4.7 million thank-yous

Next to every edit in a Wikipedia page's history there is a small link that says thank. Click it and the person who made the edit gets a notification: someone noticed. That's the entire feature. No message, no score, no badge. It launched on English Wikipedia on 30 May 2013 and it has not changed much since.

What almost nobody using it thinks about is that every press is written to a public log — who thanked whom, and when — and that the log is kept forever. It doesn't record which edit earned the thanks. Just the gesture, the two names, and the second it happened.

That makes it something rare: a complete, timestamped record of one kind of kindness inside one community, for thirteen years. We downloaded all of it.

Thanks sent

4.7M

May 2013 – Sep 2026

Accounts involved

523.3K

gave or received at least one

Relationships

2.5M

distinct thanker → thanked pairs

A steady hum

The first surprise is how flat it is. After a fast first year the button settled at around 30,000 presses a month and has stayed within shouting distance of that for a decade. The one visible swell is spring 2020, which lines up with the first pandemic lockdowns; the busiest single day in the log is 8 May 2020, with 1,865 thank-yous.

Source: English Wikipedia public logging dump, thanks events. EdgeStories analysis.

View as table

Month	Thanks
2013-06	9,403
2013-07	9,232
2013-08	9,566
2013-09	9,212
2013-10	13,427
2013-11	14,006
2013-12	15,879
2014-01	18,607
2014-02	20,276
2014-03	21,159
2014-04	19,977
2014-05	21,545
2014-06	20,477
2014-07	22,560
2014-08	22,914
2014-09	20,685
2014-10	23,053
2014-11	22,384
2014-12	24,363
2015-01	28,009
2015-02	27,341
2015-03	31,305
2015-04	27,919
2015-05	26,925
2015-06	24,861
2015-07	28,137
2015-08	29,411
2015-09	27,061
2015-10	28,027
2015-11	28,696
2015-12	27,358
2016-01	30,033
2016-02	26,677
2016-03	28,276
2016-04	27,114
2016-05	25,707
2016-06	24,081
2016-07	24,920
2016-08	26,262
2016-09	23,942
2016-10	24,988
2016-11	24,997
2016-12	27,357
2017-01	28,924
2017-02	26,053
2017-03	28,456
2017-04	27,456
2017-05	27,794
2017-06	26,201
2017-07	26,368
2017-08	28,165
2017-09	27,119
2017-10	27,907
2017-11	27,837
2017-12	29,209
2018-01	32,274
2018-02	27,705
2018-03	30,438
2018-04	30,591
2018-05	31,266
2018-06	30,703
2018-07	28,937
2018-08	30,727
2018-09	28,074
2018-10	29,726
2018-11	30,067
2018-12	31,226
2019-01	32,175
2019-02	29,352
2019-03	31,199
2019-04	28,952
2019-05	29,974
2019-06	30,651
2019-07	31,216
2019-08	29,304
2019-09	27,945
2019-10	30,369
2019-11	30,195
2019-12	29,895
2020-01	35,121
2020-02	30,784
2020-03	33,679
2020-04	38,389
2020-05	39,463
2020-06	36,934
2020-07	37,627
2020-08	37,265
2020-09	33,651
2020-10	35,255
2020-11	34,843
2020-12	37,286
2021-01	40,981
2021-02	36,272
2021-03	38,826
2021-04	36,191
2021-05	35,844
2021-06	32,800
2021-07	33,760
2021-08	32,890
2021-09	30,294
2021-10	30,627
2021-11	31,991
2021-12	32,460
2022-01	35,169
2022-02	31,247
2022-03	33,715
2022-04	31,892
2022-05	33,112
2022-06	30,070
2022-07	29,760
2022-08	29,847
2022-09	30,682
2022-10	31,057
2022-11	27,886
2022-12	28,301
2023-01	35,087
2023-02	32,081
2023-03	34,066
2023-04	32,723
2023-05	31,544
2023-06	30,586
2023-07	32,906
2023-08	32,633
2023-09	30,278
2023-10	32,390
2023-11	32,649
2023-12	34,942
2024-01	38,081
2024-02	32,179
2024-03	33,421
2024-04	30,997
2024-05	31,766
2024-06	29,473
2024-07	30,092
2024-08	30,804
2024-09	30,879
2024-10	32,128
2024-11	33,893
2024-12	34,846
2025-01	36,293
2025-02	30,369
2025-03	32,723
2025-04	31,433
2025-05	33,447
2025-06	31,656
2025-07	32,822
2025-08	32,003
2025-09	32,183
2025-10	34,002
2025-11	35,503
2025-12	34,571
2026-01	37,447
2026-02	31,814
2026-03	34,251
2026-04	34,102
2026-05	34,973
2026-06	32,270
2026-07	32,896
2026-08	34,077

In April 2018 the feature was widened so that log actions — a page move, a deletion, a protection — could be thanked, not just edits. You can find the date on the chart if you're told where to look. You would not find it otherwise.

The number of people behind the line is just as steady. In a typical year about 34,000 accounts send at least one thank-you, and about 42,000 receive one. For scale: Wikipedia's own statistics counted 263,530 accounts as active in the 30 days before we checked. Thanking is a minority habit, and it is not spreading. Each year around 19,000 accounts send their first thank-you, down from a peak of 27,600 in 2020 — just enough newcomers to replace the people who stop.

The gratitude 1%

Nothing on the internet is evenly distributed, and neither is this. But the scale of it is worth sitting with.

291,114 accounts have sent at least one. Gini coefficient 0.90; for thanks received it is 0.89. Source: English Wikipedia public logging dump, thanks events. EdgeStories analysis.

View as table

Senders	Accounts	Share of all thanks sent
Top 1%	2,911	64%
Next 9%	26,200	25.2%
Other 90%	262,003	10.8%

Fewer than 3,000 accounts — the top 1% of senders — are responsible for 64% of all the thanks ever sent on English Wikipedia. At the other end, 57% of everyone who has ever pressed the button pressed it exactly once and never again. The most prolific single account has sent 43,911, which is one every two hours and forty minutes, around the clock, for thirteen years.

We expected receiving to be flatter than giving — a few generous people spraying thanks across a wide population. It isn't. The top 1% of recipients collect 63% of all thanks, almost exactly mirroring the senders, and 62% of recipients have been thanked exactly once. Gratitude on Wikipedia is not a broadcast from the few to the many. It is mostly a conversation among the few.

Thank you. No, thank you.

How much of a conversation? Take every ordered pair where one account has thanked another — 2.5 million of them — and ask whether the thanks ever came back.

Reciprocated

14.3%

of thanker → thanked pairs are thanked back, ever

By chance

2.8%

expected if the same people thanked at random

Fast

22%

of first thank-backs arrive within a day

One in seven does. That sounds low until you ask what it would be if the same people handed out the same number of thanks to the same number of recipients, but chose them without regard to who had thanked them. We shuffled the network that way, keeping everyone's totals fixed. By chance, 2.8% of pairs would be mutual. The real figure is five times that.

When a thank-you is returned, it's often immediate: one in ten thank-backs arrives within the hour, and 22% within a day. Then there's a very long tail — the median return takes 72 days — which is less a reply than two people who keep running into each other.

And some pairs run into each other constantly. Three-quarters of pairs consist of a single thank-you, ever. But a quarter of all thanks flow inside pairs that have exchanged ten or more, and the most devoted sender in the log has thanked the same person 5,943 times. Even among the 31,000 pairs who thank each other regularly, the exchange is lopsided: in the typical one, the quieter partner sends 29 for every 100 they get. Only a third are anywhere near even.

We're not naming any of these people, though Wikipedia's logs would let us. Our rule is that volunteers are reported in aggregate only, even when the data flatters them.

The going rate

We went into this expecting to write that gratitude and contribution are different things — that the people who get thanked are not the people who do the most work. It's a nice idea. The data doesn't support it.

Edit counts are lifetime totals as of September 2026 and include edits made before the button existed. Accounts under 100 edits are left out: they are in this data only because they sent or received a thank-you, which inflates their rate. Source: English Wikipedia public logging dump, thanks events. EdgeStories analysis.

View as table

Lifetime edits	Accounts	Thanks per 1k edits	Share of all thanks received
1–9	140,854	163.5	1.8%
10–99	195,808	20.7	3.3%
100–999	115,282	7.2	5.8%
1k–10k	39,200	5.6	14.8%
10k–100k	10,847	6.5	41.7%
100k+	1,111	5.0	32.2%

Across three orders of magnitude of effort, the rate barely moves. An editor with a few hundred edits and an editor with a few hundred thousand both collect somewhere between five and seven thank-yous per thousand edits — call it one for every 170 saves. Thanks aren't a prize that flows to a charismatic few. They behave more like a flat tax on work, levied at the same rate on everyone.

Which means the concentration in the last section is mostly the concentration of Wikipedia itself. The 12,000 accounts with more than 10,000 edits receive 74% of all thanks — and they made 78% of the edits by accounts in this data. The correlation between an account's edit count and its thanks is strong but far from perfect (a rank correlation of 0.48), so individuals vary a great deal. The averages don't.

Thanking up

Who thanks whom? We looked up the registration date of every account in the log and measured how old each account was at the moment of each thank-you.

4.7M thanks where both accounts' registration dates are known. Accounts older than the registration log (pre-2006) are counted as registered in September 2005. Source: English Wikipedia public logging dump, thanks events. EdgeStories analysis.

View as table

Thanker account age	< 1 month	1–12 months	1–5 years	5–10 years	10+ years
< 1 month	7.3%	10%	21.5%	24.4%	36.8%
1–12 months	4.3%	10.6%	23.3%	25.7%	36.2%
1–5 years	3.5%	8.2%	24.3%	26.8%	37.2%
5–10 years	2.8%	6.9%	21.7%	33.4%	35.2%
10+ years	2.7%	6.2%	17.9%	20.2%	52.9%

Every row peaks in the same column. Whatever your own tenure, the largest share of your thanks goes to accounts more than a decade old — they do much of the work, so by the going rate they collect much of the gratitude. Veterans are the most inward-looking: 53% of thanks sent by ten-year accounts go to other ten-year accounts.

The more interesting row is the first. Accounts less than a month old have sent 262,000 thank-yous and received 157,000. Newcomers give 1.7 times as much gratitude as they get — and only 7% of what they give goes to other newcomers. It's tempting to picture the button as veterans patting newcomers on the head. In practice it runs at least as much in the other direction.

Overall, 55% of thanks go to an account older than the sender's. The tilt was strongest at launch — 61% in 2013 — and has sat around 54–56% since 2015. One plausible reason, which we haven't tested: the population is ageing together, and the gap between a nine-year account and an eleven-year one means less than the gap between a newcomer and anyone.

Does it do anything?

This is the question everyone asks, and it's the one a log can't answer. We can see that 284,000 accounts first met the feature by being thanked, that 18% of them went on to thank someone else, and that 7% did so within a month. We can't see what they would have done otherwise.

Fortunately someone ran the experiment. In 2019, J. Nathan Matias and colleagues had 344 volunteers thank a random half of about 15,500 editors on the Arabic, German, Polish and Persian Wikipedias. Being thanked raised two-week retention by two percentage points and increased the thanks those editors sent to others by 43%. It had no detectable effect on how many hours they put in. An earlier observational study by Goel, Anderson and Zia found the same upward flow of thanks we see here, in other languages; English was excluded from their data.

So: a small, real effect on whether people stay, and a larger one on whether they pass it on. That is about what you'd hope for from a link that costs one click. It also puts the numbers above in a slightly melancholy light. A gesture that measurably keeps volunteers around is used by about 34,000 people a year, most of them once, on a site where a quarter of a million accounts were active last month.

Methods and limits

Source. The English Wikipedia pages-logging dump dated 4 September 2026, streamed and filtered to thanks events: 4,711,880 of them, from 30 May 2013 to the dump date. We used the dump rather than the API because Wikimedia asks bulk readers to, and because a dated file can be re-checked. As a test we counted one arbitrary day (5 March 2024) through the live API and in our extract; both gave 1,108. A further 577 events couldn't be parsed — typically because the sender's name has been hidden — and are excluded.

Accounts. Registration dates, lifetime edit counts and bot flags for all 523,299 accounts came from the public users API in September 2026. 543 accounts (0.1%) no longer resolve — renamed or otherwise gone — and drop out of the tenure and edit-count analyses only. 161 bot accounts are excluded from the edit-count analysis. Accounts that predate the registration log are treated as registered in September 2005, which understates the age of Wikipedia's oldest editors.

What the log doesn't say. It records that a thank-you happened, not what for. We can't tell you which kinds of edit earn gratitude, and anyone who claims to from this data is guessing. Unregistered (IP) editors can't be thanked, so they aren't here either.

Edit counts are lifetime totals, including edits from before 2013, measured today rather than at the time of each thank-you. That blurs the rate chart; it's why we describe the rate as flat rather than quoting differences between bars. Accounts with fewer than 100 edits appear in this data only if they were thanked or thanked someone, which inflates their rate, so the chart leaves them out. The table view includes them.

The chance baseline for reciprocity is a degree-preserving shuffle: each sender keeps their number of distinct recipients and each recipient their number of distinct senders, but who points at whom is randomised, with collisions repaired by swapping so that no pair is duplicated and nobody thanks themselves. The 2.8% figure is the mean of five shuffles.

This is one wiki. Everything here is English Wikipedia. Prior work found real differences between language editions, and nothing above should be read as "Wikipedia" in general.

Names. Wikipedia's logs are public and the community once kept its own leaderboard of thankers, frozen since 2019. We still publish no usernames and no identifiable pairs. Volunteer editors are not public figures, and our rules don't have an exception for flattering data.