Version two of Teas.co.uk went live on 10 May 2026. Ninety days later, with no paid advertising, the site had recorded 42,524 credited machine-served reads and 736 sessions arriving from an AI answer surface. Those numbers are encouraging. They are not a conversion claim, and they do not share the same measurement window.
Two launches, one direction
The first version of Teas.co.uk launched in November 2025. That initial shop established the catalogue and showed that a small independent tea business could publish unusually detailed product and editorial information. Version two went live on 10 May 2026 and rebuilt the site around a stricter idea: one governed set of product facts should be readable by people, search engines and AI systems without the business paying for placement.
The site is still a real shop with real stock, baskets and orders. It is also an instrument. Under the Hood publishes the counting rules, source tables, windows, corrections and deliberate zeros rather than reducing everything to one impressive-looking total.
The headline figures, with their windows
| Figure | What it means | Measurement window |
|---|---|---|
| 42,524 | Credited reads served successfully to named machine agents, status 200 or 206 with bytes served, excluding an agent withdrawn from credit | 2 June–8 August 2026 |
| 88,796 | All logged attempts in the machine-read ledger before the published exclusions are applied | 2 June–8 August 2026 |
| 736 | Sessions whose referrer named an AI answer surface or which were flagged AI-attributed | 10 May–8 August 2026 |
| 17 | Distinct named machine agents written to the citation table | 2 June–8 August 2026 |
| 26,428 | Stable citation anchors published across 2,474 addressable pages | Snapshot on 8 August 2026 |
| £0 | Paid advertising used to create this discovery | Version-two launch through the snapshot |
Why the date distinction matters
This is a 90-day version-two milestone, but the credited served-read ledger begins on 2 June. It would therefore be wrong to present all 42,524 reads as a complete 10 May–8 August series. The 736 AI-referred sessions do cover that full 90-day period. Publishing both windows makes the result less tidy and more useful.
The same discipline applies to the 23,406 later requests carrying a strong-confidence trace of an earlier machine offer. Those are requests, not 23,406 people, customers or citations. The site does not add them to the 736 sessions.
What happened without advertising
Major machine agents found the site, read successful responses, and returned for exact sources they had previously been offered. That supports a narrow but important conclusion: making a retail site easy to retrieve, parse and verify can earn machine discovery without buying reach.
It does not yet prove that the shop was visibly cited inside an answer. A website can see a crawler arrive and can sometimes see a person arrive from an AI surface; it cannot see what happened inside every answer engine. The site’s citation-token counter therefore remains at zero until there is stronger proof.
What the commercial evidence says
The snapshot recorded four add-to-basket events and four checkout starts from AI-referred sessions, against zero completed AI-attributed orders. Order-event coverage is still incomplete, so no AI conversion rate is published. A session is not a customer, a basket is not a sale, and a machine read is not a recommendation.
That restraint matters because the easiest way to make a young dataset sound successful is to move between those definitions mid-sentence. Under the Hood deliberately treats them as four stages: what the site published, what machines successfully read, what came back as people, and what can be verified inside an answer.
What comes next
The editorial test is now live: 25 new Wiki buying guides built around current stock and genuine comparison gaps. Every guide connects an informational question sideways to relevant Wiki explanations, brand archives, product categories and in-stock teas. The completed set ranges from Yorkshire Tea, PG Tips and Tetley comparisons to organic, chai, peppermint, Darjeeling and white-tea shortlists, plus a fact-checked Ceylon citrus comparison between Dilmah Earl Grey and Hyson Lemon.
The measurement plan is equally explicit: impressions and clicks, Wiki-to-shop journeys, assisted baskets, and any sign that a new article competes with an existing one. New words are not the result. Better routes and clearer decisions are.
Read the evidence
- Under the Hood home — the four-stage model and current headline ledger.
- The Numbers — inclusion rules and timescales.
- The Machines — named-agent behaviour.
- The Workings — corrections and reconciliation.
- The version-two launch article.
Snapshot reviewed 8 August 2026 at 11:19 UTC. The November 2025 version-one launch date is founder-supplied; the version-two date and all metrics above are reflected in the live Teas.co.uk evidence surfaces.
