Writing SMS Copy with AI: Prompts That Respect the 160-Character Limit

Written by

in

Most people who try AI for text message copy hit the same wall within a few minutes. The first draft reads nicely, but it runs to 240 characters, buries the offer, forgets the opt-out line, and suggests sending at a time when half the list is asleep. The problem is rarely the model. It is the prompt, which didn’t tell the model what an SMS actually has to do. If you write campaigns for a living, it helps to start from a library of prompt templates for marketers rather than inventing instructions from scratch each time, because a good template already contains the constraints that a generic request leaves out.

Why SMS is a harder brief than it looks

A text message feels informal, but it is one of the most tightly bounded formats in marketing. A single standard GSM-7 message holds 160 characters. Include an emoji or certain non-Latin characters and the encoding switches to UCS-2, which drops the limit to 70 characters per segment. Anything longer is split into multiple segments, each billed separately by most providers, and recipients often receive them out of order or with awkward breaks mid-sentence.

On top of length, SMS carries obligations that email does not. Marketing messages generally need a clear way to opt out, usually a reply keyword like STOP, and the sender needs to be identifiable. Many regions also restrict when promotional texts can be sent. In the United States, for example, telemarketing rules are commonly described as limiting calls and texts to a window such as 8 a.m. to 9 p.m. in the recipient’s local time, though you should confirm the specifics that apply to your program with your compliance team or legal advisor rather than relying on a blog summary.

An AI model knows none of this unless you tell it. That is the first thing a working prompt does: it states the channel and the rules explicitly.

The anatomy of a prompt that produces usable texts

Strong SMS prompts share a few components. They are short, specific, and leave the model little room to improvise on the parts that matter.

  • Channel and format: say “SMS, single segment, GSM-7 characters only, no emoji” so the output fits the encoding you actually plan to use.
  • Character ceiling: give a hard number, such as “under 150 characters including the opt-out line,” with a little headroom for merge fields like first names.
  • Required elements: list what must appear, for example the brand name, the offer, a single call to action, and the opt-out instruction.
  • Prohibited elements: ban shortened links you haven’t approved, urgency language you can’t back up, and any claims that need legal review.
  • Audience and context: describe who receives the message and why, such as “existing customers who abandoned a cart yesterday.”
  • Output count: ask for three variants, each under the limit, so you can compare tone without rewriting the brief.

A prompt built this way is easy to test. You can check every output against the same rules in seconds, which is what separates a reusable prompt from a lucky one-off.

Testing prompts before they touch your list

The phrase “prompts that actually work” only means something if you define what working looks like. For SMS, that usually means a short checklist you run against every draft before it goes into a send queue:

  • Count characters after merge fields are filled in with realistic longest-case values.
  • Confirm the opt-out text is present and readable, not truncated.
  • Check that the link, if any, resolves and is one your team has approved for SMS.
  • Verify the sender name matches what the recipient will recognize.
  • Read it aloud. Stilted phrasing that looks fine on screen often sounds wrong in a notification preview.

Keep a log of which prompt versions produced drafts that passed on the first try. Over several campaigns, that log becomes more valuable than any single clever prompt, because it shows you which constraints your model consistently respects and which ones it keeps forgetting.

Where marketplaces fit into the workflow

Many teams now look outside their own documents for prompt starting points. A marketplace of community-tested prompts can save time, but the quality varies, and a prompt written for long-form email will usually fail on SMS constraints. When you evaluate a prompt you didn’t write, run it through the same checklist above before trusting it. Look for prompts that state their channel, their length limits, and their compliance language explicitly. Those are the ones most likely to hold up when you swap in your own product details. If you are comparing options, you can browse vetted AI prompt collections organized by use case and judge them against your own SMS rules before adopting anything.

A worked example

Consider a weekend promotion for a local gym. A weak prompt says, “Write a fun text message to get members to come in this weekend.” The output is likely to be long, emoji-heavy, and missing an opt-out line.

A stronger version reads: “Write three SMS variants for existing gym members about a free Saturday class. Use GSM-7 characters only. Each variant must be under 140 characters including the final line ‘Reply STOP to opt out.’ Include the gym name, the words ‘free class,’ and one call to action to reply YES. Do not use urgency phrases like ‘last chance.’ Use a friendly, direct tone.”

The second prompt does three things the first does not. It fixes the encoding, it reserves room for the opt-out, and it gives the model a clear, checkable output. You can verify each variant in under a minute, and any that fail tell you exactly which constraint to tighten next time.

Keep humans in the loop

AI drafts are a starting point, not a send-ready message. A person should always review the final text for accuracy, brand voice, and compliance, especially when offers involve pricing, health claims, financial terms, or anything time-sensitive. Make sure the person reviewing knows the sending window, the opt-out wording you require, and who approves exceptions. A prompt can get you most of the way there, but the final responsibility for what reaches a phone sits with the team that hits send.

Takeaways

  • SMS has hard limits on length, encoding, opt-outs, and timing. Put them in the prompt.
  • Specify the channel, character ceiling, required elements, and prohibited elements every time.
  • Ask for multiple short variants so comparison is quick and objective.
  • Run every draft through the same checklist before it enters a send queue.
  • Log which prompts pass on the first try, and refine the ones that don’t.
  • Treat any prompt from outside your team as untested until it passes your own rules.

Good SMS copy has always been about respecting the reader’s time and attention. A well-built prompt simply encodes that respect into the first draft, so your team spends its effort on the ideas that matter instead of cutting words to fit a box.

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *