ShareChat: A Dataset of Chatbot Conversations in the Wild
Paper • 2512.17843 • Published • 2
Datasets:
Request access to sharetrawl
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
This dataset holds real conversations that people published from ChatGPT, Claude, Gemini, Grok and other assistants, collected from the open web. It contains personal information people chose to share and material that is offensive, explicit or harmful. Access is approved by hand, for research. By requesting access you agree to: use the dataset under CC BY-NC 4.0 (non-commercial, attribution), the terms the embedded ShareChat rows carry; make no attempt to identify the people in these conversations or to infer sensitive personal data about them; apply your own filtering and safety measures before training or evaluating models on it; respect the terms of service of the platforms the conversations came from; not redistribute the dataset; and delete any conversation the maintainers notify you has been removed.
Log in or Sign Up to review the conditions and access this dataset content.
sharetrawl
253,192 publicly-shared LLM conversations across 9 platforms including toxic output.
Last updated: 2026-10-05
This dataset extends ShareChat (paper) and is updated weekly. 75,007 of its 253,192 urls came from ShareChat's list (30%) rather than from our own discovery.
Have share links this corpus is missing? Paste them at https://sharetrawl-submit.hails.workers.dev
Motivation
Public chat datasets lag live traffic by 6 to 12 months (WildChat-4.8M → July 2025; ShareChat → October 2025); and often redact harmful/toxic output (e.g. WildChat-Full). This limits research into in-the-wild safety harms and makes benchmarking monitors on real-world use-cases challenging.
Composition
| platform | conversations |
|---|---|
| chatgpt | 114,925 |
| grok | 83,723 |
| perplexity | 21,043 |
| gemini | 19,299 |
| claude | 8,171 |
| manus | 2,348 |
| qwen | 1,394 |
| deepseek | 1,306 |
| kimi | 983 |
| total | 253,192 |
Time distribution
The earlier of created_at (when the conversation happened) and captured_at (when we fetched
it).
| year | conversations |
|---|---|
| 2022 | 8 |
| 2023 | 16,576 |
| 2024 | 30,375 |
| 2025 | 96,737 |
| 2026 | 99,832 (to 2026-09) |
243,528 of 253,192 conversations carry one of the two dates; the remaining 9,664 appear in no year above, having neither.
Harm labels
labels/sharetrawl_broad_labels.parquet scores 253,146 conversations with a two-stage cascade: a cheap gate
(z-ai/glm-5.3-flash) admits anything scoring ≥ 5, and a stronger oracle
(claude-opus-4-8) gives each admitted conversation a 0–100 score and a list of categories. Both
scores are stored (gate_score, opus_score); flagged is oracle score ≥ 50, and
15,294 (6.04%) are flagged.
Evidence window: whole-conversation (labelled 2026-10-05, declared in
labels/METADATA.json): both stages read the whole conversation, in windows of
40,000 characters with the strongest verdict kept (n_windows, windowed).
46 rows (46 of them flagged) have no url at all.
The gate's misses are measured: on 2026-09-08, 2,000 gate-rejected conversations were scored by the oracle and 1 of 2,000 came back flagged — a false-negative rate of 0.05% (95% CI 0.01–0.28%), or roughly 79 missed flags (at most 446) across the 157,849 rejected. The true rate over the labelled subset is therefore likely near 6.07% (at most 6.22%), not 6.04%.
By category
A conversation can carry several categories, so these sum to more than the flagged total. Categories with 10 conversations or fewer are omitted.
| category | conversations |
|---|---|
| misinformation | 13,147 |
| deception | 7,158 |
| defamation | 4,235 |
| hate_harassment | 2,484 |
| privacy | 2,299 |
| sexual | 1,932 |
| other | 1,832 |
| jailbreak | 1,073 |
| illicit | 1,039 |
| self_harm | 811 |
| unknown | 512 |
| violence | 508 |
| cbrn | 290 |
| harassment | 261 |
| cyber | 235 |
| harassment_hate | 22 |
By platform
| platform | flagged | scored | rate |
|---|---|---|---|
| grok | 8,448 | 83,689 | 10.09% |
| kimi | 87 | 983 | 8.85% |
| deepseek | 93 | 1,306 | 7.12% |
| claude | 556 | 8,170 | 6.81% |
| qwen | 77 | 1,394 | 5.52% |
| manus | 103 | 2,348 | 4.39% |
| chatgpt | 4,623 | 114,916 | 4.02% |
| gemini | 666 | 19,297 | 3.45% |
| perplexity | 641 | 21,043 | 3.05% |
By source slice
| slice | flagged | scored | rate |
|---|---|---|---|
| sharetrawl | 14,891 | 243,486 | 6.12% |
| sharechat | 403 | 9,660 | 4.17% |
The gap is primarily platform mix, not corpus: within platform, text-deduplicated, the two are broadly comparable.
To break the rate down by where a URL was found, group on provenance_source; it is set on
163,843 of 253,146 scored conversations, absent on the earliest-discovered
rows, and thin enough per backend that small denominators dominate.
By platform and discovery route
provenance_source is the highest-priority backend that found a URL while it was still new to us,
not the only place it appeared. Grouping backends into route classes (census refetch, web archive,
code search, social, passive scan, curated dataset, the un-provenanced fleet bulk) and splitting grok
by share host, every cell with at least 200 scored conversations:
| platform | route | flagged | scored | rate |
|---|---|---|---|---|
| chatgpt | census-refetch | 3,429 | 69,786 | 4.91% |
| chatgpt | web-archive | 238 | 15,349 | 1.55% |
| chatgpt | social | 472 | 8,545 | 5.52% |
| chatgpt | census-csv | 328 | 7,427 | 4.42% |
| chatgpt | curated-dataset | 15 | 5,056 | 0.30% |
| chatgpt | code-search | 71 | 5,020 | 1.41% |
| chatgpt | passive-scan | 70 | 3,726 | 1.88% |
| claude | pre-provenance-fleet | 330 | 3,890 | 8.48% |
| claude | code-search | 68 | 1,911 | 3.56% |
| claude | social | 105 | 1,510 | 6.95% |
| claude | curated-dataset | 26 | 452 | 5.75% |
| deepseek | code-search | 38 | 675 | 5.63% |
| deepseek | social | 35 | 505 | 6.93% |
| gemini | pre-provenance-fleet | 416 | 11,981 | 3.47% |
| gemini | social | 135 | 2,499 | 5.40% |
| gemini | census-refetch | 33 | 1,679 | 1.97% |
| gemini | code-search | 34 | 1,423 | 2.39% |
| gemini | web-archive | 30 | 1,037 | 2.89% |
| gemini | census-csv | 14 | 570 | 2.46% |
| grok (grok.com) | pre-provenance-fleet | 6,368 | 63,765 | 9.99% |
| grok (grok.com) | social | 211 | 1,472 | 14.33% |
| grok (grok.com) | code-search | 116 | 1,024 | 11.33% |
| grok (grok.com) | census-csv | 46 | 589 | 7.81% |
| grok (grok.com) | census-refetch | 42 | 558 | 7.53% |
| grok (x.com) | web-archive | 1,344 | 13,190 | 10.19% |
| grok (x.com) | social | 106 | 1,801 | 5.89% |
| grok (x.com) | code-search | 168 | 943 | 17.82% |
| kimi | web-archive | 70 | 783 | 8.94% |
| manus | web-archive | 83 | 1,895 | 4.38% |
| manus | social | 3 | 231 | 1.30% |
| perplexity | web-archive | 520 | 15,219 | 3.42% |
| perplexity | census-refetch | 56 | 2,871 | 1.95% |
| perplexity | census-csv | 14 | 1,046 | 1.34% |
| perplexity | code-search | 13 | 1,022 | 1.27% |
| perplexity | social | 37 | 858 | 4.31% |
| qwen | web-archive | 74 | 676 | 10.95% |
| qwen | code-search | 2 | 634 | 0.32% |
Grok's lead over perplexity survives within routes: it flags more in every route both were found through (census-csv, census-refetch, code-search, social, web-archive), so the gap in the platform table is not route mix alone.
Turn-level labels
The cascade above scores whole conversations. Separate classifier layers additionally score every text-carrying turn, which is the unit ShareChat and WildChat report on.
| layer | model | flagged when | flagged | scored | rate | coverage |
|---|---|---|---|---|---|---|
| Detoxify | unitary/toxic-bert |
toxicity >= 0.5 |
28,328 | 1,925,918 | 1.47% | 76.7% |
| Llama Guard 3 1B | meta-llama/Llama-Guard-3-1B |
p_unsafe >= 0.9 |
88,849 | 1,925,918 | 4.61% | 76.7% |
| OpenAI Moderation † | omni-moderation-latest |
flagged, the file's own boolean |
76,038 | 2,106,237 | 3.61% | 83.9% |
| Broad cascade (turn-level) | claude-haiku-4-5-20251001 → claude-opus-4-8 |
flagged, the file's own boolean |
69,139 | 2,494,235 | 2.77% | 99.3% |
† OpenAI Moderation is still being scored. It walks the corpus in file order, so its coverage concentrates in the platforms early in the scan and is not a random sample — group by platform before trusting its rate.
Data structure
One row per conversation (not per message, unlike ShareChat's CSV release), with the turns
nested in a messages list.
| column | meaning |
|---|---|
platform |
responder platform (chatgpt, grok, perplexity, gemini, claude, manus, qwen, deepseek, kimi) |
url |
the public share URL |
source |
whose copy of a conversation this row is — sharechat (base census) or sharetrawl (our own capture) — not which corpus first found it |
provenance_source |
which discovery backend found the URL (github, wayback, …) |
provenance_ref |
the exact item that surfaced it — an archive snapshot, a scan record, or the public post, repo file or forum thread where the link was shared |
model |
the model that wrote the conversation's last answer, where the share names it — on chatgpt, the last answer's own model slug, else the model picker's setting (auto, research) when no answer names one. Null where the share names no model: every claude and kimi share, and grok rows captured from x.com |
created_at / published_at |
conversation timestamps, where available |
captured_at |
when we captured it — set only on fresh rows |
topic / language |
classified topic and language — ShareChat's labels on base rows; language on our own captures is identified by Claude Sonnet |
summary_headline / summary / summary_outcome / summary_terms |
a model-written English index entry: a headline noun phrase, two to four sentences ending with how the conversation visibly ends, an outcome (answered, refused, unfinished, empty, unclear) judged from the last visible exchange, and three to six lowercase public search terms. Written by Claude Sonnet 5 (the backfill) and GLM-5.3-Flash on zero-data-retention endpoints (the increment); names of private people and the user's own projects are deliberately absent. Empty where no pass has read the conversation yet |
summary_window |
full when the model read the whole conversation in one window; hierarchical when a long one was read in stretches, each condensed and the condensed transcript then summarised; head_tail when a long one was cut to its opening and ending — the summary then says "After an unseen stretch," before the ending |
turns_count |
number of turns |
messages |
list of {message_index, role, plain_text, thinking, tool_calls, code, analysis, version}. role is ShareChat's user / llm, plus export notice on chatgpt rows for the export's fixed Original custom instructions no longer available line — a redaction notice, not a turn anyone typed. thinking is the reasoning the platform exposes for an assistant turn, as plain text (chain-of-thought on DeepSeek, Qwen, Kimi and Grok-on-X; reasoning summaries on ChatGPT, Grok and Manus; ShareChat's base-row values flattened to their content), empty where the share carries none (Claude, Gemini, Perplexity). tool_calls is a JSON list of the tools that turn ran, {kind, name, input, output} — kind is one of code, search, browse, file, artifact, image, memory, mcp, plugin, other; name the platform's own tool name; searched and fetched pages are reduced to [{title, url}] locators while code output and error text are kept whole; null output where the platform recorded none; empty on Gemini and on every base row. code / analysis / version are ShareChat's, base rows only |
redactions/<platform>.parquet holds each platform's encrypted redaction mappings (url,
payload); see Redaction. It is not conversation data and has no use without the key.
published_urls.sha256 lists the sha256 of every url any retained revision has carried since
the repository's history was last purged, one per line — a superset of this revision's rows, label
files included. It is how a publish knows whether a since-withheld conversation was ever public
(see Removal requests), and it lets you check whether a url has appeared in this dataset without
downloading the parquets.
How the fresh slice is collected
Share URLs are discovered by unioning many backends — web archives, passive-scan corpora (urlscan.io, AlienVault OTX), code search (GitHub, SourceGraph), social/forum search (Bluesky, Reddit, Hacker News, Lemmy) and WeChat public-article search (Sogou) — then de-duplicated against what is already stored. Each URL is rendered in a real browser (please reach out for the HAR files).
Safety and exclusions
Child-safety material is excluded via an automatic screen that runs before every publish and fails
closed — a classifier refusal quarantines rather than passes, and a capture whose screen call
did not complete is withheld until a later run clears it. Quarantined conversations are held in a
content-free denylist of hashed URLs (no text, no URLs) and never enter the corpus. In the label
file, rows categorised csae retain the score and category but have url and id blanked. The
screen reads each turn's plain_text; a turn's thinking and tool_calls publish as captured and
have not been read by any classifier.
These are real conversations from the open web. They contain material that is offensive, explicit, or harmful.
Redaction
Personal data about private individuals is replaced by a placeholder before publication. Every conversation is read by Claude Opus 5.5 (Claude Sonnet 5.5 where Opus declines), which lists the names, email addresses, phone numbers, street addresses, account handles, identity and account numbers, dates of birth, public IP addresses and credentials of private people: the user and anyone they, a pasted document or a search result name who is not a public figure. Public figures in their public role, fictional characters, organisations and documentation examples are left as written. Credential-shaped strings are also caught by pattern rules, whatever the model says.
Each value becomes a typed, numbered placeholder — [REDACTED:PERSON_1], [REDACTED:EMAIL_2],
[REDACTED:OPENAI_KEY_1] — that is consistent within a conversation (the same person is
PERSON_1 wherever they recur in it) and restarts at 1 in the next, so a placeholder carries
nothing across conversations. Placeholders appear in every text field of messages and in the
summary_* columns. A conversation the models could not read is not published.
The originals are kept in redactions/<platform>.parquet: one row per redacted conversation,
its url in the clear and its placeholder-to-original mapping encrypted (AES-256-GCM, bound to
the url) under a key the maintainers hold and do not distribute. The files let the maintainers
reproduce the unredacted corpus exactly; without the key they carry nothing readable. A removal
request withholds a conversation's sidecar row with the rest of it.
Redaction is a model judgement and misses some personal data, particularly indirect identifiers (employers, small towns, unusual life events) that are not in the list above.
Removal requests
If you are in one of these conversations and want it removed, or you hold rights in one, use the
anonymous removal form at https://sharetrawl-submit.hails.workers.dev/remove. It takes the share URL (the url value) and
an optional reason; it needs no account or email, it does not record your IP address, and only
the maintainers can read what it stores. It gives the same answer whether or not the link is in
this dataset, so it cannot be used to look one up. Please do not post a share URL on this
dataset's Community tab: discussions there are public, so the link would reach exactly the people
you want it kept from. The Community tab is for questions that do not name a conversation.
Once a request is actioned, the conversation leaves the corpus at the next weekly publish, and that publish purges the repository's history (squashes it to one commit), so the conversation does not persist in an earlier revision. Otherwise revisions are kept. Removals are recorded as a content-free list of hashed URLs, never as the URL itself, and the request's link and reason are deleted once it is actioned.
Limitations
- Shared conversations are not representative of all usage. People share what is interesting, impressive, funny, or outrageous.
- Platform mix is uneven and shifts over time, driven by what is discoverable rather than by real platform share.
- Harm labels are model judgements, not human annotated nor the truth.
Citation
Please reach out to me for a citation, as the major work is not yet released.
Cite ShareChat as well:
@misc{yan2026sharechatdatasetchatbotconversations,
title = {ShareChat: A Dataset of Chatbot Conversations in the Wild},
author = {Yueru Yan and Tuc Nguyen and Bo Su and Melissa Lieffers and Thai Le},
year = {2026},
archivePrefix = {arXiv},
primaryClass = {cs.CL},
url = {https://arxiv.org/abs/2512.17843},
}
License
CC BY-NC 4.0, inherited from ShareChat — this dataset embeds the ShareChat corpus, whose data structure and distribution are licensed CC BY-NC 4.0 under the ShareChat Dataset License Agreement, so the original terms apply to the whole of it. Non-commercial use only, attribution required. The licence covers the compilation and our labels; the conversations themselves were written by their participants and the responding models, and the originating platforms' terms of service continue to apply to that content.
ShareChat's community guidelines carry over, and access here is granted on the same terms: make no attempt to identify the people in these conversations or to infer sensitive personal data about them; apply filtering and safety measures before training or evaluating on this data; respect the terms of service of the originating platforms; comply with applicable law and your institution's ethical review; the views in the conversations are not those of the maintainers; and the dataset is provided as is, without warranty of any kind.
- Downloads last month
- 328