How to research ecommerce ads

What Ecommerce advertising looks like

Ecommerce paid social runs on volume: many creatives, fast iteration, and a promotional calendar that repeats each year. Broad campaign types have pushed a lot of the optimisation work into the platform, which means creative and offer are the levers the advertiser still controls. That is convenient for a researcher, because creative and offer are exactly what the ad library shows you.

The trap is that volume looks like insight. A competitor with two hundred live ads is not necessarily outspending you, and a competitor with fifteen is not necessarily small. Some of that count is catalog ads generated from a feed, some is one creative duplicated across placements, and some is a test that will be off by Friday. Count distinct concepts rather than entries, and the picture gets honest quickly.

The rules that shape the creative

Ecommerce is regulated as commerce rather than as a profession, and the rules that bite are about offers, reviews and subscriptions.

• Reviews and testimonials. US FTC rules prohibit fake reviews and testimonials, including buying them and misrepresenting an insider review as independent. Incentivised endorsements have to be disclosed, and creator content in an ad counts.
• Reference pricing. A 'was' price has to be a price you genuinely offered. Permanent strike-through pricing is one of the more commonly enforced problems in the category.
• Subscriptions and negative options. Recurring charges need clear disclosure before purchase, informed consent, and a cancellation path that is not harder than signing up. Rules on this have been tightening, not loosening.
• Delivery and stock claims. Shipping timelines and availability stated in an ad are claims. Countdown urgency that resets on refresh is the version regulators dislike most.
• Product policy. Platform commerce policies restrict entire product categories, and the restrictions differ by country. A competitor's ad running in one market is not evidence that your version will clear in another.

This is a summary of well-established rules, not legal advice. Anything that touches a regulator should go past your own counsel before it goes live.

The angles that keep coming back

The category concentrates on a handful of proven structures, and the variation happens inside them.

• Offer led. A percentage, a bundle or a free shipping threshold carrying the whole ad.
• Problem to demo. Three seconds of the problem, then the product solving it on camera.
• Review read. A real customer review used as the script, with disclosure where the reviewer was compensated.
• Comparison grid. Feature or price grid against the generic category, rarely against a named brand.
• Bestseller framing. Rank or volume as social proof, which is only usable if the number is true.

How to research Ecommerce competitor ads

The Meta Ad Library is free, public and needs no login. In a high-volume vertical the method has to include a way of reducing what you look at.

1. Choose five, not fifty. Three direct competitors at your price point, two one tier above. Price point matters more than category, because offer structure follows margin and margin follows price.
2. Filter before you read. Set the country and platform filters, then scan for distinct concepts rather than entries. Catalog ads and placement duplicates otherwise inflate the count into something meaningless.
3. Map the calendar. Record the start date of every promotional creative across a quarter. Promotional calendars repeat annually, so last year's dates predict this year's. This is the most useful output of the exercise.
4. Separate evergreen from test. Over sixty days live is a considered bet. Under two weeks is a test. Study the first group for angle, the second for what the brand currently thinks is missing.
5. Follow the click. Open the landing page and the first email. The offer in the ad is a third of the mechanic. The threshold, the upsell and the post-purchase flow are invisible from the library.
6. Test one variable. One angle into your account, format and audience held constant, run against your control long enough to clear the noise. Copy five things at once and you learn which one moved: none.

Keyword search across ad text is how you find advertisers you have never heard of. Useful seeds:

• “free shipping over”
• “bundle and save”
• “as seen on”
• “sold out twice”
• “60 day trial”
• “shop the sale”

What the ad library will not tell you

No spend, no impressions, no results, no targeting outside the EU. Non-political ads also vanish when the advertiser turns them off, so anything you did not record is gone. In ecommerce you also cannot see margin. A competitor's discount is only copyable if your unit economics look like theirs, and the library does not publish those.

You can see what a competitor will say in public and how long they have said it. The rest is inference, so let your own creative testing settle it. The full list of blind spots is on the index.

Common mistakes

• Counting entries as budget. Ad count is a function of feed size and placement duplication as much as spend.
• Copying an offer you cannot fund. A discount is a margin decision. Match the mechanic to your own numbers or it wins revenue and loses money.
• Mistaking a test for a strategy. Half of what is live this week will be off next week. Judge by what survived, not by what appeared.
• Ignoring the post-click. The ad is the cheap part of a competitor's funnel. The threshold, the bundle and the email sequence do most of the work.

Doing this without the browser tabs

This is the vertical where manual checking fails first. Sets are large, they change weekly, and the useful output is the diff: what appeared, what disappeared, and what has now survived long enough to be worth copying.

Muze collects it with four tools: list_competitors, scrape_competitor_ads, which starts a Meta Ad Library scrape, get_scrape_status and get_competitor_ads. Call them from ChatGPT, Claude or any MCP client, over one OAuth connection that also covers Google Ads, Amazon Ads and Shopify.

The point is not API access. It is that the research and the change to the account happen in one conversation. Reading scraped ads is a read call, starting a scrape counts as a write and needs a paid plan, and the free tier is 25 read-only calls a month. Account changes preview first and wait for your confirmation, new campaigns are created paused, and Muze never takes a percentage of ad spend.

Frequently asked questions

How do I see what other ecommerce brands are advertising?

Search the brand page name in the free Meta Ad Library, set the country and platform filters, and read the run dates. The Google Ads Transparency Center covers Search, Shopping, Display and YouTube for the same brands.

Why do some brands show hundreds of ads?

Because entries are not concepts. One creative duplicated across placements, plus catalog ads generated from a feed, inflate the count. Count distinct concepts and the number usually collapses.

Can I see which ecommerce ads are performing?

Not directly. The only free signal is how long an ad has been running, and that is an inference. Brands do leave underperforming ads live, especially at low delivery.

Are catalog and dynamic product ads visible in the library?

They appear, but what you see is a sample. The product card is assembled at delivery time from the advertiser's feed, so no library can show every combination.

How far back does the Meta Ad Library go?

Years for political and social issue advertising. For ordinary commercial ads it shows what is running now, so an ad that stops disappears. Keep your own history.

Is copying a competitor's ad legal?

Reading public ads is fine. Reproducing their photography, video or copy is a copyright question. Take the structure, the offer mechanic and the angle, then make the assets yourself.

Other verticals

Keep reading: all eight verticals, ROAS, ad examples, safe zones, Amazon Ads with Muze, Meta Ads with Muze, or how the Muze MCP connection works.