Ecommerce SMS Marketing: Draft With AI, Then Check Consent Rules

Draft the texts with AI, but build consent and timing checks into your process, not the prompt. US marketing texts generally need documented written consent. UK ones need consent too, with a narrow exception for existing customers. Send only in local daytime hours, include a clear opt-out and have a person approve every campaign. This is a pointer, not legal advice.

The common mistake is treating a phone number collected at checkout as permission to market by text. It is not, in most cases. A number given so the courier can reach you is for delivery. Marketing needs its own, separate, recorded yes. Everything else here, from the length of a message to the wording of the opt-out, is easier than that single point.

Follow me on Instagram@sagnikteaches

Consent first: what permission you need, and how to show it

Ask a qualified adviser to review your set-up, especially if you sell to customers in more than one jurisdiction. What follows are the pointers an owner should know before drafting a single message.

Connect on LinkedInSagnik Bhattacharya

United States. The federal rules that matter here come from the Telephone Consumer Protection Act (TCPA) and the regulations the Federal Communications Commission (FCC) wrote under it. The rules define "prior express written consent" as an agreement in writing, bearing the signature of the person called, that clearly authorises the seller to deliver advertisements or telemarketing messages using an automatic telephone dialling system or an artificial or prerecorded voice, and that names the number to be used. The same definition requires the agreement to tell the person that they are authorising such messages and that they do not have to sign it as a condition of buying anything. In practice, shops capture this with a sign-up form or checkout tickbox whose wording says exactly that, kept as a dated record. The regulations apply to text messages as well as calls. The cost of getting this wrong is real: the statute allows a person to recover $500 for each violation, or actual loss if greater, and a court may increase that up to three times for a wilful or knowing violation. Claims can arrive over a handful of texts, so consent records are not paperwork for its own sake.

Subscribe on YouTube@sagnikteaches

United Kingdom. Marketing texts fall under the Privacy and Electronic Communications Regulations (PECR), whose rules on "electronic mail" cover texts as well as emails. The default is that you need the recipient's consent, and it should be a clear positive act: data-protection law's consent standard says silence, pre-ticked boxes or inactivity do not count. There is an exception, usually called the soft opt-in, that can let you text your own customers without fresh consent. In outline, every condition must hold: you collected the number yourself, during a sale or negotiations for a sale; you market only your own similar products or services; and you gave a simple, free way to refuse when you collected it and give one again in every message. Bought-in lists cannot use it. The Commissioner's guidance sets the conditions out in full, so read it and take advice before relying on the exception, and when in doubt collect explicit consent.

A practical rule for a shop selling in both: collect explicit, separate consent for marketing texts from everyone, and keep proof. It is simpler than running two regimes, and it holds up under both.

The consent wording and the record behind it

Put the text-marketing question on its own line, unticked, apart from your terms. Do not bundle it with "I agree to the terms". Here is wording a shop might start from, to be checked by an adviser against current rules:

[ ] Yes, send me offers and updates by text message from [Shop name] to the mobile number above. These may be sent using automated technology. Consent is not a condition of buying anything. Message frequency varies (about [number] a month). Reply STOP at any time to opt out. See our privacy notice: [link].

Customers who subscribe through a pop-up or a form should see the same wording beside the field. If the first text you send says what they signed up for ("Welcome to [Shop] offers. About 4 texts a month. Reply STOP to opt out"), you also confirm the opt-in in a way many platforms treat as good practice.

Then keep a record for every number. At minimum: the phone number, the date and time of the opt-in, the exact wording shown, the source (checkout, pop-up, in-store tablet) and, where your platform provides it, the IP address or other proof of submission. Most messaging platforms store this in the subscriber record. Check yours, because when someone complains, the record is your defence and an empty field is a problem. Do not import numbers from a spreadsheet of past customers without a record of consent, and never buy a list.

Timing: quiet hours that work for every customer

The FCC's rule bars telephone solicitations, which include marketing texts, before 8 a.m. or after 9 p.m. at the called party's location, and some US states set tighter limits (at least one closes the window at 8 p.m.), so ask an adviser what applies where your customers live. For UK recipients, read the Commissioner's current guidance on timing; whatever it says, daytime sends are what customers expect everywhere. The simplest policy is one stricter window for everyone: send marketing texts between 9 a.m. and 8 p.m. in the recipient's local time, and none on the early mornings, late evenings or public holidays you choose to skip.

The phrase "recipient's local time" is where small shops slip. A shop that schedules a 10 a.m. send in its own time zone reaches customers on the other side of the country at 7 a.m. Use your platform's "send in recipient time zone" setting if it has one. If it does not, group subscribers by the area code or shipping region and schedule one send per group. Ask the assistant to build the schedule so you do not do the arithmetic by hand:

We send marketing texts only between 09:00 and 20:00 in each recipient's local time. Our shop is in [time zone]. Our subscribers are in these time zones with approximate shares: [list]. Propose three send times in our own time zone for a Thursday campaign so that every group receives it inside the window, with the local arrival time for each group, and say which groups cannot be reached by a single send.

Check the output yourself with a world clock. A common error is mixing up daylight saving dates, which differ by place and change twice a year. If your subscribers span many zones, use the platform's time-zone feature rather than a hand-built schedule.

Opt-out wording and what happens when someone replies

Every marketing text needs an easy way to stop. The standard solution is the line "Reply STOP to opt out" at the end of every marketing message. Some platforms only insist on it in the first message and at intervals, but it costs a few characters, removes the doubt, and for UK customers texted under the soft opt-in every message must offer a way out anyway. The US rule treats replies of "stop", "quit", "end", "revoke", "opt out", "cancel" or "unsubscribe" as a valid revocation, and requires you to treat other wording as a request to stop if a reasonable person would understand it that way. So "please remove me" or "no more texts" must be honoured, not ignored because it is not the keyword.

Make sure that happens in practice. Check three things in your platform: that it honours those keywords automatically, that someone reads replies that are not keywords, and that opted-out numbers are blocked from every future campaign including automations such as abandoned-cart texts. A customer who replies STOP and then receives a cart reminder an hour later is the kind of complaint that becomes a legal claim. Send the one confirmation text your platform allows ("You're unsubscribed from [Shop] texts. No more messages.") and then nothing.

The AI can help you design the reply handling, but test it with your own phone before you launch. Subscribe, receive a message, reply STOP and confirm that you get the confirmation and nothing afterwards. Then reply "please stop texting me" from a second test number, and see what your platform does. If it does nothing, you have a gap you must close by reading replies daily.

Length: 160, 70 and the emoji that doubles the bill

A standard text holds 160 characters in the usual GSM alphabet. Twilio's glossary says that once a message contains a character outside that set, such as an emoji or a curly apostrophe, it is sent in a different encoding that holds only 70 characters, and longer messages are split into segments of 153 or 67 characters that are billed one by one. Many platforms bill per segment, so a message you thought was one text may be sent and charged as two or three.

A quick sum with illustrative numbers shows the cost. A 120-character message with plain characters is one segment. Add one emoji and the same message switches to the 70-character limit, so it is 2 segments. At 8,000 subscribers that doubles the number of segments sent for the campaign, from 8,000 to 16,000, for the sake of one emoji. Your platform's price per segment, which varies by provider, turns that into a bill. Platforms often have a preview that shows segment counts. Use it every time.

Ask the assistant to write within the limit and to avoid curly punctuation:

Write 3 versions of a marketing text for a small homeware shop's weekend sale: 20% off all candles until Sunday midnight, code CANDLE20, link [short link]. Each must be under 140 characters including the opt-out line "Reply STOP to opt out", use only plain keyboard characters (straight quotes, no emoji, no accents), start with the shop name "[Shop]", and include the end date. Show the character count for each.

Then count the characters yourself, because assistants often miscount. A sample output: "[Shop]: 20% off all candles until Sun 11:59pm. Code CANDLE20. [short link] Reply STOP to opt out" is about 100 characters. Check that the discount, the end date and the code match your actual offer, and that the link works on a phone.

A before and after shows what the edit catches. An unedited draft, pasted from a phone's notes app, read: "[Shop] [party emoji] It's here! Our biggest candle sale ever - 20% off everything with code CANDLE20 until Sunday. Don't miss out! [short link]". On the phone the apostrophes were curly and the dash was an en dash, so with the emoji the message went out in the 70-character encoding: about 130 characters, two segments. It also claimed "biggest ever", which the shop would need to be able to back up, and it had no opt-out. The edited version: "[Shop]: 20% off candles with code CANDLE20 until Sun midnight. [short link] Reply STOP to opt out". That is 97 plain characters before the placeholders are filled, one segment, with the offer, the deadline and the way out all visible.

Drafting with AI: voice, offer and a person's sign-off

Texts are the shortest marketing you write, which makes the offer and the first five words do all the work. AI is good at producing variants quickly, so give it three things: who the customer is, the one thing you want them to do, and the rules above. Ask for variants, then pick, and rewrite in your own voice, because texts that sound like every other shop's blur together.

You write marketing texts for [Shop], a small online homeware shop. Customers opted in to hear about offers. Voice: friendly, plain, no hype words, no exclamation marks, no emoji. Each text: under 140 characters, starts with "[Shop]:", has one offer, one deadline, one link placeholder [link], and ends "Reply STOP to opt out". Write one text each for: (1) a weekend sale; (2) a restock of a sold-out best seller; (3) a back-in-stock reminder to people who asked to be told; (4) a thank-you with a small discount for repeat buyers. Do not invent offers, dates or discount amounts: use [offer] and [date] placeholders where I have not given them.

Notice the last sentence. Without it, assistants will invent a "20% off" and a Friday deadline. A promotion you did not mean to run and a text already sent is an expensive mistake. Fill the placeholders with real values after reading the drafts.

The person's sign-off is the final control. Before any campaign goes out, a named person reads the final text on a phone, checks the offer against the live store, the link and the code, confirms the audience is the opted-in list only, and confirms the send time. A five-line checklist pinned above the scheduling screen works well:

1. Audience is the opted-in segment, with opted-out numbers excluded.
2. Text names the shop, carries one offer and an end date, and ends with the opt-out line.
3. Characters plain, segments counted, no emoji unless you accept the double count.
4. Send time inside the recipient-local window.
5. Offer, code and link tested on a phone today.

A month of texts for a small homeware shop

To see the plan in use, picture a shop with 6,000 opted-in subscribers and a four-week plan. The rule it follows: no more than four marketing texts a month, because frequency is the main reason people opt out. Week one: a sale announcement on Thursday at 11 a.m. recipient time. Week two: nothing. Week three: a restock text, sent only to the 400 people who tapped "tell me when it's back" on that product, whose consent covers exactly that. Week four: a thank-you text with a small discount to customers with two or more orders. Two sends are to the whole list, two are to small, relevant groups.

The owner tracks three numbers for each send: the click rate on the link, the opt-out rate and the revenue attributed to the link. If an ordinary send brings more than about one opt-out per hundred recipients, she shortens the next message or skips a week. The shop's rule of thumb comes from its own numbers, not from a benchmark, and you will build yours the same way. The tutorial on building an upsell and cross-sell sequence covers sequencing across email and text, and finding bundle pairs in your orders gives you offers worth a text.

Segments make a small list work harder

Sending everything to everyone is the fastest way to lose subscribers. A small shop gets more from three or four segments, each defined by something the customer did and each covered by the consent they gave. Useful ones: people who asked to be told when a product returns, customers with two or more orders, people who joined in the last 30 days (who get a short welcome series), and everyone else. The first group's consent is narrow, so only text them about that product. Keep the definitions in a one-page note so a colleague sending next month does not widen them by accident.

Ask the assistant to propose segments from the fields you actually hold, listing them first so it does not invent data you do not have:

Our customer records hold only: opt-in date, opt-in source, number of orders, date of last order, total spent, product categories bought, and "notify me" requests. Suggest four text-message segments from these fields only, say what each should be texted about, and flag any segment where the customer's consent might not cover the topic.

A sample answer would propose a welcome segment, a repeat-buyer segment, a lapsed segment (no order in 180 days) and a notify-me segment, and would flag that the notify-me group agreed to hear about one product only. Take that flag seriously. It is the assistant noticing the consent boundary, which is exactly where a text-marketing programme usually goes wrong.

What went wrong when a shop texted its checkout numbers

Here is an illustrative mistake of the kind that arrives as a letter. A small shop exported every phone number captured at checkout over two years and loaded the list into a texting platform for a clearance sale. The checkout form said the number was for delivery updates. Several customers complained, one replied "STOP" and still got the next campaign because the import had not been matched against the platform's block list, and the shop received a demand letter citing the statute's per-text damages.

The recovery was costly: the shop stopped all marketing texts, deleted the imported list, kept only the numbers with a documented opt-in, and rebuilt its list from scratch through a checkout tickbox and a pop-up. It took months to get back to a quarter of its old list size. The cheap fix is to start the right way: tickbox first, record kept, test STOP before launch. A smaller list of people who said yes is worth more than a large list of people who did not.

Texts that are not marketing: service messages

Order confirmations, dispatch notices and delivery updates are service messages, and they follow different rules from promotions. The line is the content: a text that says your parcel is out for delivery is service, while one that adds "and enjoy 10% off your next order" has become marketing and needs marketing consent. Keep the two apart in your platform, with separate templates, and do not tuck an offer into a delivery notice for the sake of one extra click. If you send service texts, still honour opt-out requests and tell customers how to stop them. For the conversational side of customer contact, the scripts in seven store conversations to copy apply equally to replies by text.

Review every quarter

Rules and platform features change, so put a review in the diary every three months. Check that the consent wording still matches your practice, spot-check ten random subscribers and confirm each has a complete consent record, review your opt-out rate and message frequency, and read the platform's current policy page for carrier or compliance changes. Ask your adviser to look at the wording once a year, or whenever you start selling into a new market. A short log of what you checked and when is cheap, and it shows regulators and platforms that you take permission seriously, which is the quality that matters most in this channel.

Questions about texting customers

Can I text customers who bought from me?

Not automatically. A purchase is not consent to marketing texts in the US, where marketing messages need documented permission. In the UK a narrow exception lets you contact your own customers about similar products if you gave them an opt-out when you collected their details and in every message. Ask a qualified adviser which applies.

What time can I send marketing texts?

US federal rules set calling hours of 8 a.m. to 9 p.m. at the recipient's location, and some states are stricter. Applying a stricter window to everyone, such as 9 a.m. to 8 p.m. local time, keeps you safe and is kinder to customers anyway. Send by the recipient's time zone, not yours.

How must a customer opt out?

Make it easy and honour it quickly. In the US, replies such as STOP, QUIT, END, REVOKE, OPT OUT, CANCEL or UNSUBSCRIBE are treated as revoking consent, and other clear wording must also be treated as a request to stop. Put the opt-out instruction in your messages and process requests the same day.

Can AI write my texts and send them itself?

AI can draft well, but a person should approve every campaign before it is sent, and consent, timing and opt-outs are handled by your messaging platform and your records, not by the AI. Never paste customer phone numbers into a general chatbot to build a send list.

Further reads

• How to Set Up AI Abandoned Cart Emails for a Small Online Store — The email version of cart recovery, with its own consent rules.
• AI Marketing for Salons: Fill Diary Gaps With Posts, Emails and Offers — A four-week plan mixing posts, emails and offers.
• Food Truck Marketing: Build an Email and SMS Regulars List With AI — Build a regulars list by email and text.
• AI for Boutique Owners: A Weekly New-Arrivals Email Routine — A weekly rhythm for new-stock announcements.
• AI Tools for Ecommerce Websites: What a 50-Orders-a-Month Store Needs — Which AI tools a small store needs first.
• ChatGPT Ads Cost for a Small Store: CPC, CPM and a Test Budget — Budget a test of another paid channel.
• AI for Car Washes: Membership Marketing and Churn Messages — Turn a member export into segments, then draft win-back texts, cancellation-reason replies and plan offers for repeat single-wash customers.
• Pizza Shop Marketing: Fill Slow Nights With AI-Written Offers — Find the slow nights in your orders, have AI write five offers, then test each one's margin before it reaches a customer.
• Tyre Sales Promotion Ideas: Seasonal Offers Written With AI — Six seasonal tyre promotions with the margin sums done first, then AI prompts and sample copy for texts, email and posts, plus the claims to cut.
• Best AI for HVAC Business Owners: Find Ageing Systems to Replace — Find customers whose heating or cooling systems are old enough for a replacement conversation, with age bands, letter and text templates, and reply tracking.
• House Cleaning Cancellation Policy: Wording and Reminder Texts With AI — A worked cancellation policy for a house cleaning company, with fees derived from your costs, lockout wording and a nine-text reminder sequence.
• AI Insurance Lead Generation: Follow-Up Rules Before You Automate — An eight-check pre-send gate for insurance lead follow-up: consent records, calling windows, opt-outs, and what AI may draft versus a licensed person.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: US federal telemarketing rules in the Code of Federal Regulations (47 CFR 64.1200: written-consent definition and disclosures, revocation by reply words, 8 a.m. to 9 p.m. hours for telephone solicitations including texts) and 47 U.S.C. 227(b)(3) (damages); a state telephone-solicitation statute with an 8 p.m. limit; the soft opt-in conditions of the Privacy and Electronic Communications Regulations as summarised in the Information Commissioner's guidance and legal commentary; the GDPR recital on consent by clear affirmative act; Twilio's SMS character-limit glossary. Pointers only, not legal advice.