Creative Intelligence from Stored Ad Evidence

Stored ad observations

Data as of 2026-10-14 04:44:15.773215+00. These counts cover stored records in this topic.

Ads observed

14955

Advertiser labels

4583

Ready page screenshots

27

The advertiser count uses distinct stored labels, not verified business identities. A missing screenshot remains unavailable.

Niche directory

• NewsBreak make money online · 649 stored ads
• NewsBreak gifts home decor · 601 stored ads
• NewsBreak blood sugar · 477 stored ads
• NewsBreak weight loss · 469 stored ads
• NewsBreak brain health · 346 stored ads
• NewsBreak credit debt · 335 stored ads
• NewsBreak family · 239 stored ads
• NewsBreak apparel ecommerce · 223 stored ads
• NewsBreak home improvement · 207 stored ads
• NewsBreak insurance · 247 stored ads
• NewsBreak heart circulation · 201 stored ads
• NewsBreak health general · 192 stored ads
• NewsBreak benefits eligibility · 185 stored ads
• NewsBreak skin health · 176 stored ads
• NewsBreak neuropathy · 168 stored ads
• NewsBreak lung respiratory · 163 stored ads
• NewsBreak quiz engagement · 151 stored ads
• NewsBreak senior health hook · 145 stored ads
• NewsBreak utility app · 141 stored ads
• NewsBreak hearing · 131 stored ads
• NewsBreak telecom · 128 stored ads
• NewsBreak home goods · 129 stored ads
• NewsBreak mens health · 103 stored ads
• NewsBreak games gambling · 98 stored ads
• NewsBreak dental · 109 stored ads
• NewsBreak auto general · 99 stored ads
• NewsBreak consumer tech · 95 stored ads
• NewsBreak app install · 90 stored ads
• NewsBreak pets · 90 stored ads
• NewsBreak retirement · 87 stored ads
• NewsBreak joint pain · 88 stored ads
• NewsBreak news faith utility · 81 stored ads
• NewsBreak investing · 79 stored ads
• NewsBreak medicare · 78 stored ads
• NewsBreak housing assistance · 85 stored ads
• NewsBreak vision · 75 stored ads
• NewsBreak content clickbait · 65 stored ads
• NewsBreak streaming entertainment · 64 stored ads
• NewsBreak marriage · 66 stored ads
• NewsBreak travel · 61 stored ads
• Rumble beauty skincare · 66 stored ads
• Rumble first aid · 43 stored ads
• Rumble heart circulation · 31 stored ads
• Rumble shooting sports · 31 stored ads
• Rumble creator tools · 26 stored ads
• Rumble blood sugar · 23 stored ads
• Rumble app install · 22 stored ads
• Rumble brain health · 18 stored ads

Analysis limits

Hook frequency and landing-page pattern shares are unavailable. The source does not supply reviewed classifications for these measures.

Advertiser profile links are unavailable until the business identities pass a separate review.

Define the research question

Creative intelligence connects the words, images, and sequence in a creative to a research question. A useful analysis preserves the difference between a visible feature, a repeated pattern in a defined sample, and a hypothesis for a new test.

Choose one decision for the first review. A question about an opening message needs different evidence from a question about a destination page. Record the fields you need before you select a sample, so that the available fields do not silently change the question.

Compare sources by their evidence

SwipeHarbor supports NewsBreak, Rumble, YouTube, and Meta research. TikTok is coming soon. The example module on this page shows the stored records available through its source reader. It does not claim that those examples represent every ad on a platform.

Use the comparison directory to check tool scope. Keep vendor statements separate from observations in your own sample. A feature list does not establish the depth or completeness of a specific search result.

Official transparency sources can help with advertiser and platform research. Their fields and selection rules differ. Preserve the source name with every note rather than merging unlike records into a single unsupported measure.

Move from observation to an original brief

Record the ad reference, the visible message, and the source dates. Add destination evidence only when it is present. State any gap before you write the hypothesis. A blank field can otherwise look like a complete review to the next reader.

Describe the original test you propose and the question it will answer. Keep claims review separate from creative preference. An example in a library is not an endorsement and does not verify an advertiser statement.

Define the units of a creative

Creative intelligence depends on breaking an ad into parts that can be compared. A practical set of units is the opening, the subject or problem, the explanation or demonstration, the proof elements, the offer, and the request to act. For video, add the order and timing of these parts. For images, add the visual hierarchy: what the eye meets first, second, and third.

Agree on these units with your team before you tag any records. If one reviewer calls a line the hook and another calls it the headline, the tags will not line up and any count across them will mislead.

Write a short definition for each unit with one neutral example drawn from your own past work. Keep the definitions in the same place as the tags so new reviewers can apply them consistently.

Observations first, labels second

A creative label such as curiosity hook or authority angle is an interpretation. Before you apply a label, record the observation it rests on: the exact words, the visible scene, the spoken line. Another reviewer can then check whether the label fits.

Keep a small, stable label set. Every new label makes past tags harder to compare. When a record does not fit an existing label, note it as unclassified and review those records together later, rather than inventing a label for each one.

Labels are research tools, not facts about the advertiser’s intent. An ad tagged as a fear appeal may have been written with a different goal in mind. Keep the label in your notes and out of any claim about what the advertiser meant.

Counting patterns honestly

Creative intelligence often turns into counting: how many records in a sample use a question opening, how many show a product in the first scene. Counts are useful only when the sample is defined. State the platform, the niche, the date window, and the selection rule for every count, and include the total number of records in the sample.

A pattern that is common in a sample is common in that sample. It may reflect how the library captured ads, how the sample was chosen, or which advertisers were active, as much as anything about the category. Do not describe a frequent pattern as an effective one. Frequency in a library says nothing about results.

When you compare counts across two samples, keep their definitions side by side and note any difference in size or source. A small sample can change sharply when a single advertiser adds several versions of one idea.

Use transcripts and screenshots for different questions

A transcript is the best evidence for spoken structure: how a presenter opens, how an argument builds, where a qualification appears. It cannot show on-screen text, product shots, or the pace of edits. A frame-by-frame review answers those questions.

A destination screenshot answers a different question: what the reader met after the click. Creative intelligence that ignores the destination misses the second half of the message. Where a screenshot exists, tag the page with the same discipline as the ad: headline, first visible section, proof elements, offer, request to act.

Keep each tag linked to the evidence type it came from. A tag drawn from a transcript should not be presented as a visual observation.

Version families and repeated ideas

Advertisers often run several versions of one idea with small changes: a different first line, a new thumbnail, another background color. Treat these as a family when the evidence shows they share a core, and keep each version as its own record. Counting every version as a separate idea inflates the apparent popularity of a pattern.

Do not merge records into a family on appearance alone. A shared phrase or presenter is a reason to compare, not proof that two ads are versions of one another. Record the basis for any grouping you make.

Version families are useful research. Comparing the parts that changed with the parts that stayed the same can suggest which elements an advertiser treated as fixed. It cannot tell you which version worked better.

From pattern to hypothesis

The output of creative intelligence is a hypothesis for your own test, not a conclusion about the market. Write it in the form: we observed this pattern in this defined sample; we propose to test an original creative that uses a similar structure for our offer, because of this reason specific to our audience.

Keep the reason grounded in your own knowledge of your product and customers. A pattern borrowed without a reason is guesswork with an extra step.

Plan the test so that it can actually answer the question. If you change the opening, the visual, and the offer at once, a result cannot tell you which change mattered. Change one major element per comparison where your budget allows it.

Coverage and sources for creative research

SwipeHarbor captures NewsBreak and Rumble creatives with observation dates, destination screenshots where available, and transcripts for video where one is ready. SwipeHarbor covers NewsBreak, Rumble, YouTube, and Meta. TikTok is coming soon. Taboola is on the roadmap.

For Google Search and TikTok, use the official libraries: the Google Ads Transparency Center at https://adstransparency.google.com/, and the TikTok Commercial Content Library at https://library.tiktok.com/. Tag records from those sources with the same units and labels so that your analysis stays consistent, and record the source for each.

Respect ownership throughout. Creative intelligence studies structure and choices. The words, footage, and images in the records belong to their advertisers.

Angles, hooks, and formats are different layers

Creative teams use the words angle, hook, and format loosely, and the looseness causes trouble in analysis. A practical split is this. The angle is the reason the ad gives for caring: saving time, avoiding a risk, joining others. The hook is the specific device that opens the ad: a question, a surprising image, a direct address. The format is the container: talking presenter, product demonstration, slideshow, article-style page.

Tag each layer separately. An ad can share an angle with another ad while using a different hook and a different format. If you collapse the layers into one tag, you lose the ability to see those combinations.

When you write a test hypothesis, name which layer you are changing. A test that swaps the hook while keeping the angle and format fixed answers a cleaner question than one that changes all three.

Language analysis without false precision

Text and transcripts invite word counts and phrase frequencies. These can help you see vocabulary in a sample, but they are easy to over-read. A phrase that appears often may come from a few advertisers repeating one script. Before you report a phrase as common, check how many distinct advertisers used it and how the sample was built.

Read the phrase in context before drawing a conclusion. The same words can serve as a question in one ad and as a qualification in another. Counting without reading treats them as one thing.

Report language findings as descriptions of the sample, with the sample definition beside them. Avoid turning them into rules about what copy should say.

Visual analysis needs agreed descriptions

Visual tagging is harder to keep consistent than text tagging, because people describe images differently. Agree on a short vocabulary for the elements you care about: product visible or not, person visible or not, text overlay present or not, setting indoors or outdoors. Prefer descriptions that two reviewers would apply the same way.

Avoid subjective visual tags such as premium look or authentic feel unless you define them with examples. A subjective tag applied by different reviewers produces counts that measure the reviewers as much as the ads.

For video, tag the elements in the opening interval and in the closing interval separately. The opening and the request to act often use quite different visual choices.

Bringing your own results into the analysis

The most valuable creative intelligence links competitor observations to your own tested results. When you test an idea that came from research, tag your own creative with the same units and labels you used for competitor records. Over time, you build a record of which structures your team has tested and how they performed in your own accounts.

Keep your results and competitor observations in clearly separate fields. Your results are measured. Competitor records are observed without outcomes. A shared tag set lets you compare structures across both, while the separate fields keep you from mixing measured and unmeasured data.

This record becomes a research asset that no outside library can supply, because it is built from your own tests.

Destination patterns as part of the creative

The page after the click is part of the creative system, and the same discipline applies to it. Useful page units include the headline, the first visible section, the explanation or story, proof elements, the offer, and the request to act. Record the page type the screenshot shows, such as an article, a product page, a quiz, a form, or a video page, and leave it blank when the capture is missing.

Link each page note to the ad that led to it. A pattern that only appears when you look at ad and page together, such as a question in the ad answered by the page headline, is invisible in an ad-only analysis.

Keep in mind that a page capture is one visit at one time. When you count page patterns across a sample, state how many records had a usable capture and exclude the rest from the count rather than guessing their type.

Tools by research job

These are vendor descriptions. Check current terms with the vendor before a purchase.

Read the full platform comparison and evidence limits

Common questions

Related research