A communications office, hour by hour

One office’s day, from the morning briefing to the report in the evening. It watches the topics, people and accounts it names, state by state, in twelve Indian languages.

Media Verdict console

The questions the office is asked

  1. 01

    What is being said, and where?

    The topics rising now, per platform and per state, each measured against its own usual level.

  2. 02

    Will it be bigger in an hour?

    Rising, about the same or falling for each busy topic, with a range and the reasons.

  3. 03

    Is it for the government or against it?

    Whether posts are for, against or neutral toward the government, and how strongly, with each post weighted by its reach.

  4. 04

    Is it real interest or a push?

    Whether accounts are acting together within seconds of each other, with the accounts and the evidence.

  5. 05

    Is this video doctored?

    A verdict on the image, video or audio: what was found, how certain, where in it, and where it first appeared. Media Verdict never certifies media as real.

  6. 06

    What do AI assistants tell people about it?

    What ChatGPT, Gemini, Google AI Mode and Perplexity answer about a keyword, and which websites they cite.

What it uses

CapabilityWhat it gives the office
Social listeningThe accounts, keywords and hashtags the office names, across platforms and twelve languages, each item counted once however often it is copied.
Trend detectionTopics rising now, as a sudden spike or a rise held over a day, with the states their posts come from. A watch list can be set to a state, a district or a constituency.
ForecastsFor each busy topic, whether it is likely to rise, stay about the same or fall in the next hour, and why, so the office can see which topic to prepare for.
StanceWhether posts are for, against or neutral toward the government, and how strongly, in each language.
Coordinated activityBefore the office answers a surge, whether it is ordinary interest or accounts acting together, shown with the accounts and the evidence.
Media forensicsA verdict on an image, video or audio clip, with where in it and where it first appeared. A clip forwarded on WhatsApp can be submitted as a file.
AI answersWhat AI assistants tell people who ask about the office’s keywords, and the websites they cite, so the office can see which sources the answers rest on.
Approved responseReplies, posts and platform reports from the office’s own official accounts, sent only after a second person approves them.

Election and other busy periods

The office can declare a busy period, such as the phases of an election. Three things change while one lasts.

1. Volume. The system is designed for bursts of 500,000 posts an hour, though that size has not yet been load-tested at full scale. When more media arrives than can be examined at once, the items reaching the most people are examined first.

2. Analysts. The staffing agreed for an installation is two analysts in ordinary weeks and four during a declared period, with one approver and one policy owner throughout.

3. On call. During a declared period someone is on call and acknowledges within 15 minutes. Outside one, support answers by the next business hour.

Media Verdict console
1 2 3
Illustration

Stored in India, on your servers if you wish

All data, logs and reports are stored in India: collected posts and media, the records built from them, evidence packs, reports and the audit trail.

The same build runs as the office’s own installation, on its own servers in India, with a database no other client uses. Who runs it day to day is agreed with the office.

Some analysis uses services that run outside India, for example the language model that writes the headline paragraph of a report. What they return is stored in India with everything else.

Users sign in through the office’s own identity provider. Approvers and admins sign in with two factors, and every write records who made it.

Drafted by one person, approved by another

Nothing goes out from the office’s official accounts without an approver who is not the drafter.

1. A reply, comment, message, post or platform report is drafted by an analyst, or proposed by a rule and then edited by a person.

2. An approver who holds the approver role, signs in with two factors and did not draft or edit it approves it. A drafter who tries to approve their own draft is refused, and whoever confirmed a verdict cannot approve an action that cites it.

3. Only then is it sent, from the official account named on it, to one target. The draft, the approval and the send are kept with who did each and when.

A draft is due in the approver’s queue within 60 minutes of the firing or the request, a target set because the Election Commission’s takedown window for platforms is three hours.

Media Verdict console
1 2 3
Illustration

Questions

Toward whom is stance read?

Toward the government, and toward any other body the office names, each with the names and other words people use for it in each language.

Can we follow one state or one constituency?

Yes. A watch list can be set to a state, a district or a parliamentary or assembly constituency, and each topic shows the states its posts come from, with the share that could not be placed.

Can staff send in a clip forwarded on WhatsApp?

Yes, as a single file. WhatsApp itself is not read: a campaign organised there is seen only when it reaches public platforms.

Who can see our watch list and results?

Only your own users, signing in through your identity provider. Each user has an audit trail of their own, and no other client sees any of your work.

Bring a day your office has already had, and see where each part of it would appear.

Request a demo