ShareChat keeps each message's outbound links - the paper's SourceLink field - so citation behaviour is directly measurable. - 5 questions, shown as one
conversation. · × exit demo
round 1 of 5
What share of conversations include links or citations, per platform?
Links are recorded only for Grok and Perplexity; the release has no such field for the other platforms, so they count as 0% here because the field is missing, not because none occurred.
% of conversations with links/citations by platform
grok59.61 (n=13,149)
perplexity57.09 (n=14,612)
What this shows: Each bar is one platform; the value is % of conversations with links/citations. grok is highest at 59.61%; perplexity is lowest at 57.09% across 2 groups.
Only Grok and Perplexity record source links, and more than half of their conversations carry them. The release has no link field for the other platforms, so they are left off the chart rather than shown at 0%.
2 row(s) · 0.054s on the full corpus
compiled SQL (read-only, parameterized)
SELECT grp, AVG(t.has_links) * 100.0 AS value, COUNT(t.has_links) AS n
FROM (
SELECT c.platform AS grp,
c.id AS cid,
c.n_messages AS n_messages, c.turns_count AS turns_count,
cs.has_thinking AS has_thinking, cs.has_code AS has_code,
cs.has_links AS has_links, cs.is_complete AS is_complete
FROM conversations c
LEFT JOIN conv_signals cs ON cs.conversation_id = c.id
WHERE 1=1 AND c.platform IS NOT NULL
) t
GROUP BY grp
ORDER BY value DESC
LIMIT 20
-- params: []
On Grok, are conversations that cite sources longer?
avg messages per conversation
avg messages per conversation
n
8.73
7,838
What this shows: One overall figure: avg messages per conversation is 8.73, computed over 7,838 rows. Scope: platform = grok; has_links = true.
Round 2 narrows to Grok's link-bearing conversations - compare with the next round. (Grok, because its cited and uncited conversations come from the same period; Perplexity's older pages were captured without links.)
1 row(s) · 0.007s on the full corpus
compiled SQL (read-only, parameterized)
SELECT AVG(t.n_messages) AS value, COUNT(t.n_messages) AS n
FROM (
SELECT c.id AS cid,
c.n_messages AS n_messages, c.turns_count AS turns_count,
cs.has_thinking AS has_thinking, cs.has_code AS has_code,
cs.has_links AS has_links, cs.is_complete AS is_complete
FROM conversations c
LEFT JOIN conv_signals cs ON cs.conversation_id = c.id
WHERE c.platform = ? AND cs.has_links = ?
) t
-- params: ['grok', 1]
And how long are Grok conversations without any links?
avg messages per conversation
avg messages per conversation
n
5.73
5,311
What this shows: One overall figure: avg messages per conversation is 5.73, computed over 5,311 rows. Scope: platform = grok; has_links = false.
Same metric, inverted filter: a follow-up refines the previous plan instead of starting over. Cited Grok conversations run noticeably longer.
1 row(s) · 0.006s on the full corpus
compiled SQL (read-only, parameterized)
SELECT AVG(t.n_messages) AS value, COUNT(t.n_messages) AS n
FROM (
SELECT c.id AS cid,
c.n_messages AS n_messages, c.turns_count AS turns_count,
cs.has_thinking AS has_thinking, cs.has_code AS has_code,
cs.has_links AS has_links, cs.is_complete AS is_complete
FROM conversations c
LEFT JOIN conv_signals cs ON cs.conversation_id = c.id
WHERE c.platform = ? AND cs.has_links = ?
) t
-- params: ['grok', 0]
How did Grok's citation usage change month by month?
Dates: Perplexity's are the page's last-updated dates, not creation times, and Claude conversations have no date at all.
Imbalanced groups: 2025-02 has 9,156 rows, 161x 2025-09 (57); compare rates or averages, not raw counts.
2 group(s) with fewer than 30 rows behind them are not shown.
Missing data: 1,446 conversations in the corpus have no timestamp and cannot appear in a by-month breakdown.
% of conversations with links/citations by month
What this shows: % of conversations with links/citations over 9 months: it rose from 57.23% in 2025-01 to 82.46% in 2025-09, peaking at 82.46% in 2025-09. Scope: platform = grok.
SourceLink × Timestamp: the share of cited conversations as a trend line - steady at a little over half. Months with fewer than 30 conversations are left out.
9 row(s) · 0.01s on the full corpus
compiled SQL (read-only, parameterized)
SELECT grp, AVG(t.has_links) * 100.0 AS value, COUNT(t.has_links) AS n
FROM (
SELECT substr(cs.created_day, 1, 7) AS grp,
c.id AS cid,
c.n_messages AS n_messages, c.turns_count AS turns_count,
cs.has_thinking AS has_thinking, cs.has_code AS has_code,
cs.has_links AS has_links, cs.is_complete AS is_complete
FROM conversations c
LEFT JOIN conv_signals cs ON cs.conversation_id = c.id
WHERE c.platform = ? AND substr(cs.created_day, 1, 7) IS NOT NULL
) t
GROUP BY grp
ORDER BY grp
LIMIT 200
-- params: ['grok']
Topic is the dataset's own per-message classifier label (about half of the messages have none); a conversation counts under every topic its messages carry.
Imbalanced groups: specific_info has 9,198 rows, 174x generate_or_retrieve_other_media (53); compare rates or averages, not raw counts.
What this shows: Each bar is one topic; the value is % of conversations with links/citations. purchasable_products is highest at 93.59%; analyze_an_image is lowest at 49.82% (1.9× spread) across 20 groups. Scope: platform = grok.
Final round: grounding differs by subject - product and data questions are the most often sourced.
20 row(s) · 0.2s on the full corpus
compiled SQL (read-only, parameterized)
SELECT grp, AVG(t.has_links) * 100.0 AS value, COUNT(t.has_links) AS n
FROM (
SELECT DISTINCT m.topic AS grp, c.id AS cid,
c.n_messages AS n_messages, c.turns_count AS turns_count,
cs.has_thinking AS has_thinking, cs.has_code AS has_code,
cs.has_links AS has_links, cs.is_complete AS is_complete
FROM messages m
JOIN conversations c ON c.id = m.conversation_id
LEFT JOIN conv_signals cs ON cs.conversation_id = c.id
WHERE c.platform = ? AND m.topic IN ('specific_info', 'other', 'unclear', 'mathematical_calculation', 'tutoring_or_teaching', 'computer_programming', 'relationships_and_personal_reflection', 'greetings_and_chitchat', 'write_fiction', 'argument_or_summary_generation', 'edit_or_critique_provided_text', 'create_an_image', 'data_analysis', 'health_fitness_beauty_or_self_care', 'how_to_advice', 'translation', 'analyze_an_image', 'asking_about_the_model', 'purchasable_products', 'personal_writing_or_communication', 'games_and_role_play', 'cooking_and_recipes', 'creative_ideation', 'generate_or_retrieve_other_media')
) t
GROUP BY grp
ORDER BY value DESC
LIMIT 20
-- params: ['grok']