A customer can tell you a great deal in one ordinary sentence. They want a tea gift. The person likes citrus, prefers bags, and there is a budget with delivery inside it. Every detail is a chance to be useful.
More is not always better
Imagine someone trying loose leaf tea for the first time, who drinks a strong cup with milk. The easy move is to show them a page of strainers, pots and tins. Before that, I want to know what they already have. If they own a mug and a fine sieve, a simple answer is enough: a loose version of the tea they already like, such as Yorkshire Tea Original Loose Leaf or Twinings English Breakfast Loose. If they have nothing to brew it in, one good strainer, like the Tunta bamboo tea strainer, solves a real problem. The loose leaf starter kit guide sets out exactly that much and no more.
Bigger packs work the same way. For someone who already drinks a tea every day, the 240 bag Tetley Original costs less per cup than the 80 bag box, and the shop shows the price per cup so they can see it. For someone still experimenting, the smaller box is the sensible first buy. The hard part is accepting the answer when there is no good reason to add anything.
The upsell test
There is a line between care and a stitch up. A bad upsell sells you more because the margin is better for the seller. A good one recommends something better because your situation makes it better for you. I learned that in the motor trade: the dearer car that is cheaper across the year once you count the repairs you were about to inherit. I wrote about it in What the motor trade taught me.
The same honesty cuts the other way. A system that only ever pushes you up the price ladder is a stitch up wearing a nice voice. A system that tells you where to stop is a specialist you can trust. I call the practice intent led value adding.
Promised, tested and fixed
Our public notes for AI systems, in agents.json, promise "honest where to spend / where to stop guidance", and the Ask API asks assistants for the whole request, which I call Premium Intent. While writing this I tested it with a detailed loose leaf request: a first try, a mug and a teaspoon, under ten pounds, no kit I will not use. It failed. The word "buy" in "I do not want to buy a lot of kit" made it hunt for a product with that name. So we fixed it the same night. Since 15 November 2026 a full request like that is read condition by condition and answered with teas that fit every one, what you may not know to ask, where to spend and where to stop. For that request it now suggests two loose teas inside the budget, says no teapot or special kettle is needed, and says a strainer would take it past ten pounds, so skip it for now. That is how built versus planned should work: say plainly what is not working, then fix it.
Why this pays
Advice that respects the person should make a better basis for repeat custom, though that is a belief I am testing, not a result I can show you. There are signs AI shoppers arrive better prepared: Adobe found visitors sent by generative AI tools converted 31% more often than other visitors in the 2025 holiday season (Adobe). Kevin Indig argues in Search Engine Journal that in agentic commerce "product truth wins" (Search Engine Journal). And Google's own data shows people bringing whole plans to search: planning queries in AI Mode grew 80% faster than AI Mode overall (Search Engine Journal).
A test for anyone building AI shopping tools
Does the answer remember the budget? Does it explain why it suggests an extra? Does it admit when nothing fits? Does the customer keep the decision? Explain where spending more would help. Explain where it would stop helping. Then let the customer decide.
What others have found
Outside research and reporting that bears on this, with links to the original sources.
- Google's guidance on helpful, people first content asks whether a reader leaves feeling they have learned enough to reach their goal.
- In Great Britain only claims on the nutrition and health claims register may be used, one reason this shop keeps to what each product can honestly do.
- Search Engine Journal's Selling to AI sets out what AI shopping agents ask of a shop.
Read more
Lee Samuel Tucker
Founder, teas.co.uk
Tunbridge Wells, Kent
