Archive
The archive contains 16,714 tweets from @jk_rowling, dated 17 September 2009 through 20 September 2026. It includes original tweets, replies, retweets, and 93 records marked as recovered deleted tweets in the source database.
Updated 20 September 2026 at 12:23 UTC. Today is partial. It is a collected snapshot, not a complete history of the account. Deleted status comes from the source archive and was not rechecked on the live account for this release.
The tweet cards are recreated from archived text and recorded engagement counts. They are not screenshots or live embeds. They use the account photo retrieved in September 2026, rather than the photo displayed when each tweet was published.
Classifications
16,714 tweets have classifications and 0 remain unclassified. The September 2026 refresh added 188 tweets and classified 243 records: those new tweets plus 55 older gaps. Existing classifications were preserved.
The stored model identifier is claude-sonnet-5. The refresh used the original system prompt and classification schema, with the same requested model. Earlier runs received tweet text and available quote text or reply metadata; the refresh also supplied available reply-parent and repost text. This difference in context can affect classification and should be considered when comparing periods. It assigned topic, stance, rhetorical-device, descriptor, and intensity labels. Some context displayed on this site was recovered later and may not have been available during classification.
Trans-related covers discussion of trans people, transition, sex and gender definitions, and related debates about law, sport, healthcare, spaces, and safeguarding. It identifies the subject of a tweet. It does not establish the identity of a person being discussed or classify every included tweet as hostile.
| Label | Coding definition |
|---|---|
| Hostile | Derogatory or dehumanizing. |
| Mocking | Ridicule or contempt. |
| Critical | Opposition without derogatory language. |
| Defensive | Defending the author or allies from criticism. |
| Neutral | No supportive or opposing stance assigned. |
| Supportive | Support for trans people or their rights. |
Intensity is an ordinal rating from 0 to 4 for hostility or derogation. Confidence is the model’s own score; it has not been calibrated against an independent accuracy measure.
Topics, descriptors, and rhetorical devices allow multiple labels. Descriptor categories cover misgendering, sexualization, political framing, jargon and acronyms, delegitimizing language, derogatory slurs, and neutral wording.
Counting rules
- Annual topic share: trans-related classified tweets divided by all classified tweets in the same UTC calendar year. Includes replies and retweets. Excludes unclassified tweets.
- Stance: one label for each trans-related classified tweet. Critical, mocking, and hostile can be summed, but remain separate categories.
- Topics, devices, and descriptors: each tweet counts once per assigned category. Category totals can exceed the number of tweets.
- Engagement: median likes on original tweets, excluding replies and retweets, from the existing quarterly study, which ends in August 2026. That older series is retained with its original measurements. Counts were recorded at collection.
- Incomplete periods: 2026 runs through 20 September 2026; the latest day is partial. Early years contain very few collected tweets.
Detailed analysis
The five findings on the analysis page are calculated from the same archive. The calculations use the stored annotations, including those added in the September refresh. The charts and supporting tables show the counts, denominators, and sample sizes used in the comparisons.
- Change in stance: each annual bar divides stance counts by trans-related tweets of the selected post type. This differs from annual topic share, whose denominator is all classified tweets. Count changes compare the two latest complete calendar years, 2024 and 2025. The partial 2026 results are also shown.
- Frequency: an active day has at least one collected post of any type or classification status. A trans-related day has at least one tweet with that topic label. The ten-day concentration sums the ten highest daily trans-related counts and divides by the annual total. Dates use UTC. No missing date is treated as evidence that the account was silent.
- Descriptors: a tweet counts once per category even if it has several matching annotations. Descriptor labels can reflect quote or reply context, not only the author’s own words. Comparisons by post type divide by all trans-related posts of that type.
- Topics: hostile counts are divided by all trans-related tweets with each topic label. Tweets can belong to several topics. The interactive comparison shows categories with at least 30 tweets in the selected period. These are descriptive comparisons, not estimates adjusted for post type, timing, or the people and events discussed.
- Engagement: compare original trans-related tweets coded hostile with original trans-related tweets assigned any other stance, within each UTC quarter of 2024–25. Replies and retweets are excluded. Missing likes are excluded and counted separately in the export; zero likes are included. A quarter is compared only if each group has at least five recorded counts. All eight quarters meet that rule. This is separate from the older quarterly study further down the page, which retains its August 2026 collection window.
Medians reduce the influence of extreme engagement counts. Quarterly comparisons reduce broad changes in the account’s audience over time, but do not control for topic, reach, within-quarter timing, deletion, or tweet age at collection. The analysis cannot establish causation, audience motives, classifier accuracy, or statistical significance. The classifications themselves are an additional source of uncertainty.
Counting an accusation
The analysis page counts how often Rowling calls trans people fetishists. Three decisions shape that number, and each can be checked.
- The words: any word beginning “fetish” — fetish, fetishes, fetishism, fetishist, fetishists, fetishistic, fetishise and their variants. Adjacent terms such as autogynephilia, paraphilia and kink are not included in the count.
- Whose words: only her own text. A post is not counted because someone she replied to, or a post she quoted, used the word. Matching is case-insensitive.
- Which sense: 41 posts use such a word. 33 apply it to trans people, to trans women as a class, or to an individual she describes as a man dressing as a woman. The other 8 use it in a different sense and are excluded: 3 non-sexual uses such as “fetishise victimhood”, 2 where she repeats the phrase “genital fetishist” to reject it as a smear against lesbians, 2 jokes not about trans people, and 1 describing women as the people being fetishised.
Deciding which sense a post carries is an editorial reading, not a model output. Each of the 41 posts is recorded individually with its classification and a short reason, and the build fails if that record does not cover exactly the posts the word search returns, so a later snapshot cannot quietly drop one or add an uncoded one. 4 of the 33 counted posts quote or parody someone else’s words rather than stating the claim directly; she endorses each. Every post can be read in full through the archive search.
Annual shares divide these posts by all trans-related classified posts in the same UTC calendar year, so they answer how much of her writing about trans people carries the accusation, not how much of it she posts overall. 2026 is partial. The count is a floor: it finds the word, not every paraphrase of the idea.
Limitations
Collection gaps, deletions, and missing context affect the record. Brief replies and sarcasm are particularly difficult to classify. A model label cannot, by itself, substantiate a factual allegation about someone mentioned in a tweet.
Engagement figures can be affected by follower growth, platform changes, tweet age, and deletion. A rise in likes does not show that a particular stance caused it.
Quoted text belongs to the identified author. Its inclusion does not endorse the claims it contains. Original links may no longer work, and a Wayback lookup may return no captures.
Editorial position
“J.K. Rowling is transphobic” is this site’s assessment of her public rhetoric. The quantitative analysis describes topics and language. It does not measure her private beliefs or motives.
The 10 homepage examples were selected for explicit misgendering, sexualized insults, and ridicule of trans people. The four lead examples deny that trans children exist, reduce a trans woman to her genitals in a sentence about rape, declare that recognising a named trans woman as a woman is a lie, and end a thread naming ten trans women by calling them men. One phrase is highlighted in each; the highlighting is added by this site and does not change the tweet text. One featured post concerns Imane Khelif, who is not transgender, and is included because it shows the same policing of who counts as a woman directed at a cisgender athlete; the note beside it says so.
All featured posts are coded hostile in the existing dataset, with intensity ratings of 3 or 4 out of 4. Every post is reproduced in full and nothing is edited out; posts longer than 900 characters are clipped visually, with a link to the complete record. The selection favours posts that state a position in her own words and need no outside context to be understood; it is not an exhaustive ranking or a representative sample.
Their authorship and wording were checked against public FxTwitter copies on 19 September 2026. Archived text is preserved, including original mentions and shortened links where the live service omits or expands them. All collected tweets, including the supportive classification, remain searchable in the archive.