Tea Shop Marketing Ideas: Tasting Notes and Gift Sets Written With AI

Taste each tea yourself, write raw notes into a fixed template, and let AI turn them into shelf cards, product pages and emails in your shop's voice. Price subscription tiers and gift sets in a spreadsheet by cup count and margin, then use AI only for names, copy and rotation plans, never for flavour facts.

AI cannot taste, so any origin, harvest date, award or health benefit it adds to a tea is invented unless it came from your supplier's sheet or your own cup. Everything below is for a retail shop selling leaf by weight, not a café serving cups, so the work is product copy, brew cards, parcels and gifting rather than a menu.

Follow me on Instagram@sagnikteaches

A tasting template that stops AI inventing flavours

Generic AI tea copy all sounds alike because the model is handed a name and asked to imagine. Hand it your cup instead. Record every tea in the same fields, split into two kinds of fact: what the supplier's sheet says, and what you tasted. The model may use those two blocks and nothing else. The third column below shows the template filled in for one tea, as an illustration.

Connect on LinkedInSagnik Bhattacharya

The prompt below turns that row into three pieces of copy. Its two important lines are the ban on facts that are not in the blocks and the list of hype words, because both stop the model reaching for the usual brochure language.

Subscribe on YouTube@sagnikteaches

You write product notes for a loose-leaf tea shop. Voice: warm, plain, specific.
Short sentences. Never use: luxurious, exquisite, divine, indulge, detox,
cleanse, superfood, "perfect for".

Use ONLY the facts in the two blocks below. If a fact is not there (origin,
harvest, altitude, awards, certifications, health effects), do not write it.

Write:
1. A shelf card of 15 words or fewer.
2. A product-page note of about 60 words.
3. One "not for you if" sentence.

SUPPLIER SHEET: oolong, roasted. Origin and harvest not stated. No certification.
MY TASTING: dry leaf dark, tightly twisted, toast and dried dates. Wet leaf warm
bread crust, a little caramel. Clear amber. Roasted grain first, then sweet.
Medium body, slightly syrupy. Finish medium-long, stone fruit, no bitterness.
Four good steeps. Tested: 3 g in 200 ml, 90 C, 3 min then +1 min each.
Pairs with dark chocolate and toasted nuts. Not for anyone wanting a grassy,
floral oolong.

A reply in the right shape, shown here as an illustration, reads like this.

Shelf card: Roasted oolong: toast and dried dates, a syrupy cup, four good steeps.

Page note: A dark, tightly twisted leaf that smells of toast and dried dates. The
cup is clear amber, starting with roasted grain and turning sweet, with a medium,
slightly syrupy body and a long stone-fruit finish, no bitterness. Brew 3 g in
200 ml at 90 °C for 3 minutes, then add a minute per steep; four infusions stay
good. Lovely with dark chocolate or toasted nuts.

Not for you if: you want a grassy, floral oolong.

Now compare the "before". A model given only a tea's name and asked for a description tends to return something like this (an illustration, not a quote from any shop): "Indulge in this exquisite roasted oolong, a luxurious experience with rich, complex notes that will transport you." Nothing in that sentence can be checked against the cup. The second version can, because every phrase traces back to a row in the table. When a model still slips something in, it is typically "hand-picked", "grown at high altitude" or "rich in antioxidants". Strike all three. None came from your sheet or your palate, and the last is a health claim.

To make the check routine rather than occasional, keep a small claims ledger beside the draft. It takes a minute per tea and settles every argument about whether a line stays.

Brew cards and cup counts: the numbers customers actually use

A tasting note sells the tea; a brew card gets it brewed well, which is what brings the customer back. The card is a line of settings you have tested, and AI only formats it. Tea retailers' guides disagree on exact figures, by several degrees and a minute or two, so the ranges below are common starting points, not a standard. Taste, adjust and write down your own numbers for each tea before they go on a page.

The cup count is the number customers compare with a tea bag, and a spreadsheet should calculate it, not a model. Take 2.5 g a cup. A 50 g pouch then makes 20 cups, and at $9 that is $0.45 a cup. A 100 g tin at $16 makes 40 cups, $0.40 each. Put the formula in the sheet, put the result on the page ("makes about 40 cups"), and tell the prompt to copy the figure rather than recalculate it. Models like round numbers, and a tin that "makes about 50 cups" is the sort of small overstatement a customer notices by week three.

Batch workflow: 64 teas, two tasting sessions, one review pass

The example shop has one counter, a web store and 64 teas. Rewriting everything at once is a project nobody finishes, so start with the 20 teas that sell most or that staff recommend. The steps run in this order.

1. Cup ten teas in a session, with the template open. Speak each note into your phone as you taste, using the field names as headings, and let a transcription tool turn it into text. Typing mid-tasting breaks the rhythm and shortens the notes.
2. Paste the transcript into the tasting column and the supplier's facts into a separate column. Keeping them apart is what lets the prompt treat them differently.
3. Add a word bank to the prompt: the 40 or so flavour words you actually use (toast, hay, malt, stone fruit, cut grass, toasted rice). Tell the model to prefer those words. This is the cheapest way to make the whole catalogue sound like one person wrote it.
4. Run the prompt in batches of five teas, and store the three outputs in their own columns: shelf card, page note, "not for you if" line.
5. Review each one against the claims ledger. Read the note aloud, and if you would not say it across the counter, rewrite it.
6. Publish the page note to the product page (the tutorial on Shopify product descriptions with AI covers the store side) and print the shelf cards.

The sums, using timings that are assumptions: about 3 minutes a tea to taste and dictate when the leaf is already out, and about 4 minutes to review. For 64 teas that is 64 x (3 + 4) = 448 minutes, roughly 7.5 hours across two or three sessions. Writing three versions of each note from scratch at about 15 minutes a tea would take 16 hours. The saving is real, but most of what remains is your tasting time, which is the part customers value.

Subscription tiers priced from cup counts and parcel weight

Design tiers from the weight in the parcel, then price them from cost. The numbers below are illustrative assumptions: your leaf costs 40% of its shelf price, which at an average shelf price of $16 per 100 g is $6.40 per 100 g; packaging is $2.50 a parcel ($3 for the larger one); postage is $4.80 ($5.60 for the larger one); and card fees are 2.9% plus $0.30, which is Shopify Basic's list online rate. Swap in your own figures.

The cheapest tier is the one that breaks. At $15 the Taster earns $2.16 a parcel, and one failed delivery or a refunded parcel wipes out several months of profit. Raising it to $19, or moving to a lighter letterbox parcel if your carrier offers one, fixes it before a single customer has signed up. This is the job to give a spreadsheet and to ask AI to check, not to do: paste the table, ask it to find the weakest tier and explain why, then confirm the arithmetic yourself. The subscription box margin model tutorial goes deeper on this, and costing each blend covers the leaf-cost side for house blends.

Once the tiers hold, AI is useful for four jobs. It writes the tier names and one-line promises. It drafts the welcome email, a skip confirmation and a renewal reminder. It writes the insert card for each parcel from the three tasting notes in it. And it proposes a rotation from your catalogue sheet, with rules written into the prompt: no tea repeats within six months, every Explorer parcel has one caffeine-free pouch, and only teas marked "in stock" may be chosen. A model must be told to choose from the list, because left alone it will happily propose a tea you do not sell.

Check each proposed month against the three rules before you order stock. A draft that puts a smoked black in the month 1 wildcard slot looks like a sensible contrast to the green and the oolong, yet it leaves that parcel with no caffeine-free pouch, and a subscriber who drinks tea in the evening notices.

On the store side, Shopify's own Subscriptions app lets customers skip or cancel from their account, and you can skip, pause, resume or cancel a contract from the admin. Shopify's help page says cancelling cannot be undone and that uninstalling the app deletes the plans and contracts made in it, along with discounts that use it, so settle on the app before you take sign-ups. Check the app listing for any fees. A shop too small for an app can run a tea club by invoice for the first ten members, which is also the cheapest test. The tutorial on testing a subscription offer with AI first shows how to count sign-ups before you build anything.

Gift sets: three bundles priced to keep the margin

Gift sets sell on occasion, not on tea type, so build one for each: the morning set, the evening set and a discovery set for people who do not know what to choose. The numbers are illustrative and use the same assumptions as above, with leaf at 40% of its shelf price; the Morning set's $23 shelf value includes a $6 tin, which carries no leaf cost. Postage is charged on top for online orders.

Each set takes roughly a 9-11% discount on shelf value and still clears 38% or more after fees. If you want a deeper bundle discount, the tutorial on bundle pricing and discount margin shows how far you can go before it stops paying. Two practical checks apply to every set. Look at the best-before dates on every pouch going in, because a gift opened in three months should not arrive with six weeks left. Weigh a packed set with its box and card before you set the postage line.

AI does the words. Give it the set contents and the tasting notes and ask for a name, a 40-word description and a gift-card insert. The prompt and a typical result for the morning set follow.

Write a gift-card insert (max 45 words) for the Morning set: a malty black tea and
a sencha-style green tea, plus a tin. Use only the tasting notes below. No claims
about health, energy or origin. End with one brewing tip.

Notes: Black: malt, toast, rounded, holds a splash of milk. Green: fresh cut grass,
sweet finish, brew at 75 C for 2 minutes.

Two teas for the start of the day: a malty black that takes milk well, and a green
that tastes of fresh-cut grass with a sweet finish. Brew the green at 75 °C for two
minutes, and let the water cool a little before you pour.

That reply is 46 words, one over the limit, and "let the water cool a little" is a tip the notes did not supply. A model asked for a word limit often misses it by a few, and one that is asked for a brewing tip may invent one. Fix both by hand, or tighten the prompt to "use the brewing line from the notes only". Finally, add a short staff script from the same notes: "If they like milk in their tea, the Morning set; if they drink tea late, the Evening set; if they are buying for someone whose taste they do not know, Discovery." Staff find a line like that easier to use at the counter than a long description.

Six marketing ideas that reuse the same notes

Once the notes exist, each of these is a small job, and AI can draft the words for every one.

• Shelf cards with a QR code. The 15-word card sits on the shelf, and the code leads to the product page with its brew card. Count scans per tea, and promote the ones that get scanned but not bought.
• A weekly counter flight. Three teas chosen to contrast (a green, a roasted oolong, a malty black), poured in small cups with a note card each. The prompt takes the three notes and writes the cards and the Instagram caption. Count how many flight tasters leave with a pouch.
• A sample sachet in every parcel. Pick a tea in the same flavour family as their order and add a "if you liked X, try Y" line built from your notes. Count reorders of sampled teas within 60 days.
• A brew-of-the-week email. About 90 words: one tea, one brew card, one pairing. As of October 2026 Mailchimp's Free plan allows 250 contacts and 500 sends a month, so four weekly sends to 125 people use the whole allowance, and it cannot schedule a send, so someone presses the button each week; paid plans start at $13 a month for 500 contacts.
• Refill pouches. A cheaper pouch than a tin, with a card that says "refill your tin" and the tin's note repeated. Count the share of customers who refill.
• Workplace tea boxes. A small-office version of the Explorer tier with three teas, 150 g, monthly. Have AI draft a three-line introduction to send to ten nearby offices, then follow up once.

What AI must never write on a tea page or label

The brief for every run is the same. Origin, harvest season and estate are supplier facts and appear only when the sheet states them. Certification words such as organic or fair-trade appear only with a certificate on file. Health and wellness claims do not appear at all, and the tutorial on health food store posts without health claims shows what to write instead. Caffeine statements should be general, such as "contains caffeine" for true teas, rather than numbers per cup, because levels vary with the leaf and the brew. Allergen lines come from the supplier's specification for flavoured blends. Scarcity wording such as "limited edition" needs a real limit behind it.

Subscriptions add one more rule. Say the price, how often you charge and how to skip or cancel in plain words on the sign-up page and in the first email. Consumer rules for recurring payments vary, so check yours with an adviser before you launch.

A four-week test before you rewrite all 64 pages

Treat the first month as an experiment. Rewrite the notes for 10 of your 20 best sellers and leave the other 10 alone. Over four weeks, compare each rewritten product's add-to-basket rate with its own rate over the previous four weeks, then set that beside the same comparison for the ten you left alone, which shows what the season did to everything. Small shops see small numbers, so a gap of a few percentage points on a few hundred visits is not proof; look for a consistent direction across the ten and treat anything else as a draw. If the rewritten notes do no worse and customers have stopped asking "what does it taste like?", carry on.

Test the subscription separately, and before building it. Offer the Regular and Explorer tiers to your email list for 14 days as a pre-order with a start date. If fewer than ten people choose a tier, the idea needs a different offer, not more automation. For gift sets, make a limited run of 40 and see how many sell by the middle of the gifting season, then reorder boxes for the rest. These thresholds are my own rules of thumb, so adjust them to your list size.

Keep the work honest by dating it. Each tea row carries a "last tasted" date, and a new supplier batch triggers a retaste, a new note and an update to the shelf card. Once the loop runs for a month, the tasting notes, tiers and sets can feed one another: a tea that earns repeated compliments in the counter flight becomes the wildcard in next month's parcel and the lead in next season's gift set.

Questions a tea retailer asks after the first draft

Can I say a tea helps with sleep, digestion or stress?

Leave health claims out of AI-written copy and your own. Rules on health claims for food and drink are strict in many places and AI will happily add them. Describe taste, aroma, ritual and brewing instead. If you want to say anything about effects, ask a food-standards adviser or solicitor for the rules that apply where you sell.

Does a tasting note stay valid when the next harvest arrives?

No. A note describes one batch. Add a 'last tasted' date to every row, re-taste when the supplier ships a new lot, and rerun the prompt on the new note. Keep the old note live only while the old stock lasts, because the flavour can shift between lots.

Can I use AI-generated pictures of my teas on product pages?

Use real photos of your real tins and leaf. A generated image shows a tea you may not sell. If you list products on Google, its product-feed rules ask for metadata marking AI-generated images, so check Google's current page for the exact wording before uploading any.

What about allergens in flavoured blends?

Take allergen information from the supplier's specification for each blend, never from an AI inference. Flavoured and dessert-style blends can contain nuts, cereals or milk derivatives. Put the supplier's wording on the label and page, and ask the supplier for a fresh sheet whenever a recipe changes.

Further reads

• AI Marketing Automation for Small Businesses: Chocolate Gift Seasons — A gifting-season plan from another small food retailer.
• How AI Can Help in Sales: A Gift Shop's Bundle and Follow-Up Plan — Bundle and follow-up ideas from a gift shop.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Shopify's help pages on Shopify Subscriptions; tea retailers' brewing guides (ranges differ, so treated as starting points); Shopify's list online card rate (used as an assumption); Mailchimp plan pages; Google's product-feed guidance on AI-generated images.