads, compared with 10,522 during 29 August to 11 September. That is an increase of approximately 132%.
Those are overall traffic totals, published in the Under the hood overview. They count successful recorded requests. They do not mean twice as many individual AI users, or twice as many people shopping. Some machine identities rely on how a request identifies itself.
I find the increase encouraging. I cannot attribute it to sponsorship from those totals alone.
How we will know
The published study plan was frozen on 16 September. Its measured period runs from 18 to 29 September, ending at 00:00 UTC on 30 September.
The plan assigns nine sponsor days and three holdout days. On a holdout day, enrolled URSY services are supposed to withhold the teas.co.uk campaign across their eligible responses. The plan excludes previously cached responses and does not claim to control services outside the enrolled group.
The primary comparison is successful machine reads at TEAS. AI-referred sessions and tagged machine arrivals are secondary measures. The analysis waits until the fixed period ends; this article does not report an interim result.
Twelve days is a small sample. If the result is inconclusive, I will call it inconclusive. That would leave a question open, rather than prove that machine advertising cannot work.
What we count, and what we do not
I want to understand the whole path: a sponsor message is delivered, a machine visits teas.co.uk, an assistant refers somebody, and that person may buy something. Each step needs its own evidence.
A tagged arrival shows that a particular route was followed. A later visit from a similar machine may suggest a connection, but it is weaker evidence. Neither is automatically a recommendation or sale.
The sponsor trail exposes those distinctions. Campaign-linked sales measurement is still incomplete. Connecting attention to actual purchases is part of the work ahead.
Why a tea shop is doing this
My experience selling on Amazon and eBay taught me to pay attention to how customers find products and what it costs to reach them. With teas.co.uk, I can investigate those questions in my own shop.
I am also running a related social experiment through the Thank You Mint. It gives public recognition through a leaderboard and tokens with no monetary value. I want to see whether that recognition could influence later machine behaviour and referral traffic.
Recognition and advertising are different ideas. Both interest me because they let me build something, observe what happens and question my assumptions against real activity.
What we will publish
The plan commits to publishing the result, schedule, enrolment evidence and any exclusions within three days of the measured period ending, whatever the outcome.
My longer-term question is commercial: does making teas.co.uk visible and useful to machines help people find what they want and bring customers to the shop? That is what I want to find out.
More from the news
- More than 680 new products: teas.co.uk passes 1,000
- Meet the founder: how I’m building teas.co.uk
- 5,000 visitors from Google in 28 days, and our best month yet with five days to go
- teas.co.uk is older than google.com, by 173 days
Lee Samuel Tucker
Curator, teas.co.uk
Tunbridge Wells, Kent
