Trending on 2 September 2026

Rising this week

Terms that build over weeks rather than spiking for a morning, counted across link aggregators, news, code repositories and preprints. The figure is the last seven days; the pace beside it compares those days with the three weeks before them.

• data centersRising · 140 · 3.9×
• open-weight modelsRising · 121 · 2.9×
• diesel pricesRising · 87 · 12.4×
• AI slopRising · 82 · 1.8×
• american airlinesRising · 80 · 16.4×
• vibe codingRising · 79 · 2.0×
• agentic commerceRising · 72 · 2.3×
• generative engine optimizationRising · 70 · 2.7×

All watched terms →Rising16Established17Quiet2

By source

What each source said on its own, before any crossing. A source that did not answer is recorded as unavailable with the reason — never quietly dropped, and never filled in from a neighbour.

Who is reading this

Audience counter in standby — no analytics token is connected yet; nothing here is estimated.

For machines

The same day, in the forms a machine reader can use. All of them are static files, built from the archive with the page, so a reader and a crawler never see two different states.

/trending/latest.json

the newest day in full — every source, every signal, the converging topics and the summary

/trending/2026-09-02.json

any committed day by its date, in the same shape, never edited after the morning it was written

/trending/index.json

the archive register — every committed day with its page, its JSON and what converged on it

/trending/feed.xml

RSS, one item per day, with the converging topics as its text

/trending/latest.md

the same day as Markdown, for readers that parse text better than markup

/trending/topics/

the slower half of the ledger — the terms watched over weeks, each with its own page

/trending/topics.json

the whole watchlist in one file: counts per platform and window, status, receipts, and what the discovery run proposed

/llms.txt

where this sits in the house, and what a retrieval agent may do with it

• Data: CC0 1.0
• Cadence: rebuilt daily around 06:40 UTC
• Cite: Frank Bültge, Common Ground — trending ledger, frankbueltge.de/trending/2026-09-02

Method

Every morning the pipeline reads 8 open sources and records what each one declares as trending, with that source’s own label, its own rank and its own count. Labels are reduced to tokens — case folded, diacritics stripped, hashtags split, stopwords dropped — and matched across sources by an exact token, by overlap above a committed threshold, or by one label being contained in another. A topic scores by the number of independent platforms carrying it and then by its rank within each of them, and it is shown under the label of the highest-priority source that carries it. Every day is committed as JSON and never edited afterwards, so first-seen and days-hot are read out of the archive rather than out of a memory.

Method sheet →

No language model writes anything on this page: every label is a source’s own label, every count a source’s own count; the cross-check is a token-matching rule disclosed on the method sheet.