Three and a half years of coverage, read end to end. Around twenty thousand articles across ten publications, from MIT Technology Review through the Financial Times, The Economist, The Guardian and the New York Times to the management and consulting view of Harvard Business Review, the MIT Sloan titles, McKinsey and BCG. What rose, what faded, and what reversed.
ChatGPT launched on 30th November 2022. Since then a wide span of titles has worked out what it meant: the technical press, the business press, the mainstream press, and the management journals. This page reads the whole record at once: when each idea arrived, which advice inflated and quietly died, and where on the media spectrum each idea lives. Use the switch to move between the four tiers and the individual publications.
Monthly AI article counts. A news story usually spikes and decays; this one is still climbing three and a half years on, with no peak in sight. Switch to indexed to set each title to 100 at ChatGPT's launch and compare growth rather than raw volume — on a log scale, because a title that started small (the New York Times began near zero) can grow many times over. The monthly journals carry far fewer articles than the dailies, so indexing is the fairer way to see who accelerated. Click any title in the key to add or remove it.
The date each name first appeared in each title (hollow rings mark the metadata-only titles: the New York Times and MIT Sloan Management Review). Where more than one carried it, a line joins them. The order of arrival is the early history of the field, told by its nouns. Click the key to add or remove titles; use the view switch at the top for tiers or individual publications.
Not products, arguments. Each panel tracks one idea's share of monthly AI coverage, smoothed over three months. The grey area is total attention; the coloured lines are the four media tiers (or, with the view switch, the individual publications). The badge marks its trajectory; the italic line beneath is its story. The key controls every panel below — click a tier or title to add or remove it. Discovered by reading the articles, not by guessing keywords.
Ideas and advice that inflated, peaked, and faded, or reversed outright. Each is plotted with its peak marked. This is the answer to "were there bubbles of advice?": yes, and here they are, dated.
For each title: the share of issue headlines mentioning AI, month by month since 2018. The vertical line is ChatGPT. Watch the pivots: flat, flat, flat, then vertical.
The recurring arguments of this email, each plotted against attention across the whole spectrum (the management and consulting tier — HBR, the MIT Sloan titles, McKinsey and BCG — is the one most likely to share this territory). Three kinds of territory: ideas the press is writing about now, ideas the press covered and moved on from, and ideas the press has barely touched: white space, the differentiated ground. The badge and the classification fold in every active title; click the key to focus on a tier or publication. Full text everywhere except the New York Times (headline and abstract only) and MIT Sloan Management Review (headline and teaser only).
What we charted, and what we left on the table. The summary counts every idea the analysis surfaced; the list below is just the ones that did not make a chart, too broad to mean anything, too sparse to draw, or set aside. Left in deliberately, as a reading list of where the AI conversation went that this page did not follow.
Press corpus, matched on full article text where available and arranged as a media spectrum across four tiers (technical, business news, mainstream news, and management & consulting): 1,146 from The Economist; 5,285 from the Financial Times; 484 from BCG; 4,787 from The Guardian; 811 from Harvard Business Review; 340 from MIT Sloan's Ideas Made to Matter; 961 from McKinsey; 1,985 from MIT Technology Review; 3,427 from The New York Times; 370 from MIT Sloan Management Review. Collected from paid or authorised subscriptions and open APIs for private research; monthly journals run at lower volume than the dailies by design. The view switch flips every chart between the four tiers and the individual publications. Newsletter corpus: issue headlines from the newsletter archaeology archive (28 titles), masthead stripped before matching. Ideas in sections III, IV and VI were discovered by a language model reading stratified samples of each corpus, then validated by counting pattern matches across every article; concepts whose patterns matched implausibly broadly or too narrowly to chart were set aside and are listed in the cutting-room floor. Counts are shares of monthly AI coverage, smoothed over three months. Keyword matching undercounts ideas expressed obliquely. Newsletter titles were audited individually: titles with unrecovered or gap-ridden archives (Import AI, TLDR AI) are excluded, and titles without a meaningful pre-ChatGPT sample show their launch date instead of a baseline. The dashed titles are matched on metadata rather than full text — the New York Times on headline and abstract (its API returns no article bodies), MIT Sloan Management Review on headline and teaser (full text is subscription-gated) — so their matches are weaker evidence than the full-text titles; McKinsey and BCG are smaller, still-growing corpora. Status: Raw AI output. Not yet CEO'd (Checked, Edited and Owned).
This is an artefact created for David's weekly email. See the others here.