A social media comment export tool for research turns the comment section of a public post into a file you can code, count and quote: one row per comment, with the author, the text, the like count, the time and the reply structure. You have two kinds of options. The platforms run researcher programmes that grant API access after an application, and no-code exporters turn a post link into a CSV, Excel or JSON file in seconds. This guide compares both, shows how ZocialComment handles TikTok, Instagram, Facebook, YouTube, Reddit and Threads in one shared format, and covers the data-quality traps that decide whether your numbers hold up.
We run that exporter, so we will be specific about what it does well and where you need something else.
What a research-grade comment dataset needs
Most comment studies fail on the dataset rather than the analysis. Before picking a tool, check that the output gives you:
- Every public comment, not a sample the platform chose. The app shows "top" comments first and hides the long tail. Your analysis needs the long tail.
- Replies with a link to their parent. Without a parent id you cannot tell a reply from a top-level comment, and any conversation or stance analysis is guesswork.
- A stable comment id. You need it to remove duplicates when you export the same post twice, and to trace a quote back to its row.
- A timestamp, so you can cut the data by day, campaign phase or event.
- The same columns across platforms. A cross-platform study with six different file layouts spends its first week on reformatting.
- A record of how it was collected: which posts, when, and how many rows each.
Hold each option below up against that list.
Official researcher access, platform by platform
Several platforms now run programmes for approved researchers. They are the right choice for some projects and a poor fit for most.
- TikTok offers the TikTok Research API to academic researchers at non-profit institutions in eligible regions. You apply with a project description, wait for approval, and then query videos and comments with code under a daily request quota.
- Meta (Facebook and Instagram) runs the Meta Content Library, accessed through an approved research environment. Data analysis happens inside that environment, with rules about what may leave it.
- YouTube exposes comments through the public Data API: the commentThreads endpoint returns comment threads under a daily quota tied to a Google Cloud project.
- Reddit sets the terms for programmatic access in its Data API documentation, with rate limits and registration for apps.
What these routes have in common: an application or developer setup, someone who writes code, quotas to plan around, and weeks between the research question and the first row of data. In exchange you get scale (thousands of posts) and, on some programmes, data that is not visible to the public. If your project is a funded, multi-year study of platform-wide behaviour, apply.
If your project is "the comments under these 40 campaign posts" or "every reply on the three most viewed videos about this topic", the approval cycle costs more than the collection.
The no-code route: one link in, one file out
ZocialComment is the tool we built for the second case. You paste the link of a public post, the exporter reads its comment section the way a logged-out visitor sees it, and you download the result as CSV, Excel or JSON. There is no application, no code and no developer account, and the first 100 comments of any post export with no signup.
Why we recommend it for most research projects:
- One format for six platforms. TikTok, Instagram, Facebook, YouTube, Reddit and Threads (and Lemon8 and Google reviews) all start with the same columns:
author,username,text,likes,replies,created_at,language,is_pinned,id,reply_to_id,avatar_url. Platform-specific fields follow after them, so nothing is thrown away. - Replies as rows. Each reply is its own row with
reply_to_idpointing at the parent, so threads can be rebuilt in a spreadsheet or in code. - Many posts at once. With a pass you paste up to 20 links in one go, on any platform page.
- Analysis-ready files. The CSV opens correctly in Excel, accents and all, and the JSON keeps the full raw record for anyone who wants every field.
- AI analysis when you want a first read. Pass holders can run a sentiment and theme summary on an export. Treat it as a pilot read to shape your codebook, not as your coding.
Pick the platform page for your data: TikTok, Instagram, Facebook, YouTube, Reddit or Threads.
Choosing between the two
| Your project | Better fit | Why |
|---|---|---|
| Thesis or class project, 10–200 posts | No-code export | Data the same day; no approval wait. |
| Brand or market research on named posts or campaigns | No-code export | You already know which posts matter; you need their full comment sections. |
| Cross-platform comparison | No-code export | One column layout across platforms saves the reformatting step. |
| Platform-wide sampling of millions of posts | Official researcher API | Search and sampling across the whole platform is what those programmes are built for. |
| Non-public data (reach, demographics) | Official researcher programme | Only the platform can grant it; a public export only sees what visitors see. |
Many projects use both: the API or a search tool to find the posts, then a full comment export of the posts that made the cut.
Where this guide comes from: what running the exporter shows
This guide comes from operating the exporter, not from reading about it. A few things we see every week that change how a research dataset should be read:
- The comment count on a post is not the number of comments you can collect. Platforms count comments that moderation later hid, filtered or deleted. On one Instagram post we checked against several independent methods, the badge said 1,189 and only 165 comments were visible to anyone; on a Facebook giveaway post the badge said 784 and 202 were visible. Report rows collected, and say why they differ from the badge.
- Empty text is common on TikTok. On a TikTok post with 1,340 comments we looked at this week, 571 rows had no words at all: 519 were stickers or GIFs and 52 were photos. The export marks those rows with
image_typeand keeps the picture link inimage_url. Those links are signed by TikTok and expire after roughly 30 days, so download any images you need right away. - A post with zero comments is often not empty. When we audited Instagram exports that came back with zero rows, about a third turned out not to be empty at all; a common cause is a giveaway post whose comments were switched off after the draw. Note the collection date; the same link can give a different answer a week later.
- Timestamps differ by platform. Most platforms give an exact time; YouTube only shows "3 months ago", so its
created_atis an estimate. Plan your time-based analysis around the least precise platform in your sample.
The column names, limits and platform list in this guide are the live product values at the time of writing. For each platform's own rules on research use, its official documentation linked above is the authority.
From six files to one dataset
- Add two columns to every file before merging:
platformandpost_url. Once stacked, you will want to group by both. - Stack on the shared columns. Keep the eleven shared columns for the merged sheet and leave platform-specific columns in the original files.
- Remove duplicates by
platform+id. If you exported a post twice to catch late comments, the second file contains the first. - Normalise time. Convert
created_atto UTC, or to the time zone of the event you study, in one step. - Flag rather than delete. Add flags for empty text, emoji-only, link-only and very short comments. Whether to exclude them is a methods decision you should be able to report, not a cleanup step that disappears.
- Separate top-level comments and replies with
reply_to_id. Many coding schemes treat them differently, and replies can outnumber top-level comments on heated posts.
For qualitative coding, draw a random pilot sample of 100–200 rows, build the codebook on it, have a second coder code a share of rows, and check agreement before coding the rest. Stratify the sample by platform and by post so one viral post does not dominate the result.
Ethics, consent and anonymity
Public does not mean anonymous. A commenter posted to their followers, not to your paper. The AoIR Internet Research Ethical Guidelines 3.0 are the reference most ethics boards accept for this kind of work, and they ask you to weigh possible harm to the people you quote. In practice:
- Replace
usernameandauthorwith study codes, and deleteavatar_urlfrom any shared or published dataset. - Paraphrase quotes that could be pasted into a search engine to find the author, especially on sensitive topics.
- Collect only the columns your question needs, store files where your institution says, and delete them when the project ends.
- If any commenters are in the EU, the GDPR still applies; Article 89 allows processing for research but expects safeguards such as pseudonymisation.
Exporting does not notify the account owner or the commenters, which is exactly why the responsibility for handling the data well sits with you.
A collection log reviewers will accept
Keep one row per post in a separate sheet while you export: post URL, platform, account, date and time collected, comments shown on the post, rows in the file, replies included or not, and any notes ("comments switched off", "re-exported a week later"). It takes ten seconds per post and answers the first three questions a reviewer asks about your data. It also lets you or anyone else repeat the collection.
The first 100 comments of any public post export with no signup. Bigger posts and batches of up to 20 links are a one-time payment, never a subscription; see pricing for the details.
Export comments for your study →
Related reading: exporting YouTube comments for research, exporting Reddit comments for market research, analysing TikTok comments.
