Facebook Ad Copy: How to Write It When You Are Not a Copywriter

Almost every part of an ad can be bought. A photographer will shoot the product, a designer will lay out the frame, a generator will produce forty backgrounds before lunch. The words are the part that stays on your desk. They are also the part ad review reads most closely, and the part that decides whether anyone reads the second line. This is about the text and nothing else: what the three fields are for, how long each one can be before it is cut, which sentences get an ad rejected, and what to change first when nothing is working.

The three text fields, and what each one is for

Ads Manager shows three boxes and does not mention that they are not interchangeable. Underneath, in the Marketing API, they are three separate fields on the ad creative, and their real names are worth knowing because error messages use those rather than the labels on your screen. Primary text is the field Meta calls message, described in the Ad Creative Link Data reference as the main body of the post. The headline is name, the name of the link, which overwrites the title of the link when you preview the ad. The description is description, the link description, and Meta's own note on that field is the sentence almost nobody has read: you do not use this setting for Instagram ads. All three were read on September 12, 2026.

One more thing the interface hides. A video ad does not use the same fields as an image ad. On an image the headline is name and the second line is description. On a video the headline is title and the second line is link_description, and a video carries no link field at all, because the destination lives inside the call to action button. Graph silently ignores a key it does not recognize, so a headline written into the wrong field produces no error and no warning. The ad is created, it goes live with no headline, and the only way to find out is to look at it in Ads Manager while the budget is already spending.

Two of those four are Facebook-only by Meta's own documentation. If your ads run mostly on Instagram, the description and the display link are not where your effort belongs, and an evening spent polishing them buys nothing at all.

How long is too long?

This is where most advice on the subject is quietly repeating a number from 2017. The figures everyone knows, 125 characters of primary text and 40 for the headline, were not invented, but they are not limits and they are not universal. Checked on September 12, 2026, the Marketing API reference for Ad Creative Link Data documents message, name, description and caption without a single character count anywhere on the page. What Meta does publish is the Ads Guide, and that gives recommendations which change with both the placement and the format.

Here are those recommendations across placements, taken from the Ads Guide on September 12, 2026. Read the table once and the advice to write to 125 characters stops working as a general rule.

Two caveats before anyone plans a workflow around that table. These are recommendations, not caps: Meta accepts longer text and simply truncates it where it does not fit. And the Ads Guide shows one campaign objective at a time, so the same placement can carry a different recommendation under a different objective. Treat the numbers as the shape of the problem rather than as a specification. The shape is clear enough on its own. A sentence that sits comfortably in the Facebook Feed is more than three times the recommended length for Instagram Reels, and a carousel headline has half the room a Marketplace headline has. Nobody writes one line that is right for both. What you can do is write the important part early enough that it survives the shortest one.

What the first line has to do

Primary text is not cut at a fixed number of characters. It is cut where the layout runs out of room, which depends on the placement, the device, how long your words are and whether the reader has set a larger font. That is why a fixed target is the wrong mental model and a fixed habit is the right one: write so the ad still works when everything after the first line is hidden behind See more. Nobody expands an ad to find out whether it is relevant to them. They expand it because the visible part already said it was.

In practice that means the first line carries the offer and the audience, and the story goes underneath it. A line reading We have been family owned since 2004 is not an opening, it is a footnote that has been promoted. Set it against Same-day drain cleaning in Tucson, flat $149. The second has told a stranger four things, the service, the city, the speed and the price, any one of which is enough to make them stop, and it has done it in the part of the ad that no placement hides. The history still belongs in the ad. It belongs in line four, where it reassures someone who has already decided to keep reading.

Line breaks deserve a sentence of their own, because they are the one formatting decision available to you. Six lines with no break reads as an essay, and an essay in a feed is a scroll. But breaks also spend the visible space before the cut, so an ad that opens with three short punchy lines can burn its entire visible budget before saying anything concrete. One break after the first line is usually the whole technique.

Price, deadlines and guarantees without failing review

Owners strip these out of their ads constantly, because someone told them the words are risky. They are not. Meta does not prohibit stating a price, a deadline or a guarantee. What it enforces against is a price, a deadline or a guarantee that is not real. The Community Standard on Prohibited Commercial Practices, read on September 12, 2026, lists the practices it acts on, and the list reads like a description of what goes wrong rather than a banned vocabulary.

• –Misleading or unreasonable pricing. The number in the ad has to be a number the visitor can actually pay. A headline price that becomes something else two clicks later is exactly the practice being named.
• –Deceptive subscription practices, which Meta defines as enrolling customers in recurring payment plans or subscription services without clear disclosure of terms. If your offer turns into a subscription, the ad and the page behind it both have to say so before the card is entered.
• –Violation of return or refund policies, defined as failure to honor stated refund or return policies. A guarantee is safe to write and unsafe to break, and Meta is explicit that the enforcement is on the breaking.
• –Product misrepresentation: promoting products or services that materially differ from what was advertised, including significant discrepancies in quality or incorrect items. This is the rule that ties the ad text to the landing page, and it is why the two have to promise the same thing.

Three categories cannot carry a guarantee at all, whatever the wording. The same standard prohibits offers of investment opportunities where returns are guaranteed or risk-free, offers of guaranteed jobs, and loan offers that guarantee approval without any mention of standard approval criteria such as financial information from the borrower. If you sell in one of those, the word is simply not available to you. Everywhere else it is, on the condition that you honor it.

Meta also enforces on sustained unhappiness rather than on the text alone. The same standard says it weighs consistent patterns or disproportionate volumes of negative feedback and shares that feedback with advertisers. Copy that oversells passes review and then collects the complaints that get an account looked at, which is the slow version of the same failure.

What Meta actually forbids in the text

The policy that rejects the most small business copy is not about profanity and not about claims. It is Privacy Violations and Personal Attributes, and hardly anyone reads it before writing. Its rule is one sentence: ads must not contain content that asserts or implies personal attributes, including direct or indirect assertions or implications about a person's race, ethnicity, religion, beliefs, age, sexual orientation or practices, gender identity, disability, physical or mental health including medical conditions, vulnerable financial status, voting status, membership in a trade union, criminal record, or name.

The folk version of that rule, that you may not address the reader as you, is wrong, and the policy says so in as many words. Under what ads can do, Meta lists: use you/your language without a personal attribute. What is forbidden is using you to assert something about the person reading. The distinction is far easier to see in Meta's own examples than in any paraphrase of them, and these were read on the policy page on September 12, 2026.

• –Allowed: Depression counseling. Rejected: Depression getting you down? Get help now.
• –Allowed: New diabetes treatment available. Rejected: Do you have diabetes?
• –Allowed: Meet seniors. Rejected: Meet other seniors. The word other is what asserts the reader is one of them.
• –Allowed: Our new lotion and creams fight wrinkles like never before. Rejected: Ready to upgrade your skin to look younger?
• –Allowed: Services to clean up any previous offenses. Rejected: Are you a convicted felon?
• –Allowed: We print customizable t-shirts and stickers with your name. Rejected: Billy Taylor, get this t-shirt with your name in print.

Read down that list and the pattern is mechanical. Every rejected line is a question or an assumption aimed at the reader. Every allowed line describes the product or the service. Two habits follow, and between them they cover most rejections a small business will ever see: do not open with a question about the reader, and delete the word other wherever it sits in front of a group noun. The policy is also more generous than its reputation in one direction. Meta allows a passing reference to a personal attribute, including gender and age ranges, and broad references such as calling someone American or a New Yorker, so writing to a city or to an audience is not the hazard people assume.

Before and after, and promises of a result

This is the other thing everyone repeats and almost nobody has checked. Before-and-after imagery is not banned across the board. Meta's Health and Wellness policy, whose change log shows an update on July 22, 2026 and which was read on September 12, 2026, sets out what advertisers can run when targeting people 18 or older, and general cosmetic products, procedures, surgeries depicting before and after transformation appears on that list of permitted things.

What the same policy does prohibit is narrower and far more useful to know. Ads promoting weight loss, weight gain or cosmetic products cannot contain statements of inferiority about physical appearance, which Meta defines as terms, descriptions or questions that are negative and attack an individual's appearance, specific body parts, or hygiene. They cannot show a close up on specific body area by pinching fat. They cannot claim results can be achieved solely by using wearable products. And claims to cure, heal or eliminate any of a named list of incurable conditions are prohibited, with Meta noting that this does not apply to claims about treating or managing symptoms.

Then comes the clause worth copying onto a sticky note. In a health, weight loss or weight gain context, the policy prohibits clickbait tactics such as sensational language with exaggerated or extreme claims, or promises of specific outcomes within a set timeframe without disclaimers or qualifiers. That is the rule people are reaching for when they say you cannot promise a result, and it is narrower than the folklore. The objection is not to the promise. It is to a specific outcome, inside a named timeframe, with nothing qualifying it. Outside those categories the binding constraint is the one from the previous section: whatever the ad promises has to be what the buyer receives, or the complaints will do the work the policy did not.

How to get variants without writing twenty of them

The ad object itself accepts more than one of each. In the Marketing API, an ad's asset feed spec takes bodies, titles and descriptions as lists rather than as single values, and carries an optimization type whose values include PLACEMENT and ASSET_CUSTOMIZATION. That was read on September 12, 2026. Two things follow from it that are easy to miss. Variation is not a favor you are doing the delivery system, it is a field the object was designed around. And the truncation problem from earlier has a structural answer as well as a stylistic one: a short body and a long body can travel in the same ad, and the placement decides which one a person sees.

A generator, ours included, is good at the mechanical half of this. Give it one line you wrote and it returns fifteen shapes of that line: shorter, longer, question, statement, price first, problem first. It is not good at the half that matters, which is knowing what a customer complained about on the phone last week. The division of labor that works is to write one honest sentence yourself and let a tool multiply it, rather than to ask a tool for the idea and then edit what comes back. Separating whether it was the words or the picture that moved the number is its own diagnosis, and that one is written up at aevin.dev/blog/ai-generated-ad-creatives rather than repeated here.

What to test first

Test order matters more than test volume for anyone on a small budget, because most owners run out of patience long before they run out of ideas. This order is by how much a change usually moves and how cheap it is to make.

1. 1.The first line. It is the only part of the primary text every placement shows, and rewriting it costs nothing. Change what it promises, not how it is phrased.
2. 2.The offer inside the words. Free consultation and $49 diagnostic are not two phrasings of one thing, they are two offers, and the winner tells you something about the business rather than about the writing.
3. 3.The headline. It sits next to the button, it takes a minute to rewrite, and the recommendations suggest it is the most cramped field you have.
4. 4.The call to action button. A text decision with a fixed vocabulary, which makes it the fastest thing on this list to change and the easiest to read afterwards.
5. 5.The description. Last, and only for Facebook placements, because Meta's own documentation says the field is not used for Instagram ads.

Then the arithmetic, which decides whether any of the above meant anything. Two variants that differ by a fifth of a percent across forty clicks have told you nothing, and delivery will happily hand one variant most of the impressions if the test is not built to prevent it. How many results a comparison needs, and how to structure one so it stays fair, is a separate piece of work at aevin.dev/blog/how-to-test-facebook-ad-creatives. One check belongs before all of it, though. A campaign that used to work and then stopped may have no copy problem at all, and telling a worn-out creative apart from an auction that simply got more expensive is its own diagnosis, at aevin.dev/blog/creative-burnout-in-the-numbers. Rewriting the text of an ad whose price rose for reasons outside the ad is the most reliable way to spend a week on nothing.

One last pairing, because it is where good copy most often dies. The sentence that earns the click sets an expectation the landing page has to meet within a second of loading, and Meta states the requirement outright in its Advertising Standards: the products and services promoted in an ad must match those promoted on the landing page. If the ad says flat $149 and the page opens on a brand statement, the visitor has been handed a different offer from the one they accepted, and the bounce is not a design problem. What the page owes the promise the ad made is at aevin.dev/blog/landing-page-for-facebook-ads.

Where Aevin fits, and where it does not

Aevin is our own product. It connects Meta, Google Ads and TikTok in one panel, builds landing pages, generates images for creative, sends conversions back to Meta, keeps a CRM and teaches the MAD program, at $97, $197 and $497 a month. On this particular subject the honest description is narrow, so here it is. It will help you produce ad text and launch it, and the launch flow checks length before anything is uploaded rather than after a rejection. What it will not do is rewrite the copy of an ad that is already running. The changes the agent may propose on a connected account are pausing, resuming and setting a daily budget, plus a bid strategy change on Google, and every one of them waits for a person to approve it before it reaches the platform. Ad text is missing from that list deliberately. A sentence that has passed review and is collecting results is not something worth changing behind your back.