Compliance · 5 min read
Beauty Brand DMs: Describe the Product, Never the Result
A beauty brand's inbox is not a salon's inbox. A salon is asked about slots. A brand is asked about outcomes.
"Will this work on oily skin?" "Is it safe in pregnancy?" "Ye pigmentation ke liye chalega?" "Does it have parabens?" "I have rosacea, can I use this?" "Will it clear my acne in a month?"
Every one of those reads as a product question. Most of them are health questions wearing a product question's clothes, and the difference decides what your automation is allowed to say.
The only line that matters: description versus claim
A template can describe. It cannot conclude.
Describing is repeating what is already printed on your own pack and your own product page. The ingredient list. The stated use. The texture, the size, the shelf life, whether it is fragranced, whether it was tested on skin under supervision, what your own label says the product is for. These are fixed facts about an object. They do not change with who is asking, which is the whole test for whether something can be a template.
Concluding is the other thing. It is taking the sender's skin, or their condition, or their pregnancy, or their prescription, and mapping your product onto it. "Yes, that will help your pigmentation." "It is fine to use with your retinol." "Perfectly safe during pregnancy."
Those sentences are not marketing. They are advice about a person's body, generated by a system that knows one line of text about them and has no way to represent uncertainty. It will produce them fluently, confidently, and at whatever hour they were asked.
We are not going to tell you what your advertising rules permit you to claim about a cosmetic. Those rules vary by jurisdiction, they get revised, and a reply tool is a bad place to read them. Take the structure from this post and take the permitted wording from whoever advises you.
A cosmetic claim made by a template is still a claim. Nobody reading it knows or cares that a machine wrote it.
Your DM history is the training data, which is the risk
Here is the part specific to beauty brands and it is the one nobody expects.
Voice DNA is built from replies the account owner actually wrote. That is the right way to build it — it is why generated replies sound like your brand rather than like a call centre.
Now go and read what your team has been typing into DMs for the last year. In a beauty brand it is very often warmer and much braver than anything on the label: definitely will help your acne, within two weeks you will see a difference, totally safe, don't worry. Said once, to one person, by a founder who was being kind at eleven at night.
Feed that in and you have taught a system to say it to everyone.
So the beauty-specific onboarding step is not a settings change. It is an afternoon spent reading your own outbox before you train anything on it, and pulling out the messages where somebody promised a result. If those replies are the voice, then the voice is the compliance problem.
The skin photo is a consultation, not an enquiry
People send beauty brands pictures of their faces. Close-ups, bad lighting, a caption saying "what should I use for this?"
Unlike a photo with no words, this one may well match a keyword and fire an automation. That is worse, not better. A reply that recommends a product against a photograph of somebody's skin is a consultation performed by software that cannot see, cannot examine, and cannot ask a follow-up.
Route these. The keyword rule above catches many of them. For the rest, the honest reply is that a person will look, and then a person actually does.
Use the review queue harder than a low-ticket brand would
The economics are not symmetrical here. A slow answer about shade matching is a mild annoyance. A fast wrong answer about whether an acid is safe with someone's prescription is a different category of event entirely, and it is screenshot-shaped.
Two other controls do real work. Takeover — the moment anyone on your team replies by hand, automation is out of that thread, so a founder's careful answer is never followed by a template forty seconds later. And a rate budget, because one product reel travelling is how an uncapped automation sends two thousand replies in an hour and starts looking like spam to Meta.
What is genuinely safe, and it is most of the volume
Shipping times and where you ship. Returns policy, split by opened and unopened. Whether a shade is a limited run. Full ingredient lists exactly as printed. Sizes, prices already published on your own site, whether you offer samples, how to read the batch code, whether the pack is refillable.
Templated replies are free and unlimited; a credit is spent only when the AI writes a new reply. For a brand whose inbox is mostly logistics and shade questions, that means the repetitive majority is free forever. Free tier is 100 credits with no card, packs from ₹499.
And the boundaries: no catalogue or inventory access, so it never confirms stock. No payment links. No messaging anyone who has not messaged you first. No WhatsApp. Leads export as a CSV. Comments stay answerable for seven days, DM threads for twenty-four hours, and that DM clock restarts only when the customer writes again.
Build the condition refusal, then audit your outbox, then write the shipping templates. What actually blocks a reply is worth reading first, and if your brand also talks about wellbeing rather than only skin, the wellness brands post sits directly on top of this one.
