Facebook ads for skincare and beauty brands

Start with who reads it. Pew's 2025 survey puts Facebook ahead of Instagram in every American age group from 30 up, and at 50 to 64 it reaches 74% against Instagram's 40%; the Facebook examples page has the full split. If your customers skew older, Facebook is where more of them are, and that changes what proof has to look like. On our reasoning, readers judge a result by whether the skin in it is like theirs, and a 26-year-old's cheek tells a 58-year-old little about what a moisturiser will do to their skin in January.

The copy cannot close that gap. Meta's personal-attributes standard forbids asserting or implying the reader's age, and “Ready to upgrade your skin to look younger?” is one of its rejected examples. A person in the ad can close it. Meta's examples turn on the word that points at the reader, so a customer who says “At 58, this is what my cheeks do in January” is describing their own skin. That is our reading of Meta's examples, and it is why a real person talking about their own face can say what your copy cannot.

It is also where this trade's creators run into the second referee. The same person who can say “at 58” can write “it cleared my rosacea”, a drug claim, and in your ad it is your claim. The FTC's endorsement guides say an endorsement “may not convey any express or implied representation that would be deceptive if made directly by the advertiser”. In their own example an influencer writes “This lotion cures eczema”, and the advertiser who lacks substantiation “would be liable regardless of the liability of the endorser”. The FDA counts claims wherever a brand makes them: a July 2024 warning letter to a skincare company quoted three of its own Instagram posts, the oldest from March 2022, promising to help “acne and rosacea” and “increase collagen production”.

Meta sells the borrowed face as a format. Partnership ads, Meta's name for influencer and creator ads (it used to call them branded content ads), let an advertiser “run ads with partners: creators, brands and other businesses”. On Facebook they run in Feed, Reels, Stories and Marketplace, under all six objectives.

The creator shares a partnership ad code generated from one of their published posts; on Facebook it comes from the post's three-dot menu. In Ads Manager you turn on the partnership ad toggle at the ad level, paste the code, the post ID or its URL, and choose whether the header shows both accounts, only the first, or lets Meta pick (Marketplace and Stories always show both). Then either boost their post as it stands or create a new ad and write the text yourself.

Two planning notes. Our publisher posts under your own Page only, so a partnership ad is built in Ads Manager. And we could not confirm in Meta's help pages whether a creator's Instagram post can run in Facebook placements; check before you plan on it.

The second thing Meta sells is a way to find the customer again when the jar runs out. A website custom audience built on your Purchase event can hold the people who bought within a band of days you choose. The targeting section builds one, and the budget section measures what it adds, because the purchases Ads Manager reports cannot tell you which reorders the ad caused.

Ads Manager offers six objectives: Awareness, Traffic, Engagement, Leads, App promotion and Sales. Pick by where the purchase happens and what has to happen before it. A product the reader already understands, bought on your site: Sales, with your store sending Meta a Purchase event and its value. A product that asks the reader to change a habit first: Traffic, to a page that makes the argument. A capped batch or a launch list: Leads. News for your own customers: Awareness.

Before you build on the Purchase event, look at Events Manager. Meta treats “diseases, medical conditions and injuries” and over-the-counter products “for specific medical conditions” as sensitive health information, warns that sharing it can bring “performance issues, data restrictions”, and tells you by email or with an alert in Events Manager or Ads Manager. It also said that from 2 September 2025 it would start flagging custom conversions that suggest specific health conditions and stop them being used to run campaigns. Meta names no skincare category, but a store whose product names and page addresses say acne, eczema or rosacea is naming conditions. The cosmetic wording that keeps your ads on the right side of the FDA also keeps your page addresses from describing one.

The objective buys the first jar. None of them buys the second by itself: a reorder is a Sales purchase like any other, and the targeting section is what points a Sales ad set at the customers whose jar is running out.

The losing skincare account runs one product shot, a jar on travertine in morning light, to everyone, with a line about glowing skin. It gives the reader nothing to believe and nothing to check, and it looks like every other jar in the feed. The rows below are offers with an argument attached, and they differ along the line this page is built on: whether they win the first jar, on proof and risk, or the next one, on timing and price.

Each row pins four decisions: who the offer is for, which angle family argues it (each is a page in our library of ad angles), which objective buys it, and, where the examples page has one, a finished ad. Fourteen rows is not a campaign plan. Pick the first-jar offer your best product earns, and the second-jar offer that suits how long it lasts.

The first eleven rows are mainly for people who have not bought from you; the gift sets and the seasonal batch reach customers too. The next two sell the jar after that. The last is for when something goes wrong.

Health is not one of Meta's special ad categories, so none of the special-category limits apply: age, lookalikes, exclusions, custom audiences and interests are all available, though the health-data filters in the objectives section can still restrict the pixel data these audiences are built from. For the first jar, broad targeting is the starting point. A store that sends Meta its purchases teaches the delivery system who buys, and Advantage+ audience treats interests and included audiences as suggestions it may go beyond. What it holds are location, minimum age, language and custom-audience exclusions. Meta's API reference says that since version 23.0 a new ad set either defaults to Advantage+ audience or must set it explicitly; in Ads Manager, check the audience section of every new ad set.

Minimum age moves with the ad, not the brand. Meta's floor is 13, and its age rules exempt creams, cleansers and makeup, so those ads can run from there; a before-and-after runs at 18 and over, and so, as the safe default, does a sunscreen or a set with one in it. An ad set takes the strictest floor any of its ads needs. With Advantage+ audience on, the minimum you set, from 18 to 25, holds and a maximum does not; Meta's reference does not say how far below 18 it delivers when you set no minimum, so if younger buyers matter to you, check the audience section. On our reasoning, the face in the ad is the age signal you still control. Put age in the copy and it points at the reader.

Exclude recent customers. Anyone still using a jar they bought from you should not see a first-order ad, so exclude everyone who bought within your longest product's life plus about ten days (70 days if your products last about two months) from prospecting and retargeting ad sets, and sell them what they have not bought in a customer ad set of their own. Exclusions belong to the ad set, so excluding product by product means one prospecting ad set per product, each needing its own fifty results a week; only a brand with that volume should try it. Custom-audience exclusions are one of the constraints Advantage+ audience will not expand past. Meta used to offer an existing-customer budget cap on Advantage+ sales campaigns; its help centre now says the cap “is no longer available” and describes the same moves by hand: exclude existing customers, or give them a separate ad set with its own spending limit.

The second jar is a calendar, and your pixel keeps it. A website custom audience built on the Purchase event keeps each buyer for as many days as you choose, and one audience can hold an inclusion rule and an exclusion rule, each with its own window. Include everyone who bought in the last 70 days and exclude everyone who bought in the last 45, and for a set sized for about two months the audience holds the customers from shortly before it runs out to shortly after. A customer who reorders on your site drops out, because they now match the exclusion.

Filter those rules by product if your pixel sends a product ID with the Purchase event. For a store with more than one product that is not optional: without it, a serum buyer who adds a hand cream at day 40 is kept out of the band until day 85, weeks after the serum ran out. Sell the two-jar pack under its own product ID, so its buyers never match the single jar's band and get one of their own at twice the life (bought in the last 130 days, not the last 105, for a two-month product). Some customers the pixel cannot time: subscribers, whose renewals may never reach it, customers sent a replacement, and refunded customers. Keep them out of every reorder ad set with an uploaded list, and upload another for buyers older than your windows or from before the pixel.

Then keep each retargeting ad set inside its audience. For Sales, Leads and App promotion, including a custom audience makes it an Advantage+ custom audience suggestion automatically, and Meta says a suggestion can deliver ads “beyond the original selected audience”. A reorder ad shown to strangers is a first-order ad with the wrong copy. Use “Further limit the reach of your ads” on every ad set built on a custom audience, product-page visitors included.

For a product the reader already understands, the product page. It is advertising too: the FDA counts claims “on the product labeling, in advertising, on the Internet”, so the page should describe the product in the same cosmetic words as the ad, give the size and how long it lasts, and carry the full ingredient list as text, which a reader with an allergy checks before buying. Send the Purchase event with its value, and tag the link so your store records which campaign each customer's click came from.

For an ad that has to change a habit, a page that makes the argument: why a foaming wash leaves skin feeling tight, why a hand cream can dry in a minute. Run it as Traffic, optimised for landing page views, and write it under your own name. A page dressed as an article from an independent publication is the format the FTC treats as deceptive when it implies the ad “comes from a party other than the sponsoring advertiser”; the retinal serum's invented journal is the example.

For a reorder, a link that opens the product ready to buy again, or the customer's last order ready to repeat, if your store can build one, with the two-jar price and the subscription beside it.

For a subscription, two rulebooks apply. Meta's Subscription Services standard says subscription ads “must disclose information on pricing and recurrent billing”, and an ad that asks for personal information, an Instant Form for instance, is prohibited if it lacks any one of these: an unchecked opt-in box, clear language on how to cancel, the price and billing interval shown plainly rather than in fine print or behind a link.

Your checkout answers to the Restore Online Shoppers' Confidence Act: the material terms before the billing details, express informed consent before the charge, and a simple way to stop the charges. A box the customer ticks themselves is the plainest way to show consent. Put the interval in the ad as well; “$38 every 12 weeks, skip or cancel in your account” fits in one line.

For a limited batch or a launch, an Instant Form or a sign-up page, then an email on the day. Instant Form leads can be downloaded for up to 90 days from submission, so connect the form to your email tool or download the leads weekly.

For shade matching, a conversation, if someone answers it. Meta lists Messenger among the places a Sales ad can send people; our publisher sends link buttons only, so that ad is built in Ads Manager. “Which shade am I?” is a question a person answers better than a chart. Readers may also ask whether a product is safe for their rosacea or their eczema, so write the answers in advance in cosmetic terms and send anything medical to a doctor: a reply in a chat is written by you.

A partnership ad goes wherever your own ad would. Boosting the creator's post keeps their words and their disclosure, so both have to be right in the post itself. Creating a new ad lets you complete the text and media yourself and leave out any claim you could not make, but their original post still says whatever it said, so ask them to fix that too.

The rule from the examples page carries every ad here: make the product the subject, and describe how skin looks or feels. Meta's personal-attributes rule has nothing to object to in a sentence that says nothing about the reader, though its Health and Wellness standard still bars timed promises without qualifiers. The FDA reads a product that makes lines “less noticeable, simply by moisturizing” as a cosmetic, and one that claims to “increase the skin's production of collagen” as a drug. A skincare account has three authors, and the rule binds all of them: you, the tool that drafts your ads, and the creators whose faces and words you borrow.

In our batch the tool drifted one way. Eight of the ten briefs described a cosmetic in cosmetic terms. Four of those ads came back with repair, rebuilding, renewal or cold protection in the copy, and a fifth named its product “Barrier Repair Emulsion”. The qualifiers went too: “often” became nothing, and “about one in twelve” became 8.3%. Read generated copy for exactly those words and put the hedges back. Read the pictures as well: the hand-cream tube came back with an ingredient list nobody wrote, and both faces whose skin tone no brief described came back light.

Precision is a claim of its own. A figure, “clinically proven” or “clinical” needs evidence behind it: in 2014 the FTC's order against L'Oréal, over “clinically proven” gene claims for two anti-aging lines, required “competent and reliable scientific evidence” for claims like them. The hand cream's copy called itself “Clinical Barrier Repair for Nurses” on a brief that mentioned no test at all.

A borrowed face has rules of its own. When an ad represents that the person in it uses the product, the FTC's endorsement guides require them to have really used it, and the ad to stop when you no longer have good reason to believe they still do. The guides say a customer's result will likely be read as what buyers generally get; the FTC found that a note like “Results not typical” does not change that, so either the result is typical or the ad states what typical is. One of the FTC's examples is an acne treatment whose real testimonials were shown beside stock photos of people with near-perfect skin, which it says misrepresents the endorsers' improvement. A generated face is the same substitution with nobody behind it.

Borrowed words carry the strictest rule: under the FTC's endorsement guides a creator can make the claims about the product that you could make, and no more. Brief them the way you would brief a copywriter: what the product does in cosmetic words, what they may say about their own skin and how long they used it, and what is off limits (cures, heals, treats, repairs, collagen, cells). They can describe their skin as dry, oily or sensitive; they should not name a diagnosis, or say what the product did to one.

Read the caption before they publish, not before you run it. The code you boost is generated from the live post, and a paid or gifted creator's post is an endorsement from the day it goes up. The FTC also expects a brand to make “a reasonable effort to know what participants in your network are saying”, so keep reading their posts afterwards.

Agree the terms before they share the code. How long you may run the post: turning a code off stops new ads, but “Any active ads using the code will continue running”, so the end date is yours to keep, and the ad stops sooner if they stop using the product. Who may use the code: two partners can at a time, so name yourself and your agency. How long they will use the product before they post, at least as long as their words will imply. And the original file, so a new ad can be cropped to each placement.

Then disclose it in your own words. Meta requires creators to tag you with its branded content tool, and a partnership ad can show the partner's account in its header, but the FTC's advice is not to rely on a platform's tool: “it's always best to add your own disclosure even if a platform offers its own disclosure tool.” A free product is a material connection too, so a creator you only sent a jar to still has one to disclose. If you boost their post as it stands, the disclosure has to be in their caption; if you create a new ad, start its text with “Ad” or “Sponsored”.

The two purchases set the structure: a line that finds new customers, a line for people who looked and did not buy, a line that sells the next jar, and a small line for customers. They run side by side, because at any moment you have new readers, people deciding, and customers whose jar is running out. Each is judged on its own clock, shown on its card and worked through in the budget section.

Start with what a new customer is worth, because it sets what you can pay for one. Roughly:

Value of a new customer = margin on the first order + (share of first-time buyers who reorder × margin on a reorder × reorders per returning customer)

Take each figure from your own store, from customers old enough to have had the chance to reorder, over a period you are willing to wait for, a year say. Use margins after product cost, postage, payment fees, discounts and refunds; take the reorder ads' own spend off the reorder margin; and work out the reorder share without gift orders, whose buyers are not the users.

That value is the most you can pay for a first order and break even over the year, and your target cost per first order sits below it. The cost per first order that Ads Manager or your link tags show you is attribution, not effect: some of those customers would have found you anyway. One way to approach cause for the first-jar line is to alternate spend off and on over several equal periods, change nothing else, and compare the new customers your store records in each.

Then the volume. Meta says an ad set usually leaves its learning phase after about 50 results in a week, so plan on roughly 50 × your cost per purchase, as Meta counts it, as the weekly spend for one Sales ad set; before you have run anything, use your target. If your budget cannot feed even one, the ad set stays in learning, which is survivable if you stop editing it. Learning is counted from the ad set's last significant edit, so add ads and change budgets on a schedule, and judge it on several weeks of store orders.

Read the first-jar line on two clocks. Ads Manager gives you a cost per purchase within days. Your store gives you the cost per first order and, a product life later, the reorders: tag every ad link with UTM parameters so your store's analytics can record which campaign each first-time customer's click came from, and follow those customers past the date their jar would run out. A campaign whose customers never come back may be worth pausing, whatever its first week said. Gift campaigns are judged on the first order, plus any recipient reorders a card in the box captures.

Where Meta offers audience segments, which split a Sales campaign's results into new audiences, engaged audiences and existing customers once you define them in your ad account settings, define existing customers with the same window as your exclusions. A prospecting campaign should then show an existing-customer share close to zero; anything more is your exclusions leaking. The gift-set ad set, which leaves customers in on purpose, is the exception.

Net out refunds. A 60-day refund arrives long after the one- or seven-day window in which Ads Manager credited the purchase, so the refund ad's cost per purchase there is its best-looking number, not its real one. Count refunds in your store's figures.

A creator costs a fee per post plus the media behind it. You do not choose how much each ad in an ad set is shown, so a creator's ad beside your own is a comparison of what Meta chose to show, not a fair test; an ad it barely showed has not lost, it has not been tried. Rank creators on the first orders credited to their ads and on how often those customers return: that is a lead, not a verdict. To learn what a creator adds, give them their own ad set and alternate it on and off.

The second jar is measured differently, because the credit is the problem. Some customers in a reorder band were going to reorder anyway, and any who click the ad and reorder within the attribution window are counted, whether or not the ad changed anything. So build the comparison yourself, in the four holdout steps in the list below: split your customers by a fixed rule, hold one half out of the reorder ads, compare margin per customer over two jars' life, and check the gap against chance.

Count margin, not revenue, and count long enough. An ad that changes nobody's mind but hands the two-jar price to people who would have paid full price looks like a win in Ads Manager and, over a short window, in revenue too; over two jars' life, in margin, it shows up as the loss it is. If some held-out customers still see the ad, that shrinks the difference rather than faking one. At a small store, expect to read several months together.

Meta offers a model for the same question. Its incremental attribution setting “optimizes ad delivery for incremental conversions”, using “machine learning models that predict whether a conversion is caused by an ad”. A prediction is not a comparison, so keep your holdout even if you use it.

The first three are about proof: a face that does not exist, a creator's words run unread, and a result nobody has had time to see. The next three are about whom the account sells to. The three after that are words that go out unread, in a reply, a checkout and a label. The last decides what the account turns into.