How to generate ads
Generating ads is the process of turning existing customer and campaign signal into testable creative, rather than inventing an angle in a meeting. The six steps below are the whole loop: gather the signal, count the patterns, turn them into hooks, brief each one, test in tagged batches, and feed the results back. Done by hand it takes days per batch. The bottleneck is never ideas, it is that nobody has read the two thousand reviews where the answer already sits.
Marián Cabadaj
11 min read
Our articles are still written by HUMANS! Get human written articles in your Google feed.
Add as preferred sourceThe familiar version: an ad is due this week, the process starts from a blank page, one person skims reviews while another scrolls last month's campaign data, and the team settles on a hook that feels right. It lands on growth managers, performance marketers and founders hardest, because they own a testing calendar without owning a research team. The output is slow iteration, thin angles, and fatigue arriving faster than replacements.
The cost of guessing has gone up. Platform automation took targeting off the table as a source of edge, which leaves creative carrying the result: Nielsen credits creative, not targeting or media spend, with up to 89% of a digital ad's in-market success, Meta's own research agrees, and Facebook cost per lead climbed almost 21% year over year in 2025, per WordStream's benchmarks. Meanwhile the opening beat has to do more: TikTok tells advertisers to land the content proposition inside the first three seconds and reports around 90% of an ad's recall impact sitting inside the first six.
None of that is an idea problem. It is a missing pipeline. Your reviews, comments and past results already contain the objections that stall a purchase and the phrases buyers use to describe relief. The work is turning that into ad-ready material repeatably, which is what the rest of this page describes.
Product reviews, support tickets, survey answers, comments under past ads, and performance exports from Meta Ads or Klaviyo. All of it already exists and almost none of it is being read. Roughly fifty reviews per product line is where patterns start to hold; under that you are reading anecdotes.
A single articulate review is a distraction. What you want is the objection raised forty times and the desire phrased the same way by thirty different people. Count first, read second, because frequency is the only thing that separates a real pattern from a memorable outlier.
Lead with the tension the customer already feels, in the words they used. If reviewers wrote "I stopped worrying about it halfway through the first week", that sentence is the hook. A generic product claim in its place is the same ad your competitors are running.
Four fields and no more: the objection being answered, the proof point from a review or from your data, the awareness stage of the audience, and the format. A brief a creator cannot film from is not a brief, it is a note.
Three to five genuinely different arguments, not twenty variants of one, and tag each ad with the insight it came from. The tag is the whole point: an untagged test tells you which file won and nothing about why, which is a result you cannot reuse.
Read the outcome against the argument rather than the asset. Learning that sizing anxiety beat price anxiety transfers to the next ad, the next email and the next landing page. Learning that video 7B won transfers nowhere, because you cannot run that file forever.
The pipeline dies at step one if reviews live in an app nobody exports and comments live in a notification feed. Selzee connects Shopify, Meta Ads, TikTok Ads and review platforms like Judge.me so the signal arrives continuously rather than being collected by hand before each campaign.
The step humans do worst. Reading two hundred reviews, you remember the vivid ones and lose the frequency, which is exactly backwards. This is the part worth automating, and it is what Selzee actually does: rank objections and desires by how often buyers raise them, with the quote behind each.
Consistency in the brief is what makes results comparable across a quarter. If ad one is briefed differently from ad forty, the pattern you think you see is an artefact of your own paperwork. One template, every time.
Selzee renders statics from a template library with hook overlays and your brand assets. It is not a design editor, has no canvas, and does not manage a brand kit or export per-platform sizes. If you need that layer, Canva and Adobe Express do it well and this page is not competing with them.
Whatever you tag with, ad platform naming conventions or a sheet, it has to outlive the person who set it up. The compounding advantage in this whole loop lives in the tag, and it is the first thing teams drop when a launch gets rushed.
Rising frequency with falling click-through is creative fatigue and needs a new argument. A high hook rate with a poor hold rate means the opening promise and the body disagree. Both look like media problems in a dashboard and are creative problems in fact.
Three failures account for most of it, and none is a creative failure. The first is treating research as a project rather than an input: a team mines reviews once before a launch, then runs on that snapshot for two quarters while feedback keeps arriving. The second is dropping the tag under deadline, which quietly converts every subsequent test into an anecdote. The third is briefing loosely and letting the creative team fill the gaps, so two ads differ in ways nobody recorded and the comparison between them means nothing.
A fourth is worth naming because it looks like progress: using a general-purpose AI tool to write the copy. It returns polished, on-length, forgettable output, because the only input it had was your product name and your own description of the brand. It never read a review. That is not a prompt problem you can fix with a better prompt. Craft-level mistakes, as opposed to process ones, are covered in the ad creative guide.
Nobody should write step four from nothing. The ad script templates cover six formats with runtimes and funnel stages, three of them written out in full with timing cues and bracketed placeholders: problem-agitate-solve, testimonial-led, before and after, UGC unboxing, objection-handling, and hook-first short form. Pick by funnel stage rather than by preference, since a testimonial aimed at cold traffic asks strangers to care about a customer of a brand they have never heard of.
For the brief itself, the creative strategy templates give a reusable structure, and Meta ads templates covers the placement-specific variants. Creator-led formats have their own conventions, handled in the UGC script guide. All of them are the same trade: the structure is free, and the value you add is filling the brackets with something a customer actually said.
One caution about template galleries generally. A design template solves layout, which was rarely the constraint. Carousel, static and video each want a different argument shape rather than the same copy poured into a different frame, so choose the format after you know what the ad has to prove, not before.
One rule saves most of the work: the argument transfers, the delivery does not. An objection that converts on Meta converts on TikTok and in email, because it is a fact about your buyers rather than about a placement. What changes is register and crop. Rewrite the read, keep the claim, and you get four channels from one research pass instead of four research passes.
Facebook and Instagram now run as one delivery surface, so a single ad set can serve feed, Reels and Stories with the split decided by predicted performance. Produce a vertical master that survives a centre crop and you cover the set. TikTok wants first-person handheld delivery and punishes anything that reads as produced, which is a rewrite of the read rather than a new script. Google Ads splits into search, where the asset is text and the claim has to fit a headline, and display, where the constraint is sizes rather than storytelling. LinkedIn tolerates a longer setup and a more explicit business case. Amazon is closest to catalogue work, where the product image and the title carry more than any hook.
Check current sizes and aspect ratios against Meta's own ads guide and each platform's equivalent rather than a third-party list, including this one, because specs change and stale numbers get assets rejected. To see how a given argument is already being made in your category, the Meta Ad Library is free and shows every live ad; the method for reading it usefully is in competitor ads analysis.
"Teams think they have a creative problem and they usually have a reading problem. The hook that would have worked this quarter was already written, by a customer, in a review nobody opened. Generating ads is mostly just building the habit of looking there first."
The tools ranking for this term lead with named brands and headline numbers. Those results belong to those vendors and their customers, and repeating them here would tell you nothing about this process, so there are no borrowed logos or invented percentages on this page.
What teams running this loop describe is consistent in shape rather than in figures. The research pass that took a week takes an afternoon, so more tests ship per month from the same headcount. Briefs stop being debated, because the customer quote sits next to the claim and settles the argument in the document instead of the meeting. Angles surface that nobody would have reached by hand, since the useful phrase is rarely in the first fifty reviews anyone reads. And the account starts producing transferable lessons, because tagged tests accumulate into facts about buyers rather than a folder of retired files.
It works across team types for the same reason: the input is customer feedback, which every brand with customers already has. A solo operator and a ten-person growth team run the identical six steps at different volumes. When we can publish a measured result with a real brand attached, we will. Until then the honest test is your own data.
A specific claim someone could disagree with, aimed at an objection the audience actually holds, delivered in the first three seconds. Effectiveness is mostly decided before production: Nielsen credits creative rather than targeting or media spend with up to 89% of a digital ad's in-market success, and no amount of polish rescues an argument nobody cares about.
Sound-off legibility, a complete statement in the opening beat, and one claim rather than four. Meta and TikTok both reward creative that stops the scroll on its own merits now that automation handles targeting, so the differentiator is what the ad says, not who it is shown to.
Treat each card as one beat of a single argument rather than as a separate ad. Card one carries the hook, the middle cards carry proof, the last carries the CTA. The most common failure is using a carousel as a product catalogue, which gives the viewer no reason to swipe past card one.
Vertical 9:16 for Reels, Stories and TikTok, square or 4:5 for feed, and check the current specs against Meta's own ads guide rather than a blog post, because they change. Produce one vertical master that survives a centre crop and you cover most placements without reshooting.
New signups start free with no card required, which is enough to run a real research pass and judge the output. There is no time-boxed trial. Paid plans are credit-based with the same full feature set on every tier, so the only thing that changes between them is volume.
It reads from them. Meta Ads and TikTok Ads connect for performance signal, alongside Shopify and review platforms like Judge.me. Selzee does not publish ads on your behalf. You launch in the ad platform, having arrived there knowing what to launch.
The research half does, because it needs no craft: connect your sources and the objections rank themselves. The production half still needs someone who can shoot or edit, or a creator brief. What the process removes is the blank page, which is the part that stops non-specialists cold.
Selzee runs the first four steps for you: it reads your reviews, comments, product data and campaign results, ranks the objections your buyers actually raise by how often they raise them, and returns hooks and briefs with the customer quote attached to each claim. Your strategist edits and approves instead of starting from a blank page. Bring your store and the product you have been avoiding, and we will run it live so you judge the output rather than the pitch.
Request a DemoPrefer to explore on your own? Get started or browse free ecommerce tools.
Keep exploring: the ad script templates, creative strategy templates, the UGC script guide, competitor ads analysis, scoring ad creative.