05: Read any account like an analyst

Time: 15 minutes You need: XActions installed. No X account, no login, no API key. You end up with: a defensible read on how any account actually performs, and how two accounts differ.

Everything in this tutorial runs on the guest tier. You never log in.

The question

"How does this account actually do?" is usually answered with follower count, which is the least useful number available. Follower count tells you about the past. What you want to know is how often they post, what they post, how much of their audience reacts, and when.

XActions answers all of that from public data.

1. One account

xactions analyze NASA

By default it samples the last 50 posts. The report is dense, so here is what each block is telling you:

  Audience
  Followers:            92.4M  117 following
  Follower ratio:       789.4K
  Growth:               13.5K/day  lifetime average over 6,826 days

Follower ratio is followers divided by following. A high ratio means an account people seek out; a ratio near 1 usually means follow-for-follow growth. Growth is a lifetime average, so it flatters accounts that grew fast years ago. Treat it as an upper bound, not a current rate.

  Output
  Posts, lifetime:      74.2K  10.87/day
  Posts, sampled:       2.49/day  50 posts over 20.1 days
  Typical gap:          1.99h
  Last post:            7h ago

The two rates are the interesting part. Lifetime says 10.87/day; the recent sample says 2.49/day. This account posts a quarter as often now as it did on average across its life. One number would have hidden that entirely.

  Engagement
  Rate:                 0.007%  median original post, as a share of followers
  Median per post:      6,466.5  ♥ 5,426.5  ↻ 813  💬 176.5
  Best in sample:       20.9K
  Median views:         1.2M  0.44% of viewers interact, n=20

(n=20 is how many sampled posts reported a view count. A small n there means the view-rate figure is directional, not precise.)

Median, not mean. One viral post drags a mean anywhere; the median tells you what a typical post does. The engagement rate looks tiny because it is measured against 92 million followers, which is why the views-based number underneath it matters more: 0.44% of the people who actually saw the post interacted with it.

  Content mix
  Original:             36%  replies 0%  reposts 60%  quotes 4%
  With media:           80%  links 75%  hashtags 0%

Three fifths of this account's output is reposts, and zero percent is replies. That is a curation feed, not a conversation. If you are studying it to copy the strategy, that distinction matters more than any engagement number.

  Timing (UTC)
  00              ▅▅▂█▅▂▆▁▆ 23
  Best hour:            17:00  median 8,692
  Best weekday:         Thursday  median 8,377

The sparkline is engagement by hour of day, UTC, across the sample. Sample more posts for a more trustworthy peak:

xactions analyze NASA --limit 200

2. Two accounts, side by side

analyze takes several usernames and compares them:

xactions analyze NASA SpaceX

Read the comparison in this order:

Posting cadence. Who ships more? A 5x cadence difference explains most engagement differences on its own.
Content mix. Original versus repost versus reply. These are different businesses.
Engagement per view. The only figure that is fair across wildly different follower counts.
Best hour. If two accounts serve the same audience and peak at different hours, one of them is wrong.

3. Whose followers overlap

xactions audience NASA SpaceX --max 2000

This is the one command on this page that needs a session, because follower lists are not on the guest tier. Run xactions connect first if xactions doctor says you are guest-only.

Overlap answers a question analysis of a single account cannot: is this a shared audience or two separate ones? High overlap means you are choosing between two accounts for the same people. Low overlap means there is an audience you are not reaching.

--max caps how many followers are fetched per account. Start at 2000. Raising it improves the estimate and costs time linearly.

4. Keep the raw numbers

Every command above takes --output, and the format follows the extension:

xactions analyze NASA --limit 200 --output nasa.json
xactions analyze NASA --limit 200 --output nasa.csv
xactions analyze NASA --limit 200 --output nasa.xlsx

Or take JSON on stdout and cut it yourself:

xactions analyze NASA --json | jq '{
  followers:   .audience.followers,
  ratio:       .audience.followerRatio,
  postsPerDay: .output.postsPerDay,
  lifetime:    .output.lifetimePostsPerDay,
  median:      .engagement.medianPerOriginal,
  viewRate:    .engagement.viewRate,
  originals:   .mix.originalShare,
  bestHour:    .timing.bestHourUTC,
  bestWeekday: .timing.bestWeekday
}'

{
  "followers": 92356555,
  "ratio": 789372.26,
  "postsPerDay": 2.49,
  "lifetime": 10.87,
  "median": 6466.5,
  "viewRate": 0.44,
  "originals": 36,
  "bestHour": 17,
  "bestWeekday": "Thursday"
}

The top-level keys are identity, audience, output, engagement, mix, timing, topPosts, topHashtags, topMentions, meta, and signals. Run xactions analyze NASA --json | jq 'keys' to see them, and jq '.engagement' to open any one up.

signals is the observations block from the formatted report, as data:

xactions analyze NASA --json | jq -r '.signals[] | "\(.level | ascii_upcase): \(.title)"'

GOOD: Followed far more than it follows
WATCH: Engagement rate of 0.007%
WATCH: 60% of the timeline is retweets
GOOD: 80% of posts carry media
INFO: Best hour is 17:00 UTC
GOOD: Averaging 13.5K new followers a day

timing.byHourUTC is the 24-entry array behind the sparkline, so you can find the peak yourself over a large sample:

xactions analyze NASA --json \
  | jq -r '.timing.byHourUTC | max_by(.medianEngagement) | "best hour \(.index):00 UTC, median \(.medianEngagement)"'

best hour 17:00 UTC, median 8692

timing.bestHourUTC is null when no single hour cleared timing.minimumBucketSample posts, and the formatted report says "not enough data" rather than inventing a peak. Raise --limit until it resolves; the sparkline is readable either way.

5. One line per record, for agents and pipes

--json is the right shape for jq. It is the wrong shape for an LLM, which pays for every brace. --compact prints one record per line as tab-separated key=value pairs, with no colours and no spinner:

xactions profile NASA --compact

id=11348282	username=NASA	name=NASA	followers=92356563	following=117	tweets=74197	verified=false	bio=Making the seemingly impossible, possible. ✨

--fields narrows it to the columns you actually want:

xactions tweets NASA --limit 3 --compact --fields id,likes,text

id=2093063226351136958	likes=1639	text=Join us tomorrow at 11am ET (1500 UTC) as the Artemis II crew receives the Congressional Space Medal of Honor! ...
id=2092962731666051534	likes=0	text=RT @LearnWithNASA: Space telescopes help us understand the origins of the universe. ...
id=2092744659667673582	likes=6753	text=A partial lunar eclipse will pass over the Americas ...

Both are global flags, so they work on any command that emits records. One line per record also means cut, awk, and grep work without a JSON parser in the pipeline.

6. Track it over time

A single report is a snapshot. The interesting signal is the change:

# Take a reading every day, keyed by date
xactions analyze NASA --limit 200 --output "reports/nasa-$(date -u +%F).json"

Put that in cron and after two weeks you have a series. XActions also has a built-in version of this:

xactions snapshot NASA          # start auto-snapshotting
xactions history NASA           # read the series back

7. A comparison script

#!/usr/bin/env bash
# compare.sh handle1 handle2 ... — one row per account, sorted by engagement.
set -euo pipefail

printf '%-20s %12s %10s %12s %8s\n' ACCOUNT FOLLOWERS POSTS/DAY MED_ENGAGE BEST_HR

for handle in "$@"; do
  xactions analyze "$handle" --limit 100 --json 2>/dev/null | jq -r --arg h "$handle" '
    [$h,
     (.audience.followers          // 0 | tostring),
     (.output.postsPerDay          // 0 | tostring),
     (.engagement.medianPerOriginal // 0 | tostring),
     (.timing.bestHourUTC          // "?" | tostring)]
    | @tsv'
done | sort -k4 -rn | awk -F'\t' '{ printf "%-20s %12s %10s %12s %8s\n", $1, $2, $3, $4, $5 }'

chmod +x compare.sh
./compare.sh NASA SpaceX

Run it against your own account and the five accounts you compete with. The row that surprises you is the one worth investigating.

What you learned

Why median beats mean, and why engagement-per-view beats engagement-per-follower
That lifetime and recent cadence disagreeing is itself the finding
How content mix identifies what kind of account you are actually looking at
Turning reports into files, JSON, and a time series you can track
--compact and --fields when the consumer is an agent or a shell, not you

06: Everything is JSON to script all of this properly
04: Build a brand monitor to watch continuously instead of on demand
CLI reference for every flag on every command

05: Read any account like an analyst

The question

1. One account

2. Two accounts, side by side

3. Whose followers overlap

4. Keep the raw numbers

5. One line per record, for agents and pipes

6. Track it over time

7. A comparison script

What you learned

Next

⚡ Free and open source