HASHTRACKER

X and Twitter analytics 3 min read

How to export tweets and check the file before trusting it

Before you export tweets, know what a row should contain, why compliance is not a formality and the Excel problem that silently destroys nineteen digit IDs.

Cover drawn as a column chart

At some point every analysis outgrows the dashboard and you need to export tweets into a file, to do something nobody built a button for.

That is a normal requirement, and there is a right way to meet it.

Why the dataset matters more than the charts

A chart answers the question someone anticipated. A dataset answers the question you have. Cross-referencing posts against your own sales data, running a model over the text, building a chart nobody offers, or handing a reviewer something they can check: none of that happens inside a dashboard.

We deliver the report and the rows behind it, gathered within the Twitter compliance rules.

We deliver the report and the rows behind it

What a row should contain

One row per post, covering four things: the post itself, its engagement, its author and its relationships to other posts. That means text, timestamp, language and URL. Impressions, likes, reposts and replies. The account, its follower and following counts, its creation date and whether it is verified. And the IDs linking replies, quotes and reposts back to what they responded to.

With those four groups you can rebuild any figure in any report.

The compliance part, which is not optional

Public post data comes with terms attached, and they govern what can be stored, for how long and what has to happen when a post is deleted. Working inside those terms is not a formality. It is what makes a dataset usable for client work, for publication and for anything that might be looked at by the platform.

Everything we deliver is gathered Twitter compliant. If a provider cannot tell you plainly how they collect, that answer is the finding.

The Excel problem nobody warns you about

Post IDs are nineteen digits long. Open a CSV in Excel or load it carelessly into pandas and those IDs become floating point numbers. 1899862353047347394 turns into 1.8998623530473474e+18, and it no longer identifies anything. The link back to the original post is gone and you will not notice until you try to use it.

The fix is forcing those columns to text on import. It affects every ID column: the post, the author, and every reference to a quoted or replied-to post.

Sanity checks before you trust a file

Four checks catch almost everything. Row count against the total the report claims. Date range against the period you asked for. Whether reposts are in or out, because that changes every aggregate. And whether any ID column has been mangled into scientific notation.

Five minutes of checking saves a week of conclusions built on a broken file.

What to do with it once you have it

Spreadsheet for a quick split. Notebook for anything repeatable. And increasingly, feed the file to a language model and ask it what it sees in the text, which is a genuinely useful first pass on thousands of posts.

The point is that the analysis stops depending on whoever supplied the tool. Pick a plan and the file is yours.

Pick a plan

Choosing a subscription

Every subscription carries the same product. The only decision is how many tweets you need to analyse.

See the subscriptions