Quick answer
To build an ad swipe file your AI agent can use, save 30 to 50 ads you admire into one shared folder, label each one with the angle, hook, format, audience and why it works, then point your agent at that folder and an index file whenever it drafts new ads. The labels matter more than the images. They turn a pile of screenshots into patterns the agent can learn from. Use references for inspiration, never as designs to copy.
Key takeaways
- A swipe file is a labelled library of ad examples, not a folder of random screenshots.
- Start with 30 to 50 examples and add a few each week. Greg Isenberg argues the big players need thousands.
- Label every example with angle, hook, format, audience, emotion and why it works.
- Keep a simple index file (a spreadsheet or markdown table) the agent reads first.
- Include your own winners and losers once you have results. That is your most valuable data.
- Borrow ideas, not designs. Never copy another brand’s wording, layout or trademarks.
📘 Main article: This guide supports AI Marketing Agents: What They Need to Make and Test Ads Properly. It turns Greg Isenberg’s “swipe warehouse” idea into a simple system you can set up this afternoon.
In this guide
- What is an ad swipe file for an AI agent?
- Why does a swipe file make AI ads better?
- What should you collect?
- How do you build one, step by step?
- How should you label each example?
- How do you get your agent to use it?
- What mistakes should you avoid?
- Frequently asked questions
- The bottom line
- Related posts
In a recent workshop on Greg Isenberg’s channel, marketer Cody Schneider showed an AI agent producing static and video ads from a brand’s website. Greg then pointed out what most set-ups are missing: good examples. He calls it a swipe warehouse, a growing library of creative references the agent can draw on.
This guide answers one question: how do you build an ad swipe file that an AI agent can actually use? For the full workflow, read our article on what AI marketing agents need to make and test ads properly.
What is an ad swipe file for an AI agent?
It is a collection of ads and marketing examples you think are excellent, each saved with notes that explain why it works. Copywriters have kept swipe files for decades. The difference now is the reader: an AI agent needs clear, consistent labels to spot patterns, where a person can rely on memory.
Think of it as an owned asset. Your swipe file, like your email list and your brand guide, is something you build once and keep improving, whichever AI tool you use next year.
Why does a swipe file make AI ads better?
Without examples, AI agents fall back on the most average ad they can imagine. That is why so much AI creative looks the same.
Greg’s argument in the video is that ad platforms are now good at matching strong creative to the right buyers, so the quality of the creative becomes the main thing you control. Good references raise that quality by showing the agent what “strong” looks like in your market: the hooks, layouts and tone that grab attention.
💡 Tip: Your best examples are often not from your own industry. A bold hook from a bike brand might work brilliantly for an accounting app. Label the idea, not the product.
What should you collect?
Collect variety: different formats, angles and audiences, from inside and outside your niche. Good sources include:
| Source | What to save | Notes |
|---|---|---|
| Meta Ad Library | Ads from competitors and brands you admire | Free and public. Ads that run for months are often performing. |
| Your inbox | Promotional emails with strong subject lines | Useful for hooks and offers. |
| Social feeds | Posts and short videos that stopped your scroll | Screenshot or save the link straight away. |
| Landing pages | Headlines and above-the-fold layouts | Save the page URL and a screenshot. |
| Your own ads | Every ad you run, with results | Mark winners and losers once you have data. |
| Classic books | Principles and famous examples | Greg suggests feeding a creative agent classic advertising books, such as David Ogilvy’s. |
Greg mentions Sublime as one tool for collecting references. A shared Google Drive or Notion folder works just as well to begin with.
How do you build one, step by step?
- Create one home for it. A single folder called swipe-file with an images sub-folder and one index file.
- Set a target. Aim for 30 to 50 examples in the first fortnight. Ten minutes a day is enough.
- Save the original, not a crop. Keep the full ad plus the link, so you can see the context later.
- Name files consistently, for example 2026-11-meta-saas-painpoint-01.png.
- Label each example in the index file using the fields below.
- Add a “why it works” note in your own words, in one or two sentences.
- Review monthly. Remove weak examples and mark which patterns your own tests confirm.
How should you label each example?
Use the same short set of fields for every example, so the agent can sort and compare them.
| Field | Example entry |
|---|---|
| Brand and source | Project tool, Meta Ad Library |
| Format | Static image, 1:1 |
| Angle | Pain point: wasted time in meetings |
| Hook | “Your Monday meeting could have been a two-line message.” |
| Audience | Small team managers |
| Emotion | Relief |
| Visual style | Bold single colour, big headline, product screenshot |
| Why it works | Names a specific frustration, then shows the fix in one picture |
| Our result | Not tested yet / Winner / Loser |
If you use a spreadsheet, export it as a CSV or markdown file for the agent. Plain text is easier for an agent to read than a complicated document.
⚠️ Warning: Do not store anything private in a swipe file you share with an AI tool, such as customer names in screenshots of emails. Crop or blur personal details first.
How do you get your agent to use it?
Tell the agent to read the index first, pick relevant examples, and explain which patterns it is borrowing before it writes anything. A prompt you can adapt:
“Read swipe-file/index.md. Pick the five examples most relevant to this brief. For each, tell me the pattern you will borrow (angle, hook structure, layout). Then draft five original ad concepts for our offer using those patterns. Do not reuse any wording, brand names or designs from the examples.”
- Attach your brief so the agent knows the offer, audience and pain point. See how to write a campaign brief an AI marketing team can use.
- Save the prompt as a reusable skill or saved instruction, so you do not retype it. Our guide to teaching an AI agent a task by showing it once helps here.
- Ask for its reasoning each time. If it cannot explain the borrowed pattern, the concept is probably generic.
- Score the drafts before spending money, using our AI creative director guide.
- Feed results back. When an ad wins or loses, update the “Our result” field.
Next step: collect your first ten examples today, then read the full workflow in AI Marketing Agents: What They Need to Make and Test Ads Properly.
What mistakes should you avoid?
- Hoarding without labels. A thousand unlabelled screenshots teach an agent very little.
- Only saving your own niche. You end up copying the same tired patterns as your competitors.
- Copying instead of learning. Reusing a brand’s words or design can breach copyright or trademark rules and will not stand out anyway.
- Never adding your own results. Your tested winners are the most useful references you will ever have.
- Letting it go stale. Ad styles change quickly. Add a few new examples every week.
Frequently asked questions
How many examples does an AI ad swipe file need?
Start with 30 to 50 well-labelled examples. That is enough for an agent to spot patterns. Greg Isenberg suggests serious advertisers grow towards thousands over time.
Is it legal to save competitors’ ads?
Saving public ads for reference is normal practice. Copying their wording, designs, characters or trademarks into your own ads is not, so use them for ideas only.
What is the best tool for a swipe file?
The simplest tool you will keep using. A shared folder plus a spreadsheet works well. Greg mentions Sublime as one option for collecting references.
Can my AI agent collect examples for me?
It can help, for example by summarising ads from the Meta Ad Library. Keep a person choosing what goes in, because taste is the point of the file.
Should I include bad ads?
Yes, if you label them. Your own tested losers are especially useful because they show the agent what to avoid.
The bottom line
A good swipe file turns generic AI ads into creative that reflects real, proven patterns. Keep it in one place, label every example the same way, add your own results and tell your agent to explain what it borrows. It takes a few minutes a day, and it is an asset you own whichever tools you use next.
Related posts
- AI Marketing Agents: What They Need to Make and Test Ads Properly
- How to Write a Campaign Brief an AI Marketing Team Can Use
- Owned Reach vs Rented Reach: 3 Things to Build So Your Followers Aren’t Your Only Audience

[…] How to Build an Ad Swipe File Your AI Agent Can Actually Use — collect and label the ad references that make an agent’s creative better. […]
[…] you want the reviewer to learn your taste, give it your labelled swipe file too. Our guide to building an ad swipe file your AI agent can use explains […]