What stands out so far. Each finding names the messages it counts, and opens in a tool where every setting can be changed. Click any candidate, anywhere, to see what they say.
This week: Sep 21–Sep 27 on X, against the weeks before
- Website changes: the first comparison comes with the next weekly crawl (the last was Sep 25).
X: competitive-race accounts captured since Aug 1, until every account has four full weeks.
Topics by party
On their campaign websites, Democrats talk more about health care and Social Security (39% vs. 18%) and Republicans more about immigration (21% vs. 10%).
On Websites · all races
Each row compares the parties on one topic: the share of candidates' own messages that touch it, averaged across each party's candidates. Tap a topic to see every candidate behind it, then a candidate to read them. Based on 110 Democrats (1,398 messages) and 92 Republicans (899) with at least 5 messages each.
- Democrats
- Republicans
- A difference the data can back
- Could be chance
Said vs. pushed
Education takes 9% of Republicans' topic talk on their websites but 4% in their posts on X; economy and taxes gets more there (Republicans: 17% → 24%).
On campaign websites vs. X (each account from the day it was captured in full, Jul 27–Oct 4) · all races · candidates countable in both: 21 D, 14 R
In this figure
Each arrow runs from a party's campaign websites (open circle) to what the same candidates post in their posts on X (filled). A topic is measured as its share of all the topics a candidate raises, so a long page and a short post weigh alike.
- Democrats: website → X
- Republicans: website → X
Table view
| Topic | D websites | D X | R websites | R X |
|---|---|---|---|---|
| Cost of living | 15% | 19% | 10% | 16% |
| Economy and taxes † | 14% | 12% | 17% | 24% |
| Health care and Social Security † | 14% | 14% | 8% | 6% |
| Immigration † | 3% | 4% | 8% | 4% |
| Crime and public safety † | 3% | 2% | 8% | 10% |
| Abortion † | 3% | 1% | 1% | 0% |
| Democracy and government † | 13% | 19% | 9% | 11% |
| Education † | 6% | 5% | 9% | 4% |
| Energy and environment † | 8% | 3% | 7% | 4% |
| Technology and AI † | 5% | 5% | 4% | 3% |
| Disasters † | 1% | 2% | 0% | 1% |
| Foreign policy and defense † | 3% | 6% | 2% | 2% |
| Veterans † | 3% | 2% | 4% | 4% |
| Guns † | 1% | 0% | 1% | 1% |
| Culture and social issues † | 1% | 1% | 6% | 6% |
| Agriculture and rural issues † | 3% | 4% | 5% | 5% |
| Infrastructure † | 2% | 0% | 2% | 1% |
| Tribal affairs † | 1% | 1% | 0% | 0% |
Loud silences
No Republican raises corporate power / consumers on X; 27% of Democrats do.
On X (Sep 28–Oct 4) and campaign websites · all races · detailed topics
Topics one party's candidates almost never raise, though at the other party's rate they would have mentioned them dozens of times. Where a plain word backs the topic, its count is a check that doesn't depend on the labeler.
Republicans are silent on
Corporate power / consumers† on X
38 of 38 never mention it (27% of Democrats do). At Democrats' rate, 16 mentions; 0 found.
Iran war† on X
37 of 38 never mention it (34% of Democrats do). At Democrats' rate, 22 mentions; 3 found. The word “Iran”: 1 of their 402 messages, against 17 of 403 for Democrats.
Tariffs / trade‡ on their websites
86 of 92 never mention it (50% of Democrats do). At Democrats' rate, 78 mentions; 6 found. The word “tariffs”: 4 of their 954 messages, against 116 of 1,551 for Democrats.
Democrats are silent on
Crime / policing† on X
38 of 41 never mention it (45% of Republicans do). At Republicans' rate, 25 mentions; 4 found. The word “crime”: 3 of their 403 messages, against 9 of 402 for Republicans.
Transgender issues† on their websites
108 of 110 never mention it (39% of Republicans do). At Republicans' rate, 96 mentions; 2 found. The word “transgender”: 3 of their 1,551 messages, against 18 of 954 for Republicans.
Border security† on their websites
80 of 110 never mention it (64% of Republicans do). At Republicans' rate, 185 mentions; 46 found. The word “border”: 87 of their 1,551 messages, against 139 of 954 for Republicans.
Messaging vs. need
Talk of agriculture and rural issues follows farm jobs: Republicans give it 4% of their messages where farm jobs are rarest and 19% where they're most common; talk of health care costs and access doesn't follow the uninsured rate.
On Websites · all races · Census ACS 2020–24: the district for House races, the state otherwise; House races in the ten redrawn states left out
In this figure
Each dot is a candidate: across, the share of jobs in farming, forestry, fishing and mining where they run; up, the share of their own messages about agriculture and rural issues †. The line is each party's trend with its 95% band. Places differ in many ways at once (farm districts lean Republican), so this shows association, not cause. “Rarest” and “most common” in the headline are the third of candidates lowest and highest on the measure. Click a dot to open the candidate.
Democrats: r = 0.30 (95%: 0.12 to 0.51), 92 candidates · Republicans: r = 0.40 (95%: 0.24 to 0.55), 75 candidates · U.S. Census Bureau, ACS 2020-2024 5-year estimates
Table view
| Topic vs. measure | Party | Candidates | r (95% interval) | Share: lowest third → highest third |
|---|---|---|---|---|
| Agriculture and rural issues vs. farm jobs | Democrats | 92 | 0.30 (0.12 to 0.51) | 5% → 14% |
| Republicans | 75 | 0.40 (0.24 to 0.55) | 4% → 19% | |
| Housing vs. rent burden | Democrats | 92 | 0.26 (0.04 to 0.44) | 19% → 25% |
| Republicans | 75 | 0.21 (-0.01 to 0.38) | 9% → 12% | |
| Medicaid vs. medicaid | Democrats | 92 | -0.01 (-0.18 to 0.14) | 9% → 7% |
| Republicans | 75 | -0.01 (-0.12 to 0.13) | 1% → 1% | |
| Health care costs and access vs. uninsured | Democrats | 92 | -0.03 (-0.26 to 0.17) | 34% → 32% |
| Republicans | 75 | -0.08 (-0.31 to 0.16) | 17% → 11% | |
| Border security vs. born abroad | Democrats | 92 | 0.03 (-0.22 to 0.31) | 2% → 4% |
| Republicans | 75 | -0.09 (-0.28 to 0.09) | 17% → 13% | |
| Food and income assistance vs. snap | Democrats | 92 | 0.15 (-0.02 to 0.31) | 5% → 7% |
| Republicans | 75 | -0.02 (-0.25 to 0.22) | 3% → 3% |
Talking points that travel
“members of Congress from trading stocks” appears word for word from 25 candidates of both parties; “dime of corporate PAC money” from 10 Democrats.
On every channel since Aug 1 · the campaigns' own words, in messages about issues · lines from 3+ candidates
Each party's words
Democrats say “reproductive”, “corporations” and “medicaid”; Republicans say “conservative”, “regulations” and “radical” on their websites.
On campaign websites · 110 Democrats, 92 Republicans with 5+ messages · words used by 8+ candidates · no labeler
In this figure
The words most distinctive of each party, counted by how many of its candidates use them. Bars: share of each party's candidates who do. Names and places are left out.
Democrats' words
- reproductive52% vs. 2%
- corporations52% vs. 7%
- medicaid54% vs. 10%
- health care67% vs. 25%
- everyone61% vs. 20%
- democracy47% vs. 8%
- tariffs39% vs. 3%
- care91% vs. 58%
- public schools44% vs. 9%
- democratic55% vs. 20%
- climate39% vs. 5%
- health87% vs. 54%
Republicans' words
- conservative51% vs. 5%
- regulations49% vs. 10%
- radical38% vs. 2%
- fiscal41% vs. 7%
- men47% vs. 14%
- borders43% vs. 12%
- mandates32% vs. 3%
- spending67% vs. 34%
- american energy32% vs. 1%
- second amendment32% vs. 5%
- sports40% vs. 12%
- wasteful32% vs. 5%
Tick: the other party's share of candidates using the word.
Party beats place
Two Democrats from different states share more of their topic mix (0.55) than a race's own two nominees (0.47), who overlap barely more than a Democrat and a Republican from different races (0.45).
On Websites · all races · detailed topics · 38 races with one Democrat and one Republican
Each dot is a race: 1 means both nominees spread their talk over the same topics in the same proportions, 0 nothing in common. The shaded band is the middle 90% of Democrat–Republican pairs from different races. Click a race to open it.
Furthest apart: (0.24), (0.26), (0.31). Two Republicans from different races: 0.53.
What changed on their websites
The first week-to-week comparison comes with the next crawl
On campaign websites, crawled weekly
Each campaign's homepage, issues and about pages are read once a week; the last visit was Sep 25, 2026. From the second visit on, this lists what each campaign added and removed, with the text, so a dropped position or a deleted page stays on the record.
Naming Trump
Democrats name Trump about 2× as often as Republicans on their campaign websites.
The typical Republican never names Trump in their posts on X; the typical Democrat does. Democrats name Trump far more often than Republicans in their YouTube videos. Each filled circle is a candidate, placed by how often their own messaging names Donald Trump per 1,000 words. Hollow circles never name him. Hover a circle for the candidate, click to open them.
- Democrats
- Republicans
- Never names him
- Party median
- Highest in the row, named
Campaign websites
Democrats128 candidates · 38% never name him
name him
Republicans118 candidates · 47% never name him
name him
Posts on Xsince Sep 29, when every candidate's account began to be captured in full
Democrats9 candidates · 33% never name him
name him
Republicans5 candidates · 100% never name him
name him
Campaign YouTube
Democrats31 candidates · 16% never name him
name him
Republicans28 candidates · 43% never name him
name him
Blueskyalmost no Republicans post here, so it isn't a party comparison
Democrats70 candidates · 26% never name him
name him
The safer the seat, the more both parties name him: 81% of Republicans in the safest seats do on their websites, 34% in seats safe for Democrats
The share of each party's candidates who name Trump at all on their campaign websites, by how safe the seat is for their own party (Sabato's Crystal Ball's rating).
In this figure
- Safe for their partyD 71% of 34R 81% of 21
- Likely or lean theirsD 68% of 37R 50% of 16
- Toss-upD 73% of 22R 65% of 20
- Likely or lean the other side'sD 50% of 16R 44% of 32
- Safe for the other partyD 32% of 19R 34% of 29
Bands with fewer than 5 candidates aren't drawn. Candidates with at least 500 words on their websites. Ratings follow district lean, so this is the district as much as the candidate.
Table view
| Channel | Party | Candidates | Never name him | Median per 1,000 words |
|---|---|---|---|---|
| Websites | Democrats | 128 | 49 (38%) | 0.58 |
| Websites | Republicans | 118 | 56 (47%) | 0.29 |
| X | Democrats | 9 | 3 (33%) | 3.69 |
| X | Republicans | 5 | 5 (100%) | 0.00 |
| YouTube | Democrats | 31 | 5 (16%) | 1.39 |
| YouTube | Republicans | 28 | 12 (43%) | 0.11 |
| Bluesky | Democrats | 70 | 18 (26%) | 2.16 |
Over time
Topics over time in their posts on X (Jul 27–Oct 4): too few candidates each week for a headline, so read the panels.
On X · Jul 27–Oct 4 · all races
Each panel is one topic: the weekly share of own messages that touch it, averaged over candidates with at least 5 that week. Every week has the same set of candidates; a hollow point is a week still in progress. Each panel has its own scale, so small topics show their movement; read the percentages, not the heights, when comparing panels.
Websites are a snapshot of the current pages, with no weeks, so this figure uses X, where both parties post every day.
Before Sep 29, only the X accounts in competitive races were captured in full, so this covers those candidates; from Sep 29 on, every account is covered.
- Democrats
- Republicans
Compare any two groups
On their campaign websites, favored Republicans give more of their messaging to foreign policy and defense (12% vs. 6%), while trailing Republicans more to cost of living (30% vs. 15%).
On Websites · all races
Pick two groups of candidates. Each bar is group A's share of messages on a topic minus group B's, in percentage points, with a 95% interval. Dark bars are the differences whose interval excludes zero. Standing is the candidate's current Kalshi win probability; the 2024 vote is The Downballot's count for the district (House) or state.
← more in trailing Republicansmore in favored Republicans →
† Provisional: below the accuracy bar on the hand-labeled test set, but the labeler's tags for it are usually right, so it still carries signal; read it with more caution. ‡ Unverified: too few hand-labeled examples, or the labeler handles it poorly; shown only on request. Hover or tap a mark for the numbers.