Competitor content audit: turn findings into concepts with AI sketches

Most competitor ad guides stop at the spreadsheet. They tell you how to find ads, how to tag them and which patterns to spot. Then the work stalls, because turning "three of them lead with a price" into something a client can look at and choose still takes a designer a day. This page covers the whole run for an agency: where to look, what to save, how to take an ad apart, and how to get from a finding to a concept board with AI sketches, with a list of rules so you do not copy anyone.

If your question is the narrower "can I rebuild a winning ad with my own product", that is a different task, covered in recreating a winning ad with your product. This page stays on the audit and on sketching new directions. Legal points are information, not legal advice. Check local law and each platform's current rules.

What a creative audit is for

An audit should end in one decision, such as "which two hooks do we test first for this client in March". If you cannot name the decision, you are building a swipe file, and swipe files grow and never get used. Before you pull a single ad, write the decision at the top of the sheet.

Two rules keep the audit honest. First, you are looking for patterns and gaps, not ads to copy. Second, nothing you find is proof. An ad that runs for a long time is a lead, not a result.

Where to look, and what each library does not show

These are the three free libraries every guide names. Open each one in a browser and check what it currently shows, because layouts and rules change.

Library	What you usually get	What it does not give you
Meta Ad Library	Ads from an advertiser's pages: the creative, copy, format, platforms, destination and a start date	Spend, reach or results for ordinary commercial ads. Spend ranges appear mainly for political and social-issue ads. Ads that stop running tend to disappear from view
Google Ads Transparency Center	Ads shown by an advertiser across Google's ad products	Results and spend. The page is built with scripts, so check what is shown for your market
TikTok Creative Center	Trending ads and creative ideas, in leaderboards	A full history for one advertiser. It is a trend board, not an archive

A caution on sources: Meta's own library page could not be read when this guide was researched, so the Meta details above come from two vendor guides (Segwise and Creatify) that describe the library. Open it yourself and read what is on screen. LinkedIn also has an ad library, worth checking for B2B clients.

Two practical traps. Set the country filter first: an ad running in another market is not evidence for yours. And manual viewing of public libraries is one thing, while automated scraping may break a platform's terms, so check the terms before you automate anything.

Pull and save: a sheet for 15 to 20 ads per competitor

Ads that stop running vanish from the Meta library, so keep your own archive. Save a screenshot or a screen recording of each ad, with these columns:

Competitor | Library | Country filter | Page or advertiser ID |
Start date seen | Date captured | Days running | Format (image,
video, carousel) | Hook (first 3 seconds or headline) | Promise |
Proof | Offer | CTA | Landing page notes | Status: untested

Pick three to seven competitors in tiers: the direct rivals, a couple of adjacent brands and one outsider from another category who does this well. Fifteen to twenty ads per competitor is plenty for a first pass. Capture on a schedule, weekly or every two weeks, because a one-off capture misses what changes. The weekly rhythm is similar to the one in the weekly ad refresh routine.

Treat longevity as a weak hint

Several vendor guides suggest that an ad running 30 days or more is worth a closer look, and 60 to 90 days is a stronger hint. These are rules of thumb that the sources themselves admit do not prove profit, and some of the statistics they quote come without a method. I use "30 days or more" only to build a shortlist, and I write "untested" beside every pattern until it has run in the client's own account.

Take each ad apart

For each ad on the shortlist, write down five things in plain words:

1. Hook. What stops the scroll in the first three seconds.
2. Promise. What the viewer will get.
3. Proof. The reason to believe: a number, a person, a demo, a guarantee, a review.
4. Offer. Price, trial, bonus or deadline.
5. Call to action and landing page. What happens next, and whether the page matches the ad.

Score each one from 1 to 5 on how clear it is, so you can sort later. Then cluster. You are looking for patterns such as "five of seven lead with a price" or "nobody shows the product being used". The patterns tell you what is crowded. The gaps tell you where a client can look different.

Find the gap and write the hypothesis

One format keeps this tight: "For [audience], test [hook] in [format] because [pattern] may improve [measure]". For example, using a made-up case: "For homeowners comparing three roofers, test a calm one-take walk-through of a real job in a short video, because none of the audited competitors shows a finished job, and it may improve click-through rate." Every hypothesis is a guess. The sketches and the test make it a result. To pick which hooks to test first, see how to rank video hooks for a client, and for turning ideas into labelled drafts for a new client see first draft ad sets inside an hour.

From finding to concept board with AI sketches

This is the step that turns an audit into something a client can choose from. A concept sketch is a rough still or a 3 to 5 second loop. It is there to compare six to ten directions in one meeting. It is not final art, and I would keep the files internal or labelled "concept".

Use this board template, one row per direction:

Column	What goes in it
Competitor ad	Name and your saved screenshot (internal only)
Hook type	Price, question, demo, story, social proof, before and after
What they promise	One line
Gap we saw	What none of them shows or says
Our claim	Must be one we can substantiate (see the claims section)
Sketch prompt	Describes our structure, our product, our colours. No competitor brand, actor or music
Test measure	Click-through rate, cost per result or hook rate, decided in advance

A sketch prompt that stays on your own ground

Still frame for a short video ad. Subject: [our product] on a
[our setting]. Palette: [our hex colours]. Mood: [three words].
Framing: close, handheld, natural light. Leave space at the top
for a headline. No text, no logos, no people.

The tool matters less than the discipline. Any image generator works, including Midjourney (an image generator with an artistic look), Adobe Firefly (Adobe's image generator, trained on content Adobe says it is licensed to use), Canva (an online design tool) or a multi-model platform such as Higgsfield. A designer's hand sketch or a stock photo is faster when the idea is a layout and not a look. For the pitch side, see quick AI moodboards for client pitches and mood video for a client pitch. To test hooks once a direction is chosen, use the hook testing page and testing ad creative before a photoshoot.

Avoid copying: a short rule list and a copy test

Structure is reusable. Footage, scripts, characters and a distinctive look are expression, and those belong to someone else. Here are the rules I would give every designer on the team:

1. Never paste a competitor's ad, logo, character, script or named actor into an image or video generator.
2. Never prompt "in the style of [competitor] ad" or use a competitor's product photo as an image reference.
3. Do not reuse their music, voice, slogan or recognisable on-screen layout.
4. Rebuild the structure (hook, then proof, then offer) in your own words and your own visuals.
5. Do not rely on a tool's filters for legal safety. Some platforms block franchise characters, logos and named living artists, which helps, but it is not your compliance check.

The copy test: take the sketch, cover the logos, and show it to someone who has not seen the audit. Could they name the competitor within three seconds? If yes, change the structure, not just the colours.

Check the claim before you reuse the promise

A competitor running a claim does not make the claim lawful. Before you reuse a promise (fastest, cheapest, number one, a result in days), find the proof that would support it for your client, or drop it. For claims about results or health or money, the FTC (the US consumer protection regulator) publishes guidance on substantiation on ftc.gov, and other countries have their own advertising codes. If the claim is shown through a real person's result, read what an agency should never fake with AI video first.

Copyright and trade dress (the overall look that identifies a brand) questions around lookalike ads are a legal matter, and no primary legal source is quoted here. Treat this page as a prompt to ask your lawyer about edge cases.

Show the client: a one-page brief

I would hand the client one page, not the whole sheet:

• The decision (one line).
• Three to five hook patterns found, with the crowded ones marked.
• One or two gaps.
• Six to ten concept sketches in a grid, each labelled "concept".
• The two to test first, with the measure and the pass line, for example "keep it if cost per lead is below the account's current average after a fixed spend".

That last line is yours to define from the client's own numbers. Competitor data cannot tell you what a pass looks like. For the approval trail, see a client approval workflow for creative.

When to pay for an ad-intelligence tool

Paid ad-intelligence tools add search across platforms, alerts, bulk export and long history. One agency guide (Flighted) puts the point at about 10 or more competitors across several platforms, and says below that the free libraries plus a weekly manual check are usually enough. A vendor guide from Segwise says manual review does not scale past a couple of rivals. The two disagree, and both are vendors, so check the tools against your own volume before buying. If the main aim is judging which of your own videos will land, a different approach is in which video will perform.

Who should skip this

• A client in a market with almost no advertisers. The libraries will be empty, and customer interviews will teach you more.
• A team with no way to test. If you cannot run an A/B test or compare results, the sketches will not turn into learning.
• A brand whose edge is a real person or a real place. Audit for message gaps, then shoot the real thing.

How I'd approach this

I would run the audit as a half-day sprint with one question, save everything in the fixed sheet, and sketch ten directions on the concept board template. I would cut to two, write the claim check next to each, and test them in the client's account with the measure decided first. After the test, I would update the sheet with real results in place of "untested". A shared library of audits, sketches and results is what makes the next client's audit faster, and that library plus an approval step is something Blur Studio builds for agencies.