Search covers subjects, readings, releases, communities, measured statements, quotes as pages show them, the glossary and the method. It never searches posts: the posts behind each figure are its fixed sample, on its page.

Weekly release

Release 2026-W01

Posts published 29 Dec 2025 – 4 Jan 2026 on Reddit, Bluesky, YouTube, Hacker News, arXiv, Google News and Exa, each measured on its own: 97 readings.

Published 2 Oct 2026. Its figures never change. Cite it with its permanent link. What these states mean

March eraRebuilt from archives, with live capture from 7 Mar to 31 May.

Release date
2 Oct 2026 06:44:53Z
Next release
Wednesday 14 Jan 2026
Window
29 Dec 2025 – 4 Jan 2026
Method
005c-series-v3.1
Release calendarPermanent link /releases/2026-W01

This release’s conditions

As of
25 Sep 2026 19:30Z
Coverage computed
2 Oct 2026 02:23Z
Facts checked
2 Oct 2026 03:07:29Z
Posts without facts
0
Days settled
21 of 21 that can settle
Author coverage
posts with an identifiable platform identity; article sources have none by definition
Readings counted
97 readings, current: the release's stored readings that no notice names (none superseded, corrected or withdrawn); read live at page render
Spam, bot and incentive filter, not applied yet
Spam, bot and incentive posts haven’t been removed: their labels agree with a stronger model at kappa 0.43, below the threshold of 0.60
This release’s conditionsAs of 25 Sep 2026 19:30Z · 21 of 21 days settled · facts checked, 0 posts without facts
As of
25 Sep 2026 19:30Z
Coverage computed
2 Oct 2026 02:23Z
Facts checked
2 Oct 2026 03:07:29Z
Posts without facts
0
Days settled
21 of 21 that can settle
Author coverage
posts with an identifiable platform identity; article sources have none by definition
Readings counted
97 readings, current: the release's stored readings that no notice names (none superseded, corrected or withdrawn); read live at page render
Spam, bot and incentive filter, not applied yet
Spam, bot and incentive posts haven’t been removed: their labels agree with a stronger model at kappa 0.43, below the threshold of 0.60

The week in brief · 2026-W01

No prose is stored for this release.

What this release found

The first place on each of the two platforms that count people, in its reading’s own words. Two platforms, two rankings; nothing here is added across them.

01 · #1 on Reddit
ChatGPTProduct · OpenAI

7,211 distinct people on Reddit

On Reddit, 7,211 distinct people posted about ChatGPT between 29 December 2025 and 4 January 2026: 9,865 posts, an estimated 4.28% of that week's Reddit posts about AI, leaving out templated automated posts (99.9% of these posts had an identifiable author).

ChatGPT · Reddit · posts from 29 Dec 2025 – 4 Jan 2026 · n 9,865 posts · author coverage 99.9% · 7 days: 7 backfilled

02 · #1 on Hacker News
ClaudeModel · Anthropic

21 distinct people on Hacker News

On Hacker News, 21 distinct people posted about Claude between 29 December 2025 and 4 January 2026: 21 posts, 10.29% of that week's Hacker News posts about AI (100.0% of these posts had an identifiable author).

Claude · Hacker News · posts from 29 Dec 2025 – 4 Jan 2026 · n 21 posts · author coverage 100.0% · 7 days: 7 backfilled

Each card is the lead measured statement of one platform’s first place: the stored statement whose figures include that platform’s people count, shown verbatim. Platforms rank separately.

The summary table

Every subject’s place on each platform that counts people, side by side. Read down a column for one platform’s ranking; read across a row to see how differently the platforms rank one subject.

People per subject on each platform, 29 Dec 2025 – 4 Jan 2026: the first 15 in Reddit’s order
RankSubject · lineageReddit ▼ sortedn 230,309 postsbar: people, #1 = 7,211Hacker Newsn 204 postsbar: people, #1 = 21
1ChatGPTOpenAI
7,211#1
Hidden · n 6 people: n 6 people, below the threshold of 20 people.
2ClaudeAnthropic
4,842#2
21#1
3GeminiGoogle DeepMind
3,519#3
Hidden · n 6 people: n 6 people, below the threshold of 20 people.
4GPTOpenAI
2,497#4
Hidden · n 3 people: n 3 people, below the threshold of 20 people.
5GrokxAI
1,748#5
Hidden · n 4 people: n 4 people, below the threshold of 20 people.
6Claude CodeAnthropic › Claude
1,522#6
Hidden · n 11 people: n 11 people, below the threshold of 20 people.
7QwenAlibaba
1,173#7
Hidden · n 0 people: n 0 people, below the threshold of 20 people.
8SoraOpenAI
583#8
Hidden · n 0 people: n 0 people, below the threshold of 20 people.
9DeepSeekDeepSeek
546#9
Hidden · n 2 people: n 2 people, below the threshold of 20 people.
10Microsoft CopilotMicrosoft AI
537#10
Hidden · n 0 people: n 0 people, below the threshold of 20 people.
11CodexOpenAI
503#11
Hidden · n 0 people: n 0 people, below the threshold of 20 people.
12WanAlibaba
485#12
Hidden · n 0 people: n 0 people, below the threshold of 20 people.
13LlamaMeta
328#13
Hidden · n 3 people: n 3 people, below the threshold of 20 people.
14GLMZhipu AI
311#14
Hidden · n 0 people: n 0 people, below the threshold of 20 people.
15Grok appxAI › Grok
303#15
Hidden · n 0 people: n 0 people, below the threshold of 20 people.
All 85 subjects70 more in Reddit’s order, 46 of them unranked because their Reddit figure is hidden

All 85 subjects on one page

People on Reddit · n 230,309 posts#1 = 7,211

  1. 1ChatGPT7,211
  2. 2Claude4,842
  3. 3Gemini3,519
  4. 4GPT2,497
  5. 5Grok1,748
  6. 6Claude Code1,522
  7. 7Qwen1,173
  8. 8Sora583
  9. 9DeepSeek546
  10. 10Microsoft Copilot537
  11. 11Codex503
  12. 12Wan485
  13. 13Llama328
  14. 14GLM311
  15. 15Grok app303
All 85 subjects70 more in Reddit’s order; hidden ones hatched

All 85 subjects on one page

People on Hacker News · n 204 posts#1 = 21

  1. 1Claude21
  2. –ChatGPTHidden · n 6 people: n 6 people, below the threshold of 20 people.
  3. –GeminiHidden · n 6 people: n 6 people, below the threshold of 20 people.
  4. –GPTHidden · n 3 people: n 3 people, below the threshold of 20 people.
  5. –GrokHidden · n 4 people: n 4 people, below the threshold of 20 people.
  6. –Claude CodeHidden · n 11 people: n 11 people, below the threshold of 20 people.
  7. –QwenHidden · n 0 people: n 0 people, below the threshold of 20 people.
  8. –SoraHidden · n 0 people: n 0 people, below the threshold of 20 people.
  9. –DeepSeekHidden · n 2 people: n 2 people, below the threshold of 20 people.
  10. –Microsoft CopilotHidden · n 0 people: n 0 people, below the threshold of 20 people.
  11. –CodexHidden · n 0 people: n 0 people, below the threshold of 20 people.
  12. –WanHidden · n 0 people: n 0 people, below the threshold of 20 people.
  13. –LlamaHidden · n 3 people: n 3 people, below the threshold of 20 people.
  14. –GLMHidden · n 0 people: n 0 people, below the threshold of 20 people.
  15. –Grok appHidden · n 0 people: n 0 people, below the threshold of 20 people.
All 85 subjects70 more in Hacker News’s order; hidden ones hatched

All 85 subjects on one page

Distinct people who posted about each subject, 29 Dec 2025 – 4 Jan 2026, as of 25 Sep 2026 19:30Z; published. Each column is scaled to its own #1 and ranks on its own; nothing is added across platforms, because one person can post on several. Rows follow Reddit, the default platform. Hatched cells are hidden, never shown as a number. The stored reason, e.g. Hacker News ChatGPT: “n 6 people, below the threshold of 20 people”. Each column’s n 230,309 posts on Reddit and 204 on Hacker News.

Posts, no peoplearXiv and Google News are not ranked. Article sources have no authors, so people aren’t counted. Posts per subject, largest first:

arXiv n 250 posts · 85 subjects

GPT 14, Gemini 7, Qwen 6, Claude 3 posts, and 81 more subjects

Google News n not recorded · 85 subjects

Not counted this week: Google News has no day between 29 December 2025 and 4 January 2026 that AIDRAN counts, so there’s no figure to show.

The week after

The same subjects one week later, on the platform with enough people. Claude grew on Reddit while Qwen eased.

Reddit · people per subject

2026-W01
29 Dec 2025 – 4 Jan 2026
n 230,309 posts
2026-W02
5–11 Jan 2026
n 266,065 posts
ChatGPT 7,211Claude 4,842Gemini 3,519GPT 2,497Grok 1,748Claude Code 1,522Qwen 1,173Sora 583DeepSeek 546Microsoft Copilot 537Codex 503Wan 485Llama 328GLM 311Grok app 303Google Antigravity 253gpt-oss 229GitHub Copilot 171Gemma 156Gemini API (Google AI Studio) 154Veo 141Kimi 127NotebookLM 126Claude app 122MiniMax M-series 797,516 ChatGPT5,920 Claude3,964 Gemini2,634 GPT2,222 Grok1,885 Claude Code1,391 Microsoft Copilot1,011 Qwen632 Wan619 DeepSeek606 Sora543 Codex427 Gemma398 GLM307 Google Antigravity294 Grok app286 Llama232 gpt-oss230 GitHub Copilot185 Gemini API (Google AI Studio)185 Gemini API (Google AI Studi…185 Claude app148 Veo126 Kimi117 NotebookLM111 MiniMax M-seriesChatGPT 7,211Claude 4,842Gemini 3,519GPT 2,497Grok 1,748Claude Code 1,522Qwen 1,173Sora 583DeepSeek 546Microsoft Copilot 537Codex 503Wan 485Llama 328GLM 311Grok app 303Google Antigravity 253gpt-oss 229GitHub Copilot 171Gemma 156Gemini API (Google AI Studio) 154Gemini API (Google A… 154Veo 141Kimi 127NotebookLM 126Claude app 122MiniMax M-series 797,516 ChatGPT5,920 Claude3,964 Gemini2,634 GPT2,222 Grok1,885 Claude Code1,391 Microsoft Copilot1,011 Qwen632 Wan619 DeepSeek606 Sora543 Codex427 Gemma398 GLM307 Google Antigravity294 Grok app286 Llama232 gpt-oss230 GitHub Copilot185 Gemini API (Google AI Studio)185 Gemini API (Google A…185 Claude app148 Veo126 Kimi117 NotebookLM111 MiniMax M-series

Distinct people per subject on each release’s board, for subjects with n ≥ 100 posts in both weeks (25 on Reddit), as of 25 Sep 2026 19:30Z; published. Each panel on its own scale; lines join one subject’s two weeks. Each week’s n is its platform’s posts: Reddit n 230,309 and 266,065 posts. 2026-W01 and 2026-W02 follow on from each other, so their lines are joined.

Published and captured, day by day

When a post was written and when AIDRAN caught it are different days. For each platform, what was published each day against what was captured that day, and how much of the capture was catching up.What was published each day against what AIDRAN captured that day, and how much of it was catching up.

  • published that day
  • captured that day
  • of those, captured late (late_captured_n)
  • published, not collected yet
Reddit7 days: 7 backfilledown scale, top = 148,013

No capture this week: collection hadn’t started. Posts published on these days were counted from later captures.
own scale, top = 148,013

Bluesky7 days: 7 not startedown scale, top = 107

No capture this week: collection hadn’t started. Posts published on these days were counted from later captures.
own scale, top = 107

YouTube7 days: 7 not startedown scale, top = 1

No capture this week: collection hadn’t started. Posts published on these days were counted from later captures.
own scale, top = 1

Hacker News7 days: 7 backfilledown scale, top = 179

No capture this week: collection hadn’t started. Posts published on these days were counted from later captures.
own scale, top = 179

arXiv7 days: 7 backfilledown scale, top = 100

No capture this week: collection hadn’t started. Posts published on these days were counted from later captures.
own scale, top = 100

Google News7 days: 7 not startedown scale, top = 181

No capture this week: collection hadn’t started. Posts published on these days were counted from later captures.
own scale, top = 181

Exa7 days: 7 not startedown scale, top = 1

No capture this week: collection hadn’t started. Posts published on these days were counted from later captures.
own scale, top = 1

Per platform and day, 29 Dec 2025 – 4 Jan 2026: posts published that day (published_n), posts captured that day whatever their publish date (captured_n), and the part of those captured late (late_captured_n), from conditions.coverage, computed 2 Oct 2026 02:23Z. Each platform on its own scale. A flag bar marks a partial, low-capture or late-heavy day. Not-started days show published only: those posts were captured later and do count.29 Dec 2025 – 4 Jan 2026, from conditions.coverage, computed 2 Oct 2026 02:23Z. Each platform on its own scale; a flag bar marks a partial, low-capture or late-heavy day.

Families and their products

Families and their products on Reddit. A post about Claude Code usually also counts for Claude, so the bars sit together and are never summed.Families and their products on Reddit, side by side and never summed.

The family entry already includes its products; not added.

Claude family · Anthropic

Claude4,842

Grok family · xAI

Grok1,748
Gemini3,519

› Gemini API (Google AI Studio) 154 · Gemini CLI 77 · Gemini app 52

Grok1,748

› Grok app 303

People on Reddit, 29 Dec 2025 – 4 Jan 2026, as of 25 Sep 2026 19:30Z, all on Reddit’s one scale (#1 = 7,211). Groups follow each subject’s stored lineage.

Platform totals

Posts, threads and distinct people on each platform: the context for every share in this release. Each platform stands alone.The context for every share. Each platform stands alone.

Posts, threads and people per platform, 29 Dec 2025 – 4 Jan 2026
PlatformPostsThreadsPeopleAuthor cov.
Reddithiddenhidden
Blueskyhiddenhiddenhidden
Hacker News100.0%
arXivnot applicable0.0%
Google Newshiddenhiddennot applicable

29 Dec 2025 – 4 Jan 2026 · as of 25 Sep 2026 19:30Z · counted from posts’ own data · no platform coverage-limited. People count distinct platform identities and are never added across platforms, so there is no total row. People are not applicable on article sources. Reddit’s posts include its comments. Each figure’s n 230,309 posts on Reddit, not recorded on Bluesky, 204 on Hacker News, 250 on arXiv and not recorded on Google News.

Against the previous reading

Each subject against its previous stored reading under the same method, share and people, platform by platform. Each has its own page, for its size: one cell a subject and platform.

Coverage this week

What AIDRAN was collecting on each platform, each day of the week. A day can hold posts even when collection hadn’t started: they were captured later.What AIDRAN was collecting on each platform, each day.

What AIDRAN was collecting on each platform, each day, 29 Dec 2025 – 4 Jan 2026
PlatformMon 29Tue 30Wed 31Thu 1Fri 2Sat 3Sun 4
RedditMon 29 Dec: backfilledTue 30 Dec: backfilledWed 31 Dec: backfilledThu 1 Jan: backfilledFri 2 Jan: backfilledSat 3 Jan: backfilledSun 4 Jan: backfilled
BlueskyMon 29 Dec: not startedTue 30 Dec: not startedWed 31 Dec: not startedThu 1 Jan: not startedFri 2 Jan: not startedSat 3 Jan: not startedSun 4 Jan: not started
YouTubeMon 29 Dec: not startedTue 30 Dec: not startedWed 31 Dec: not startedThu 1 Jan: not startedFri 2 Jan: not startedSat 3 Jan: not startedSun 4 Jan: not started
Hacker NewsMon 29 Dec: backfilledTue 30 Dec: backfilledWed 31 Dec: backfilledThu 1 Jan: backfilledFri 2 Jan: backfilledSat 3 Jan: backfilledSun 4 Jan: backfilled
arXivMon 29 Dec: backfilledTue 30 Dec: backfilledWed 31 Dec: backfilledThu 1 Jan: backfilledFri 2 Jan: backfilledSat 3 Jan: backfilledSun 4 Jan: backfilled
Google NewsMon 29 Dec: not startedTue 30 Dec: not startedWed 31 Dec: not startedThu 1 Jan: not startedFri 2 Jan: not startedSat 3 Jan: not startedSun 4 Jan: not started
ExaMon 29 Dec: not startedTue 30 Dec: not startedWed 31 Dec: not startedThu 1 Jan: not startedFri 2 Jan: not startedSat 3 Jan: not startedSun 4 Jan: not started
What the fills and flags meanactive, episodic, not started, no lane; partial, low capture, late heavy; settled
Active. AIDRAN was collecting that day.
Episodic. Collected in batches, not continuously.
Not started. Collection hadn’t begun; posts published that day were captured later and do count.
No lane. Collection had stopped.
flag
Partial, low capture, late heavy. A day captured only in part, captured thinly, or whose posts mostly arrived late. On a phone a bar marks a flagged day.
settled
Settled. Every day that can settle has settled; the rest say not applicable. How settling works

Coverage computed 2 Oct 2026 02:23Z · as of 25 Sep 2026 19:30Z · frozen with the release, never recomputed. Each day opens what it holds.A bar marks a flagged day.

Cite this release

Published 2 Oct 2026. Its figures never change. Cite it with its permanent link.

Styles
For reference managers

AIDRAN. Release 2026-W01.

AIDRAN is the author and the publisher, so the citation names it once. To cite one subject, cite its reading page instead.

Licensed CC BY 4.0: reuse it with this citation.

Notices on this release’s readings

None. No reading in this release has been revised or corrected.