How we work

Where stories come from

We read the public feeds of the marketing trade press in India and abroad: Social Samosa, afaqs!, BestMediaInfo, Marketing Mind, ET BrandEquity, MediaNews4u, MediaNama, Passionate in Marketing, Adweek, Digiday, AdExchanger, Marketing Week, Marketing Dive and Social Media Today. Reports that cover the same event are grouped into one story; a press release printed by several sites counts as one report. Every report is linked on the story page, grouped by outlet, so you can always read the original. We would like to read exchange4media, Campaign India, Adgully, Storyboard18, Ad Age and The Drum too; their sites currently block automated readers or publish no feed, and we check again regularly.

Drafted by AI, checked against the sources

Every headline, summary and the “What happened / Why it matters / The numbers” blocks are drafted by an AI model (gpt-oss-120b, an open-weight model from OpenAI, run on Groq) from the outlets' headlines and the text they publish in their feeds: a short excerpt for most, the full article for the few feeds that carry it. The model reads up to 12 reports, newest first, the first 1,500 characters of each and about 6,000 in all. The same model then reads the draft against those same reports a second time, claim by claim: every number, name, date, role and quote is marked as stated by a report, attributed to the one outlet that reported it, or unsupported. One unsupported claim and the draft is rejected, not published. A draft that passes that check and our house-style check goes live automatically, marked as an AI summary of the reports it was checked against, with no byline. Drafts about a sensitive matter (a lawsuit, an allegation, layoffs, a death) and drafts the checks flag are not published automatically: they wait for a named editor, who reads them beside their sources and approves them under their own name. An editor can correct, hold or withdraw any live story at any time, and a story an editor has reviewed carries their name. Search engines are asked to index only stories an editor has reviewed; the rest stay readable here and in our feeds.

A story's “The other reading” block, where it appears, is also AI-drafted: the strongest counter-view in the same reports, checked against them like the rest, and never a view of ours.

Written only from the evidence

• No facts, numbers, names or quotes that are not in the sources.
• In our own words, never a source's sentences.
• When only one outlet has reported something, we say so and attribute it.
• Plain, specific headlines. No clickbait.
• One house style: who did what in the first sentence, Indian English, rupees in lakh and crore, outlets named as they name themselves, and no press-release filler. A draft that breaks it is sent back once, then dropped.

Type-of-work labels

Anything that is not a straight report is labelled on its card and at the top of the story: Analysis (an outlet's explainer or wrap), Opinion (a column or a quoted view) or Sponsored (the source reads as paid or partner content). An unlabelled story is news. Sponsored stories are never ranked as Must know and never appear in the news feeds, lists, tables or counts; they still appear, labelled, on the pages of the names they mention.

Sensitive stories get a headline and links only

Legal disputes and lawsuits involving named people or companies, allegations against named people, layoffs and job cuts, regulatory penalties, and deaths: we show only the headline and links to the original reporting, and never a summary. The model decides, and a check in code reads the sources' own words and overrules it when they say so plainly. When that code check misreads a metaphor (“an assault on linear TV”), a named editor who has read the sources may ask for the full draft; the decision carries their name, is logged, and is stated on the story page. The model's own judgement is never overruled.

Stories written by an editor

Occasionally an editor writes a story at the desk from links they supply, with no AI model involved in the text. It goes through the same queue and is approved like any draft; its page says who wrote it and from which reports.

Developing stories

When a second outlet confirms a single-source story, or coverage doubles, the AI drafts a fresher version under the same rules and the same check, and an editor decides whether to apply it. Nothing on a live story changes until an editor applies it; when they do, the story says what changed and why. If a new report names a lawsuit, an allegation, job losses or a death, the AI leaves the text alone and tells the desk. Once an editor has changed a story, the AI never drafts a rewrite of it.

How stories are ranked

Each story gets a score from how important the model rates it for people who work in marketing in India, how many independent outlets reported it, how established the best of them is, how fresh it is, and what kind of story it is. The highest become “Must know”, then “Worth knowing” and “Also happening”. A Must-know story needs two independent outlets, or one established outlet reporting something clearly important. India-first: stories with no India angle are scaled down. Must know is the same for everyone and is never personalised.

Brands, people and the tables

The brands, agencies, people and platforms named in a story are extracted by the model, only from names in the sources, and become pages that collect every story about them. Appointments, campaign launches and account wins become rows in the people-moves ledger and the campaign log, each linked to its story; the editor sees those rows beside the draft and removes any that are wrong before the story goes live.

Images and video

We never use publishers' photographs. Share images are our own. When a source links the brand's film on YouTube, we embed the official video, loaded only when you press play. Otherwise a story carries a representative photograph: of the company it is about when its Wikipedia article has one, else of the story's subject (a motorcycle, a TV studio, a grocery shelf). We use only public-domain and CC0 photographs from Wikimedia Commons, which their authors released for any use, and label them as representative: they are not pictures of the event the story reports.

Accountability

Every automated decision (grouping, section, ranking, the check) and every editor's change is logged. Corrections are marked on the story, every change after publication is listed under it, a withdrawn story's page says so and why, and all of it is listed on our corrections page.

Until our inbox is live, send messages through the Instagram page.