Which Grok Bot posts on X do people bookmark, and which do they only scroll past? We looked at 3,826 posts from 11 August 2026 to 16 October 2026 and ranked the 557 with enough reach by bookmark rate.
Posts collected
3,826
Ranked
557
Bookmarks on ranked posts
218,384
Elon Musk reposts
48
Last updated 18 October 2026 · data fetched 16 October 2026 · method and limits · PDF
Key findings
011.45% vs 0.24%
Copyable posts get saved about 6 times as often as news.
Median bookmark rate for prompts and templates is 1.45% (n = 24); for news and opinion it is 0.24% (n = 254). Courses, guides and repo lists sit in between.
19 of the top 50 mention a prompt, template or copying in their first 200 characters, against 13% of the other 507. Questions (1 of 50) and all-caps openers (0) barely appear.
23 of the 48 reposts we found show a concrete task a bot did. Median time from post to repost: 4.7 hours. 17 of the original posters had fewer than 50,000 followers.
59 posts mention money, tokens or usage in their first 200 characters; 21 of those are savings or revenue claims. None says what a specific task cost to run and how that was measured.
Posts you can reuse get saved. News gets read and left.
Bookmark rate is bookmarks divided by views. We compare medians, because a few viral posts would pull an average anywhere. The line around each median is its 95% interval: if two lines overlap, treat the groups as close.
Median bookmark rate by post type557 ranked posts. Type labelled from the post text.
Prompt or templaten = 24 · small group1.45%
Course or videon = 771.12%
Guide or tipsn = 931.02%
Repo or tool listn = 28 · small group0.90%
Showcasen = 810.68%
News or opinionn = 2540.24%
0%0.5%1%1.5%2%2.5%
Median 95% interval of the median All 557 ranked posts, 0.67%
The 120 posts that only came in through the 15 creator timelines do not change the order. Without them, prompts and templates have a median of 1.29% (n = 13) and news and opinion 0.20% (n = 218).
Table
Group
Posts
Median rate
95% interval
Bookmarks per like
In top 50
Prompt or template
24
1.45%
1.10% to 2.05%
1.78
10
Course or video
77
1.12%
0.93% to 1.21%
1.62
10
Guide or tips
93
1.02%
0.90% to 1.26%
1.56
14
Repo or tool list
28
0.90%
0.53% to 1.39%
1.45
4
Showcase
81
0.68%
0.47% to 0.88%
0.86
4
News or opinion
254
0.24%
0.19% to 0.30%
0.45
8
Source: X public metrics for 557 ranked posts (of 3,826 collected), posted 11 August 2026 to 9 October 2026, fetched 16 October 2026. Agent Receipts analysis.
Bookmarks per like tell the same story. For prompts and templates the median post gets 1.78 bookmarks per like; for news and opinion 0.45. People like the news and save the setup.
Prompts and templates are a small group (n = 24), so their interval is wide: 1.10% to 2.05%. Even its lower end is above the upper end for news (0.30%).
Median bookmark rate by category557 ranked posts. Category labelled from the post text.
Engineeringn = 1711.05%
Businessn = 1200.94%
Contentn = 300.52%
Personaln = 20 · small group0.39%
Generaln = 2160.22%
0%0.5%1%1.5%
Median 95% interval of the median All 557 ranked posts, 0.67%
Table
Group
Posts
Median rate
95% interval
Bookmarks per like
In top 50
Engineering
171
1.05%
0.92% to 1.16%
1.48
17
Business
120
0.94%
0.78% to 1.07%
1.37
12
Content
30
0.52%
0.33% to 1.16%
1.33
4
Personal
20
0.39%
0.08% to 0.74%
0.66
0
General
216
0.22%
0.18% to 0.26%
0.41
17
Source: X public metrics for 557 ranked posts (of 3,826 collected), posted 11 August 2026 to 9 October 2026, fetched 16 October 2026. Agent Receipts analysis.
What does not get saved: general news about Grok Bot (the largest group, n = 216) and personal use. Posts about personal life admin have a median of 0.39%, and the group is small, so read that number with care.
02 · Saves and reach
The most viewed posts are not the most saved.
Across the 557 ranked posts, the rank correlation between views and bookmark rate is 0.06. In plain terms: knowing how often a post was seen tells you almost nothing about how often it was saved.
Bookmark rate against views557 ranked posts. Each dot is a post; the dark line is the median rate in each view band.
Ranked post One of the 10 most viewed Median per view band
Source: X public metrics for 557 ranked posts (of 3,826 collected), posted 11 August 2026 to 9 October 2026, fetched 16 October 2026. Agent Receipts analysis.
Rate per view band
Median bookmark rate per view band
Views
Posts
Median rate
95% interval
5k–10k
155
0.51%
0.32% to 0.72%
10k–20k
142
0.76%
0.49% to 0.93%
20k–50k
142
0.89%
0.64% to 1.12%
50k–100k
54
0.78%
0.38% to 1.09%
100k–250k
33
0.86%
0.41% to 1.18%
250k or more
31
0.16%
0.05% to 0.29%
The median rate peaks at 0.89% for posts with 20k–50k views. Above 250k it drops to 0.16% (n = 31). Those 31 posts have 83% of all views of the ranked posts and 36% of the bookmarks.
7 of the 10 are news or opinion, from 7 accounts; @elonmusk alone has 4. None of them is in the top 50.
03 · Patterns in the top 50
What the top 50 have in common
The top 50 by Wilson lower bound, compared with ranks 51 to 557. Text features are keyword matches on the first 200 characters of each post, the part we stored.
Types
Post types in the top 50
Type
In top 50
Of that type
Of the top 50
Prompt or template
10 of 24
42%
20%
Guide or tips
14 of 93
15%
28%
Repo or tool list
4 of 28
14%
8%
Course or video
10 of 77
13%
20%
Showcase
4 of 81
5%
8%
News or opinion
8 of 254
3%
16%
10 of the 24 prompt and template posts made the top 50. Of 254 news and opinion posts, 8 did.
Text features
Text features in the top 50 and the rest
In the first 200 characters
Top 50
Ranks 51–557
Median rate with
Without
Mentions a prompt, template, copy, steal or paste
19 (38%)
65 (13%)
1.29% (n = 84)
0.53%
Numbered steps, arrows or bullets
10 (20%)
73 (14%)
1.16% (n = 83)
0.54%
Points to official material (SpaceXAI, a workshop, a session, engineers)
20 (40%)
159 (31%)
1.07% (n = 179)
0.41%
Names a dollar amount
2 (4%)
25 (5%)
0.86% (n = 27)
0.65%
Starts with "I" or "My"
5 (10%)
45 (9%)
0.50% (n = 50)
0.71%
Tags another account (not @bot or @grok)
3 (6%)
21 (4%)
0.22% (n = 24)
0.72%
Asks a question
1 (2%)
15 (3%)
0.20% (n = 16)
0.70%
Asks readers to bookmark or save it
3 (6%)
5 (1%)
1.13% (n = 8)
0.65%
Opens in capitals
0 (0%)
22 (4%)
0.33% (n = 22)
0.68%
What works: pointing to something reusable. 19 of the top 50 mention a prompt, template or copying, against 13% of the rest. What does not: first-person stories (median rate 0.50%), tagging other accounts (0.22%) and questions (0.20%), all below the 0.67% median of all ranked posts. These are associations in our sample, not causes.
Creators
The top 50 come from 27 creators. Three of them hold 19 places: @unicodef1wn (8), @0xRafy (6), @1920web1080 (5). Read this with care: 25 of the top 50 were found through the timelines of the 15 creators we collected separately, so those creators are over-represented. Median views in the top 50 are 21,160, against 17,651 for the rest: a top post does not need a big audience.
04 · What Elon Musk reposted
48 reposts: use cases with proof, picked up within hours
A separate sample: 48 Grok Bot posts by other accounts that @elonmusk reposted (28) or quoted (20) between 12 August 2026 and 16 October 2026; 26 of them in October. Search returns at most 25 results per query, so the real number is higher.
What the 48 reposted posts are aboutOne topic per post, labelled by hand from the first 200 characters.
A concrete task the bot did23 of 48
Opinion, praise or hype11 of 48
Product update6 of 48
Event or livestream4 of 48
Guide, template or directory4 of 48
Source: search on X for reposts and quotes by @elonmusk of Grok Bot posts by others, 12 August 2026 to 16 October 2026. Agent Receipts labels.
Median time to repost
4.7 hours
31 within 6 hours, 47 within 24.
Posters under 50,000 followers
17 of 48
5 under 10,000; the smallest had 2,172. Median: 127,056.
Median bookmark rate
0.05%
Against 0.67% for our ranked posts. A repost brings views (median 690,048), not saves.
About cost
0 of 48
None reports what a task cost. Some mention a price or a revenue figure.
The common shape is a short first-person post about one everyday task, with a screenshot or video as proof: an inbox cleaned up, a dentist booked, a contract checked, a phishing mail caught. Next come opinion, praise or hype (11) and product update (6). Accounts that appear more than once: @DirtyTesLa, @XFreeze, @bot, @icreatelife.
Only 3 of these 48 posts are in our collection of 3,826, and 2 in the ranked set, even though each had at least 158,326 views. Our search sample misses many widely seen posts. That is a limit of the main dataset, see method.
05 · The gap
Nobody posts what a task costs
59 of the 557 ranked posts mention money, tokens or usage. We read all of them. 0 report a measured cost per task: the tokens or usage one run of a defined task took, and how that was measured.
What the 59 cost mentions are aboutRanked posts whose first 200 characters mention cost, price, tokens, credits, usage or a dollar amount. Labelled by hand.
Savings or revenue claim21 of 59
Other meaning ("spent 9 hours")20 of 59
Plan price, credits or limits12 of 59
Tips to use fewer tokens6 of 59
Source: X public metrics for 557 ranked posts (of 3,826 collected), posted 11 August 2026 to 9 October 2026, fetched 16 October 2026. Agent Receipts analysis.
Savings and revenue claims are the biggest group: a stack of bots that "replaces" a salary or a SaaS bill, or a side business with a monthly figure. They are not measurements, and we could not check them. Plan prices, credits and usage limits come next, followed by tips to burn fewer tokens. The tips are popular, which suggests people do run into limits.
This is the gap Agent Receipts is built for. We are measuring 40 fixed everyday tasks with tokens, pass or fail and share of the weekly usage pool. The first round has not run yet, so there are no numbers to show. The costs page says what each result will include.
06 · Safety notes
27% of ranked posts carry a safety flag
A language model read each post and flagged three risks; a keyword list added a fourth, weaker signal. Linked prompts, repos and videos were not opened, so no flag is not the same as safe.
Safety flags by group
Posts
Any flag
Broad account access
Financial or scam risk
Asks for keys, passwords or logins
Keyword only
Top 25
6 (24%)
1
4
1
0
Top 50
10 (20%)
4
5
1
0
All 557 ranked
152 (27%)
92
37
7
16
All 3,826 collected
897 (23%)
434
320
34
109
Source: X public metrics for 557 ranked posts (of 3,826 collected), posted 11 August 2026 to 9 October 2026, fetched 16 October 2026. Flags from the post text only. Agent Receipts analysis.
Broad account access is the most common flag: 92 of 557 ranked posts suggest connecting a bot to email, socials or a computer with wide permissions. Financial or scam risk follows with 37: trading agents, tokens and income claims. In the top 50 the order flips: 5 financial, 4 broad access.
Before you copy a setup
Give a bot only the accounts and permissions the task needs. Try it on a test inbox or a spare account first.
Do not paste API keys, passwords or SSH keys into a chat or into a template from someone you do not know.
Treat trading agents, tokens and income claims as high risk. We do not test them and do not recommend them.
Read the linked prompt or repo yourself before running it. Our flags only cover the post text.
07 · The list
Top 25 of all time
Ranked by the 95% Wilson lower bound of bookmarks per view, which pulls posts with few views down a little. One line per post in our own words; open the link for the post itself. We do not copy prompts.
Once a week, the setups people save most. Cost measurements too, once they exist.
Free. We mail a confirmation link first. At most one mail a week; leave any time. Privacy.
08 · Method and limits
How we got these numbers
Collection: read-only search on X public metrics, fetched 16 October 2026. Per day since launch, the query "Grok Bot" OR grokbot OR "grok bots" plus topic variants, the timelines of 15 creators who post often about Grok Bot, and quote posts of 8 top posts. 3,826 posts from 2,623 accounts, posted 11 August 2026 to 16 October 2026.
This is a sample, not a full count. Search returns about 25 posts per page and we read the 2 most relevant pages per day. Even posts with a lot of reach can be missing: of the 48 posts Elon Musk reposted, only 3 are in our collection. Read the numbers as patterns in a large sample of Grok Bot posts, not as totals for all of X.
Ranked posts: at least 5,000 views, at least 7 days old, not a reply, on topic. That leaves 557 posts with 218,384 of the 249,769 bookmarks in the set.
Labels: category, type and safety flag come from a language model reading the post text, with a keyword list as an extra safety signal. 319 of the 557 ranked posts have a low-confidence label (1,899 of all 3,826). We did not relabel them by hand.
Text features use the first 200 characters of each post, which is all we stored. Longer posts are cut off, so a feature further down a post is missed.
The 15 creator timelines add posts that search alone would not find, so those creators are over-represented. Without the 120 timeline-only posts the type ranking stays the same.
Medians come with a 95% interval from order statistics, which needs no assumption about the distribution. Groups under 30 posts are marked as small.
The Elon Musk reposts are a separate search sample; the topic labels and the cost labels are ours, made by hand from the first 200 characters.
Views and bookmarks keep growing after collection. That drift changes the numbers more than the choice of ranking method does.
We publish handles, links, metrics and our own labels and summaries. No post text, no prompts.