How to Create Content Based on Posts That Already Worked

Working from posts that already performed is sound practice and the most commonly botched research task in social media. The trap is obvious once stated: the post worked for an account with a particular audience, at a particular moment, with a particular history behind it. Copy the words and you inherit none of that.

What transfers is structure. The hook pattern, the format, the length, the specific thing it promised and delivered. This covers where to find that reliably, what the data is worth from each source, and how to get from a spreadsheet of other people’s posts to something you can publish.

Start with your own data

This is the step everyone skips because competitor research feels more productive. Your own top posts are better training data than anyone else’s, because the audience that responded is the audience you still have.

Export twelve months from each network’s native analytics, sort by whichever metric maps to your actual goal, and take the top twenty. Then answer three questions about them. What format were they — single image, carousel, video, text? What was the first line doing — a question, a number, a claim someone could disagree with, a result? And what specifically did they promise?

Pick the metric carefully, because this is where the exercise usually goes wrong. Sorting by likes selects for posts that flattered your existing audience. Sorting by shares selects for posts that reached past it. Sorting by saves selects for reference material. If you are trying to grow, shares and saves are the honest signals and likes are a vanity sort.

Twenty posts is usually enough to see two or three repeating patterns, and those patterns are worth more than any competitor analysis you could run.

Public sources worth the time

For other people’s posts, several platforms publish this data themselves, free, with no scraping involved. These are better than any workaround because they are stable and they are not going to get your account restricted.

TikTok Creative Center is the strongest of the set and the most under-used. It shows top-performing organic content and ads by region and category, trending sounds, and trending hashtags with volume over time. If you work on short video anywhere, the patterns transfer to Reels and Shorts. For the tools that actually publish it, see best TikTok schedulers.

Meta’s Ad Library lists every active ad on Facebook and Instagram, searchable by advertiser. Ads are not organic posts, but an ad that has been running for four months is a message somebody is paying to keep live, which is a stronger signal of what converts than any engagement count.

YouTube search, sorted by view count and filtered to the last month, tells you which framings of a topic earn attention right now. Titles and thumbnails are the lesson; the videos themselves are mostly not.

Reddit and niche forums answer a different question — not what performed, but what people are confused about. A thread with four hundred comments arguing about something is a content brief, and the top comment is usually the angle.

Tools that do this properly

If you want competitor numbers in one place rather than six tabs, a few tools in this directory read them through official APIs and partner data.

Metricool is the value pick at around $20 a month for up to five brands, and competitor benchmarking is included rather than being an upsell. You add competitor profiles and it tracks their posting frequency, engagement rate and best-performing content alongside yours. For most people researching a niche, this is the whole answer.

Agorapulse and Sprout Social go further into listening — tracking mentions and conversation volume rather than just competitor posts — and are priced accordingly at $79 and $199 per user per month. Worth it when reacting to conversation is somebody’s actual job, and overkill when you want to know what format to post on Thursday.

Worth knowing: most schedulers in this directory have weak analytics and no competitor data at all, SchedPilot included. Publishing tools and research tools are different products, and the common setup is one of each.

Where scraping fits, honestly

Scraping comes up constantly in this context, so here is the practical reality rather than either the scare version or the sales version.

Collecting public data at low volume for your own research — a few hundred public posts to spot a pattern — is ordinary competitive research, and courts in several jurisdictions have treated scraping genuinely public pages as lawful. Where it stops being defensible is specific and worth knowing. Logging in and scraping from behind authentication puts you in breach of terms you accepted. Defeating rate limits, CAPTCHAs or bot detection moves it from a terms problem toward a computer-misuse problem in some jurisdictions. Collecting personal data on individuals brings GDPR and similar regimes into it regardless of whether the page was public. And republishing someone’s content, rather than analysing it, is a copyright question with a clear answer.

Set that aside and the engineering argument still points elsewhere. Scrapers break. The platforms change their markup deliberately and often, so a scraper is a thing you maintain forever to get data that TikTok’s Creative Center publishes for free and Metricool gives you for $20 a month. The number of teams whose research genuinely requires a scraper is small, and most of them are building a product rather than planning next week’s posts.

If you do need volume, use the official research channels first: TikTok and Meta both run research APIs, and licensed data providers sell access to historical social data. Slower to set up, and it does not stop working on a Tuesday.

Turning findings into your own content

You now have a list of posts that worked. The step that produces something publishable is extraction, and it takes about an hour for twenty posts.

For each one, write down the structure and throw away the words. “Opens with a counterintuitive number, three-line explanation, one concrete example, question to the reader.” That sentence is reusable across any topic. The original caption is not.

Then look for what repeats across the whole set. Patterns appearing in twelve of twenty posts are how the format works on that platform. Patterns appearing once are that account’s voice, or luck, and copying them gets you neither.

Now write against the structure using something only you have — your data, your client’s result, your mistake, your product’s behaviour. The structure is borrowed and the substance is yours, which is the difference between studying a format and plagiarising a post.

Test it properly. Three or four posts on a new structure, same week, same times, and compare against your own baseline rather than against the account you took it from. Their numbers reflect their audience and tell you nothing about yours.

Keeping it running

Done once this is a project. The version that compounds is a monthly habit: pull your own top twenty, check two or three competitor accounts, add any new structure to a running list, and retire the ones that stopped working.

The mechanical half of that automates well. A workflow can pull your analytics on the first of the month, pull competitor metrics from whichever tool you use, write both into a spreadsheet and flag anything that beat your median by a wide margin. Our n8n workflow tutorial covers the publishing side of that pipeline, and the same approach applies to collecting the numbers.

What does not automate is the extraction. Reading twenty posts and naming what they have in common is judgement, and handing it to a model gets you generic observations about hooks and value. Thirty minutes of your own attention beats it comfortably.

If you would rather keep the research and the content calendar in one place, CoSchedule is built around that pairing. It will not do the extraction for you either.

Three ways this goes wrong

Rewriting a competitor’s post with AI. It produces something recognisable to anyone who follows both accounts, and it reads like a worse version of the original because the specifics that made it good were theirs.

Chasing formats that select for the wrong audience. Engagement-bait structures work, in that they generate engagement. They also fill your audience with people who will never buy anything, and that is hard to undo.

Copying a big account’s playbook. An account with half a million followers can post a one-line observation and get thousands of likes, because the followers are the asset. Study accounts roughly your size, or one step ahead.

Frequently asked questions

How do I find a competitor’s best-performing posts?

For TikTok and short video, the TikTok Creative Center publishes top content by category and region for free. For Facebook and Instagram messaging, Meta’s Ad Library shows every ad an advertiser is running. For engagement numbers across networks in one place, Metricool’s competitor benchmarking at around $20 a month is the cheapest credible option.

Is it legal to scrape social media posts?

Collecting genuinely public data at low volume has been treated as lawful in several jurisdictions, but it still breaches most platforms’ terms of service. It becomes considerably riskier if you scrape from behind a login, defeat bot protections, or collect personal data on individuals, and republishing content rather than analysing it is a separate copyright problem. This is not legal advice — if you are building something that depends on it, ask a lawyer in your jurisdiction.

Which metric should I sort by?

Shares and saves if you want to grow, because they indicate the post travelled beyond your existing followers or was worth keeping. Likes mostly tell you which posts pleased the audience you already have.

Can I just ask an AI to analyse top posts for me?

It is useful for summarising and tagging a set you have already collected. It is weak at the part that matters, which is noticing why a specific post worked for a specific audience, and it tends to return generic advice about hooks and storytelling that you could have written without the data.

How many posts do I need to look at?

Twenty of your own and twenty from two or three comparable accounts is plenty. Beyond that you are collecting rather than learning, and the patterns stop changing.

How often should I redo this?

Monthly for your own numbers, quarterly for competitors. Formats on short video turn over in weeks; on LinkedIn they last the better part of a year.