Paste 5 to 30 posts from a competitor, a creator or your own account. AI finds the hooks, formats and gaps, then writes your next 10 content ideas.

Choose a competitor, a creator you admire, or your own content. Add one line about your brand and one about your audience.
Copy 5 to 30 posts into the box, newest first. Add views, likes, comments, shares and saves if you can see them.
Click Copy and open Claude or Copy and open ChatGPT. Paste into the new chat and press enter. Reply DASHBOARD for a visual version.
Before sharing the Decoder, we ran it on our latest 10 TikToks, scripts included. One was a paid partnership, so it was left out of the comparisons. Here’s what it found.
It also flagged the limits: 10 posts is a small sample, and several patterns rest on one or two posts. So we’re testing them, not treating them as rules.
One AI tool or system you can use, every week. Free. Get the newsletter →
| Platform | Usually visible on other people’s posts | Tip |
|---|---|---|
| TikTok | Views, likes, comments, saves, shares | Views show on the profile grid. Open each video for the rest. |
| Reel views, comments, often likes | Saves and shares are private. Use what you can see. | |
| Reactions, comments, reposts | Impressions are private. Reactions are a fair stand-in. | |
| YouTube | Views, likes, comments | Shorts and long videos behave differently. Don’t mix them. |
Decoding your own content? Use your app’s analytics instead. You get saves, shares and watch time, which makes the results much sharper.
Copy the prompt below, fill in the four details at the top, paste your posts at the bottom, and send it to Claude or ChatGPT.
You are a content strategist. Decode the social posts I paste below and turn them into a content plan for me. MY DETAILS (fill in, or leave blank) - Mode: [COMPETITOR / CREATOR I ADMIRE / MY OWN CONTENT] - Whose posts these are: [@handle and platform] - My brand or what I do: [one line] - My audience: [one line] RULES 1. Use only what is in the posts. Never invent posts, numbers or facts. Point to evidence by post number, e.g. (P4). 2. Metrics: if views, likes, comments, shares or saves are included, rank performance with them and say "best performing". Use medians, not averages. If you give a rate, state the formula once (e.g. saves ÷ views). If there are no metrics, say "most common" and never claim a pattern performs better. 3. If one post has more than 5x the median views, call it an outlier and say the conclusion leans on one post. Treat any group of 2 posts or fewer as a weak signal. 4. Paid partnerships and ads: list them, then leave them out of performance comparisons. 5. Hooks: if only captions are provided, label hook findings "caption-based" and say the spoken or on-screen hook may differ. Only call them video hooks if a script or on-screen text is included. 6. Gaps: only say a topic or format is "underrepresented in this sample of [X] posts". Never say "never covers", "ignores", "has no content on" or "does not appear". 7. Under 10 posts: add "Early signals only. Add more posts for stronger conclusions." If captions are only hashtags and there is no script, say the analysis is limited and ask for the script or on-screen text. 8. Patterns show what went together, not what caused results. Say "linked with", not "caused". 9. Keep it tight: short lines, tables where they help, no filler. If my brand and audience are blank, keep the ideas general and tell me filling them in gives better ideas. OUTPUT, in this exact order ## Top 3 findings Three bullets, the most useful discoveries, each with evidence. ## Do this next Three actions for my next posts, each with the one metric to watch. ## 1. Snapshot One table: account, platform, posts analysed, date range, metrics included (yes/no), paid posts excluded, confidence (low / medium / high). ## 2. Hook patterns Table: Hook type | Example opening line (post #) | Posts | Performance (median views, or "no metrics") | Source (caption-based or script). ## 3. Formats and length Table: Format (e.g. list, tutorial, ranking, demo, opinion) | Typical length | Posts | Performance. ## 4. Topics The 3 to 5 topics that come up most, each with post numbers. ## 5. Calls to action What viewers are asked to do (comment a keyword, follow, save, click) and how often. ## 6. What's working 3 bullets. With metrics: what the best performing posts share that the weakest don't. Without metrics: the most consistent patterns. Note any outlier or weak signal. ## 7. Underrepresented in this sample 3 topics or formats underrepresented in these [X] posts that could be an opening for me. One line each on why. ## 8. My next 10 content ideas Numbered list. For each: - Hook: the exact first line, ready to use - What you show: one line on what the viewer actually sees or learns - Format and length - Why: the pattern it borrows, with post numbers At least 5 of the 10 must bring a genuinely new angle (a use case, demo, comparison or audience problem these posts don't use), but every idea must still cite the pattern it builds on. Adapt every idea to my brand and audience. Never reword the original posts. Finish with one line: "Want this as a visual dashboard? Reply DASHBOARD." If I reply DASHBOARD: build a single-page interactive dashboard from this analysis using your artifact, canvas or code tool. Include the top 3 findings, the snapshot numbers, a bar chart of hook types by post count (and median views if metrics exist), the format table, the underrepresented list and the 10 ideas as cards. Use only numbers from the analysis. If you cannot build interactive output, give me a clean text version laid out like a dashboard instead. POSTS Paste 5 to 30 posts using this format, one block per post: P1 Caption: Script or on-screen text (recommended): Length: Views / Likes / Comments / Shares / Saves: Paid or ad? (yes/no): P2 ...
Stop copying the prompt every time. Save it once, then just paste posts.
Free plans included.
Claude Pro and above: install it as a skill. Claude then uses the Decoder automatically whenever you paste posts and ask what’s working.
Free plans included.
Apify is a web data tool with ready-made scrapers. The free plan includes $5 of credit every month with no credit card. Pulling 30 TikTok posts uses roughly 10 to 15 cents of that free credit, so you can run it many times a month without paying anything.
text, createTimeISO, playCount, diggCount, commentCount, shareCount, collectCount, videoMeta.duration
On TikTok, diggCount is likes and collectCount is saves.
Get the scripts too. Captions often say something different from the video, and the hook usually lives in the first spoken line. Set Subtitles & Transcription to download subtitles before you click Start. The export then includes a subtitle link for each post: open the links for your best and weakest few posts and paste the spoken lines in. Or use the power path below, where the AI fetches them for you. Transcribing videos that have no subtitles costs extra, about 5 cents per minute of video.
Let the AI pull the posts itself. Connect Apify, then ask:
Pull the last 30 TikToks from @handle and decode them.
https://mcp.apify.com
Only scrape public content, and check each platform’s terms before you use the data.
Prices checked October 2026.
When the analysis finishes, reply with one word: DASHBOARD. The AI builds a one-page visual version with your top findings, a chart of hook types, the format table, the gaps and your 10 ideas as cards.
The tools, prompts and systems we actually use. Free, unsubscribe anytime.
Get the newsletterWe set up AI that tracks your competitors and your own content, and turns it into fresh content ideas for your team.