AI prompts for Facebook and Instagram ads

Primary text length: a recommendation, not a hard limit

Unlike Google Ads' strictly enforced 30-character headline cap, Meta does not hard-cap primary text length at upload — longer text is accepted, but Meta's own guidance recommends keeping the most important message within roughly the first 125 characters, since text beyond that point gets truncated behind a "See more" tap on most placements, and a large share of viewers never tap through.

This changes what a prompt should ask for: not a strict character maximum, but a front-loaded structure where the strongest hook and the core offer both land before the likely truncation point, with supporting detail allowed to run longer for the smaller audience that does expand the text.

A prompt for primary text that front-loads the hook

The forbidden-phrase list is not a style preference — "struggling with debt" and "tired of feeling exhausted" are exactly the kind of phrasing Meta's Personal Attributes policy exists to catch, because they imply the ad knows something specific and sensitive about the individual viewer.

Meta's Personal Attributes enforcement covers a wide list beyond health and finance, including implied assumptions about a viewer's relationship status, religion, race, national origin, disability, age, and sexual orientation — Meta's Advertising Standards page groups all of these under one policy because the underlying concern (an ad appearing to profile an individual by a sensitive trait) is identical across each category.

The Personal Attributes policy is where AI-generated copy fails most often

Meta's Personal Attributes policy prohibits ads that assert or imply knowledge of a person's race, religion, health condition (including mental health), financial status, sexual orientation, or other sensitive characteristics, regardless of whether the implication is accurate. A model asked to write empathetic, persuasive ad copy reaches for exactly this pattern by default — "feeling overwhelmed by anxiety?" or "behind on your bills?" — because direct-address emotional hooks are common and effective in general advertising writing and the model has no built-in awareness that Meta specifically restricts this category.

Special Ad Categories change what a prompt is even allowed to produce

Housing, employment, credit, and social issue/election/politics ads fall under Meta's Special Ad Category framework, which restricts detailed audience targeting (no age, gender, or zip-code-level targeting for housing ads, for example) and imposes additional review on the claims themselves, independent of whether those claims are factually accurate. An accurate claim about a job's salary range can still require careful handling under employment-category rules.

Declaring the correct Special Ad Category in Ads Manager is a platform-configuration step separate from the copy itself, but a prompt that ignores the category entirely will often produce copy that needs to be rewritten once the declaration is made and the associated targeting and claim restrictions apply.

Health and wellness claims: the strictest review category on the platform

Health-adjacent ads — supplements, fitness programs, mental health services, weight management — face some of Meta's most consistent rejection rates, particularly for specific outcome claims ("lose 20 pounds in 30 days") and before-and-after imagery, both of which Meta's policy restricts even when the claim is true for some customers, because the policy concern is about setting unrealistic expectations at scale rather than about individual honesty.

Where no real substantiation exists, the honest instruction is to write about the product's function rather than its outcome — "formulated with [real ingredient]" rather than "lose weight fast" — because an unsubstantiated outcome claim in this category is one of the more reliably rejected patterns in Meta's ad review.

Creative brief prompts: connect the copy to the actual image or video

Meta's ad review evaluates image and video creative alongside copy, and a prompt limited to text generation misses half the review surface. A creative brief prompt should describe the actual asset rather than assuming the model can see it.

This prompt cannot replace an actual visual review, but it catches an underappreciated failure mode: copy and creative that are each individually fine but combine to make an implied claim neither makes alone.

A concrete example of the combination failure: copy reading "get back on track today" paired with an image of a person looking visibly distressed reads, together, as implying knowledge of the viewer's emotional or mental health state even though "get back on track" alone is generic enough to pass a text-only Personal Attributes review, and a distressed-looking stock photo alone is not itself policy-violating either.

Meta processes billions of ad impressions daily across Facebook, Instagram, Messenger and the Audience Network, and its automated review systems flag a substantial share of policy issues before any human reviewer sees the ad — Meta's 2024 Community Standards Enforcement Report describes automated detection as the first line of review across its advertising and content systems generally, which is part of why combination failures between text and image can slip through initial automated screening even when either element alone would be caught.

Testimonials and results claims need real substantiation, not a plausible quote

The FTC's Endorsement Guides, most recently updated in 2023, require that testimonials and endorsement-based claims in advertising be truthful and substantiated — a standard that applies to Meta ads exactly as it applies to any other channel, and one a fabricated customer quote or invented results statistic violates regardless of how natural it reads.

Where no real testimonial or verified result exists yet, the safer prompt asks for benefit-focused copy grounded in the product's actual features rather than results copy dressed up with an invented customer voice.

Meta's own political and social issue ads library, launched in 2018 and required by the platform for elections and issue-based advertising in over 40 countries as of 2025, is a useful reference for the level of specificity and disclosure regulators expect from this ad category — reviewing a handful of archived ads in the library before writing a social-issue campaign gives a concrete sense of the disclosure standard Meta actually enforces, which is more reliable than inferring it from the written policy text alone.

Ad review is inconsistent — build a review pass into the workflow, not just a policy check

Meta's own Business Help Center documentation acknowledges that ad review combines automated systems and human reviewers, and that review outcomes can vary between similar ads — the same copy can clear review once and be rejected on a near-identical resubmission. This means a prompt-level policy check reduces risk but does not eliminate the possibility of a rejection that needs appealing rather than rewriting.

A practical workflow: run new ad copy through a policy-focused review prompt before submission to catch the clear, avoidable violations (personal attributes, unsubstantiated claims, missing Special Ad Category handling), then treat any rejection that survives that check as a candidate for Meta's appeal process rather than an automatic signal the copy itself was wrong — appeals overturn a meaningful share of initial rejections precisely because the review process itself is imperfect.

Creative variation beats copy micro-optimization once the baseline is compliant

Once an ad clears policy review and reaches a reasonably well-targeted audience, the highest-leverage remaining variable is usually the creative concept — a genuinely different image, video hook, or angle — rather than small copy tweaks to an already-working ad. Meta's Advantage+ and dynamic creative tools are built around testing multiple creative concepts simultaneously for exactly this reason.

Asking explicitly for different angles, not just different phrasing of the same angle, is what produces creative diversity a testing program can actually learn from — four rewordings of the same hook tell you nothing that one test didn't already tell you.

Meta's Advantage+ shopping campaigns, expanded significantly across 2023 and 2024, are built specifically to test combinations of 5 or more creative assets against each other automatically using machine learning to allocate budget toward the best-performing pairings — a structural argument for supplying genuinely distinct creative concepts rather than a handful of similar variants, since the system's own optimization has more real signal to learn from with 5 distinct angles than with 5 versions of one angle.

Headlines and link descriptions carry their own limits

Beyond primary text, Meta ads include a headline field (Meta's own guidance suggests keeping it under roughly 40 characters for reliable display across placements) and an optional link description (around 30 characters before truncation on most surfaces). Both fields sit below the main image or video and are easy to treat as an afterthought, but a strong headline can meaningfully improve click-through when it states a concrete benefit rather than repeating the primary text.

Testing headline variants alongside primary text variants, rather than fixing one field while only testing the other, surfaces which field is actually driving performance differences — a distinction that matters when deciding where to invest further iteration time.

Meta's placement system serves the same ad across Feed, Stories, Reels, and Marketplace by default when automatic placements are enabled, and each surface renders headline and link description slightly differently — Stories and Reels place less visual weight on this text than Feed does. A headline that reads well in a Feed preview is worth checking in Meta's own placement preview tool before assuming it performs identically everywhere the ad runs.

Age-restricted and regulated products need the claim checked before the copy is written

Alcohol, gambling, financial services, and pharmaceutical-adjacent products carry additional restrictions on Meta beyond the general Advertising Policies — age-gating requirements, region-specific eligibility rules, and in some cases a requirement to hold specific advertiser certification before Meta will even run the ad category at all. A prompt for this kind of product needs the regulatory constraint stated up front, not discovered after a rejection.

Treating the model's policy knowledge as a starting hypothesis rather than a final answer matters more in this category than almost any other, because the cost of an incorrect assumption is a compliance failure with real regulatory exposure, not just an ad rejection.

Retargeting and lookalike audience copy needs a different starting assumption

Copy aimed at a cold, top-of-funnel audience and copy aimed at a warm retargeting audience — people who already visited a site or engaged with a page — should not read identically, because the retargeting audience already has context the cold audience lacks. AI-generated copy defaults to a cold-audience assumption unless told otherwise, explaining a product from scratch to someone who already added it to a cart last week.

This distinction is a genuine performance lever, not a style preference — retargeting ads that re-explain the product from scratch waste the scarce attention of an audience that has already signaled interest, on information that audience does not need.

The same logic extends to frequency. Meta reports frequency — average number of times a person has seen the ad — directly in Ads Manager, and a retargeting ad shown to the same person 15 times in a week benefits from a message that assumes prior exposure, whereas a first-impression cold ad does not need to account for repetition at all. A prompt for a retargeting campaign running longer than a week or two is worth pairing with an instruction to write 2-3 message variants specifically to reduce the fatigue that comes from showing identical copy to the same audience repeatedly.

What to check before launching any AI-drafted Meta ad

Five checks apply before any AI-assisted Meta ad goes live. First, scan for Personal Attributes language — any phrase implying knowledge of the viewer's health, finances, or other sensitive trait, even phrased as a rhetorical question, since this is the single most common avoidable rejection cause. Second, confirm the correct Special Ad Category is declared if the ad falls into housing, employment, credit, or social issues, and that targeting settings match that category's restrictions.

Third, verify any statistic, testimonial, or results claim traces to real, documented, permission-cleared source material rather than a plausible-sounding invention. Fourth, review the creative and copy together rather than separately, since a compliant-looking sentence next to a specific image can combine into an implied claim neither makes on its own. Fifth, for any regulated product category — alcohol, gambling, financial services — confirm the current Meta policy requirements directly in Ads Manager rather than relying solely on a model's general knowledge, since these rules are reviewed and updated more frequently than the platform's general advertising standards.

A sixth, lower-stakes but still worthwhile check: preview the ad across at least Feed and Stories placements before launch, since the two surfaces render primary text and headline differently, and copy that reads cleanly in one preview can crowd an image awkwardly in the other. Meta's own ad preview tool inside Ads Manager shows this directly and takes under a minute to check.