How to find viral reels for ads when views lie

A store owner on Reddit described two videos(opens in new tab) that each reached 600,000 views, one on TikTok and one on Instagram. The result for the business: almost no clicks to the site and zero sales. The audience turned out to be mostly in another country. One reply summed it up: views don't mean much on their own.

That's the trap when you look for a reel to base an ad on. The reel with the most views is often a big account having an ordinary day, or a reel that reached people who will never buy. The one worth copying beat its own creator, is still gaining views, and pulled the kind of attention your ad needs.

You don't have to post reels yourself or scroll through hundreds of them. You need one reel that already worked in your niche, and the way it was built. This guide is for people who make product ads: founders, marketers and agencies. It uses three real searches from November 2026.

What you need

• an agent connected to Clipwright.io(opens in new tab) over MCP, the standard way AI agents call outside tools (Claude Code, Cursor or any MCP client);
• your product in a sentence, and the words a buyer would type about the problem it solves.

You ask your agent to find reels that took off on a topic. Clipwright.io searches TikTok and Instagram together and brings back up to 10 reels per platform, one per creator, ranked by how far each reel beat its creator's usual views. Every number is as of the moment it was measured.

Step 1: Rank by the creator, not by views

By the end of this step you can tell a reel that took off from a big account's normal day.

A search for "collagen cream firmness" on 10 November, TikTok and Instagram, last 30 days, returned 11 reels. Here are the five with the most views. All were older than two days, so their numbers had settled. "The creator's usual" is the middle value of the views on that creator's latest videos: a fair picture of a normal day for the account.

By views, reels 1 to 3 look alike. Against their creators' usual numbers they split apart. Reel 1 got 16.5 times its creator's usual: it took off. Reel 3 had about two thirds of reel 1's views, but its creator usually gets 59,000, so 96,000 was a good day, not a breakout. Reel 4 had 14,000 views on an account that usually gets 136,000: a flop by its own standard.

A search for "ai video ads" on 7 November, last 7 days, shows a second trap. Its most-viewed reel had 3.2 million views, about half of what its creator usually gets, but it was less than a day old, so its numbers were still moving (Step 2). Of the reels older than two days, only one from a new creator reached 2×, and Step 2 explains why that doesn't count. Big numbers, and nothing that had clearly taken off yet.

Step 2: Do the math, and know when it lies

By the end of this step you can check any reel yourself and know when to ignore the number.

The math is one division: the reel's views ÷ the creator's usual views. Using the middle value rather than an average means one huge or one dead video doesn't drag "usual" around.

Three cases where the number misleads:

• The creator is new. When their usual is under 1,000 views, a few hundred extra views produce a big multiple. Clipwright.io marks these reels as new creators, doesn't count them as taking off, and ranks them by views per hour instead.
• The reel is less than two days old. Its views are still climbing, so the multiple is too. It's marked as preliminary. Check it again later.
• The reach is tiny. 6× on 2,800 views is a small account's good day. Look for a multiple and views that both hold up.

A second comparison helps when the creator's own history is unusual: views against accounts of about the same follower count. In the "ai video ads" search, one TikTok reel, almost six days old, was only 1.5 times its own creator's usual but 38 times what accounts its size usually get. The creator already outperforms accounts like it, so the reel is still worth a look.

Step 3: Check it's still gaining

By the end of this step you know whether a reel is still gaining, slowing or finished.

Total views tell you a reel's past. A reel the search has seen before, at least an hour earlier, also carries its speed now (views per hour since then) and whether that speed is up or down on the stretch before it.

Two reels, followed over two days:

Two reels from November searches: one still gaining, one finished.

The teal line is reel 5 from the collagen search, the code-word reel in Step 4. Its multiple is low, so it isn't a pick, but it shows how speed reads. At the search it was gaining 53 views an hour, up from 40 in the stretch before. A day later it had slowed to about 33 an hour. The grey reel sat at 269 views for the whole two days: speed zero, nothing to compare. Always read speed next to the time it was measured.

Late takeoffs. A reel older than two days that is still at 3 times its creator's usual or more is marked as a late takeoff. Past two days the multiple is no longer preliminary. Check that it isn't from a new creator, because the mark doesn't exclude them. The 16.5× collagen reel was 7.8 days old when the search found it.

When a reel is both not slowing and well above its creator's usual, Clipwright.io labels it "rising" (2 times or more) or "surging" (3 times or more), comparing it with the creator's other videos at about the same age when there are at least three. None of the searches in this guide caught one. When one turns up, put it near the top of your shortlist.

Step 4: Read the comments and the code word

By the end of this step you know which reels were built to collect leads rather than views.

Each reel also shows how many comments it got per view, compared with the creator's usual rate. A high number means the reel made people write.

Reel 5 in the collagen search had 12,541 views, only 0.3 times its creator's usual. By views it was a flop. But it got 86.8 times the creator's usual rate of comments, and its caption asked viewers to comment a code word: COLLAGEN. The caption made it a lead magnet, built to turn viewers into comments and DMs.

Clipwright.io picks the code word out of the caption and shows it on the reel, with the reason "asks viewers to comment a code word". Code words were common in the B2B search: four of the nine "ai video ads" captions carried one (ADS, MARKETING and two product names).

What to do with it:

• If your ad's job is to collect leads, a code-word reel with an unusual comment rate is a mechanic worth studying, even with low views.
• If your ad's job is reach or a sale on the page, skip it. Its comments most likely came from the offer in the caption, not from the video.

Step 5: Word the search the way a viewer types

By the end of this step your search returns reels from your buyers' corner of the platform, not from creators talking about your industry.

Write three to seven words, the way a viewer would type them. "Collagen cream firmness" is how a buyer talks. "Peptide-based anti-aging topical" is how a label talks, and few reels use it.

Narrow product words can return too little. A search for "freelancer invoice" on 6 November returned 12 reels. Only one beat its creator by 2 times or more, and that one came from a new creator. Most had under 1,000 views. That's a weak result. When fewer than two reels take off, Clipwright.io suggests up to three other ways to word the topic. For a search like that one, it would suggest a shorter form ("freelancer"), a hashtag form ("#freelancerinvoice") and, for English searches, a buyer's search form ("best freelancer invoice"). Run one of them as a new search.

Three things you can ask your agent for:

• A time window. Anything from the last 2 days to the last year; the default is 90 days. A short window finds what's moving now; a long one finds ideas that kept working.
• A language. It filters TikTok results only.
• A different order. Re-sorting a search you ran in the last hour is free: by multiple, by views, by views per hour, or by comment rate.

B2B searches vary. In the "ai video ads" search, the big accounts posted near their usual numbers. A search for "invoicing app" on 6 November, last 90 days, found four reels that beat their creators by 2 times or more, not counting new creators. When the category name returns little, try the problem your buyer has ("send invoices faster").

Step 6: From the reel you picked to your MP4

By the end of this step you have an ad built on the reel's mechanic, approved and priced before it renders.

The path after the pick is the same for any product:

1. Your agent breaks the reel down: its hook, structure, pacing and who speaks.
2. You decide whom your viewer trusts, and where. The maker on the factory floor, a seller in a shop, an expert at the customer's home, or a user at home. Argue it from the reel and its comments.
3. Your agent writes your script on the reel's mechanic, in new words, and plans the ad. Read what the plan won't reproduce, send notes, and approve the version you want.
4. You see the price, then it renders. Where the reel's presenter moves, the plan can add a live shot, priced on its own line. The MP4 comes back to your agent.

How to create a video ad with AI from one brief walks through every step with a real run, and thirty hooks by Friday turns it into a loop for many variants.

What to take from a reel, and what to leave

Take the mechanic:

• the type of hook and how fast it lands;
• the structure: problem, demo, proof, call to action, in whatever order worked;
• the pacing, the number of cuts, where the product appears;
• who speaks and from where.

Leave everything that belongs to the creator or the reel's product:

• the words. Write new ones. A reel's script is someone's work.
• the faces and footage. Your ad gets its own presenter and shots.
• the music.
• the facts and claims. Their numbers describe their product. A claim lifted from someone else's reel is a claim you can't back.

What it costs

A search is 20 credits, charged only when it finds reels. A search that finds nothing costs nothing, and the same search again within the hour answers for free, as does re-sorting it. An account can run 10 searches an hour. The breakdown is 70 credits, charged when it succeeds. A presenter ad is priced per second and quoted before it runs.

The 420 credits an account gets at sign-up cover the whole chain about once: a search (20), a breakdown (70) and an ad of about 30 seconds with the default presenter (about 300), around 390 in total. The ad's charge follows its measured length. A remake plan also adds live shots where the reference presenter moves, priced on the quote before anything runs, so the total can go past the sign-up credits. Treat it as an estimate. Prices change, so take today's from the quote or the price list(opens in new tab).

Common mistakes

Picking by views alone. In the collagen search, reels 1 and 3 both had around 100,000 views; only reel 1 took off. Sort by the multiple, then check the views and the age.

Trusting a new creator's multiple. A usual under 1,000 views makes any good day look like a takeoff.

Judging a reel by its first day. Under 48 hours, the multiple is still moving. Check again.

Reading speed without its time. Speed is true for the hour it was measured. A day later the same reel can be coasting.

Copying a lead magnet for a reach ad. A code-word reel's comments mostly answer its offer, so its pull won't carry over to an ad without one.

Searching in label words. Write what the buyer types, and try the suggested wordings when the result is weak.

FAQ

What counts as a viral reel for an ad?

One that beat its own creator's usual views by a wide margin, 3 times or more is a strong signal, with enough views to mean something. Raw views alone mostly tell you how big the account is.

How do I calculate views against the creator's usual?

Divide the reel's views by the middle value of the views on the creator's latest videos. 138,240 views against a usual 8,369 is 16.5 times.

Is reel views to follower ratio a good measure?

It's a rough one: follower counts predict reel views only loosely. Comparing a reel with the creator's own usual views is more direct. When you do want a size comparison, compare with accounts of about the same follower count, which the search also shows.

Can I copy a viral reel for my ad?

Copy the mechanic: the hook type, the structure, the pacing. Write new words, use your own presenter and footage, and make only claims you can back.

How often should I search?

Whenever you plan a new batch of ads. A short window, a week or two, shows what's moving now. The same search within an hour is free, so re-sort it instead of re-running it.

Your Breakout 5

The Breakout 5 is this week's top five reels in your niche, ranked by how far each beat its own creator, with views, the multiple, age and one line on why each made the list. In a quiet niche, fewer than five will have clearly taken off; the search then suggests other wordings to try. Paste this to your agent, with your own words in the quotes:

Find reels that took off for "send invoices faster" in the last 7 days. Show me my Breakout 5: views, times the creator's usual, age and why each one is there. If fewer than two took off, try the suggested wordings first. Then break down the one I pick and plan an ad for my product on its mechanic, with the price before anything renders.

For your agent

The search is the Clipwright.io MCP tool find_trends (quote it first with quote_trends). The fields this guide talks about:

• x_median: views over creator_median_views; top is true at 2 or more for a creator who isn't new.
• new_creator: the creator's median is under 1,000; such reels rank by views_per_hour.
• x_preliminary: the reel is under 48 hours old. late_takeoff: older than 48 hours with x_median of 3 or more.
• x_size_norm: views over the usual views for accounts with about as many followers.
• speed_now and accel: views per hour since the previous measurement, and that speed over the one before. takeoff is "rising" or "surging".
• comment_rate_x and funnel: comments per view against the creator's norm, and the caption's code word.
• suggested_wordings: returned when the result is weak. posted_within_days, language and sort set the window, the TikTok language and the order.

The chain after the pick is analyze_reference, plan_reference_remake (pass whom the viewer trusts as audience), approve_workflow_plan, quote_workflow_plan, then run_workflow_plan.