The Fact Checking platform provides an extension to the
Messenger Communication Platform, or short,
MCP. It's based on the same primitives: Incoming end user
messages are grouped in conversations, so-called tickets.
These tickets can be assigned to and handled by agents. An
agent, that's you and your team members.
Even when using Fact Checking, you do, of course, still have full access to MCP and all its features!
- How does it work?
- Getting Started
- Research Tasks
- Management API Reference
If you miss anything here or have suggestions or simply want to ask a human being, you can send us a message at the bottom of this page. All messages sent via that form will be read by the development team directly.
What the Fact Checking Platform provides on top of that is some new
primitives. As a fact checking team, your end users will have
questions or claims for you: Is this true? Did she really say that?
Is that image a fake?
As more end users become aware of your channel, more of these questions will appear, often the same in different words. You won't have an answer for each of them right away, probably, so you will need to look into the topics first. Here, the Fact Checking Platform steps in: You can create a new research for one or more end user messages. As you gather information about something, you can add notes to the research task, append a link to your website and discuss it with your colleagues.
The target at this stage is formulating an answer for your end users that covers the questions as completely as possible: The outcome should be reusable for other end users with the same question.
If you feel you have written a satisfactory answer, you can assign it
to someone else so they can review it.
This is not strictly necessary or enforced—assigning it to yourself is perfectly valid, too. The point of the review stage is taking a last look on the answer before dispatching it to your end users. If you or the assigned reviewer is happy with the answer, it will move to the last stage. Here, you can send it to every end user that had the same question!
Welcome to the team, glad to have you on board! This section contains a few ideas to get you up to speed with the Fact Checking Platform. First off, you'll probably want to enable the Slack integration. To do so, navigate to the settings page and click the Add to Slack button: Take me there.
If you haven't activated your own phone number yet, you're probably
in WhatsApp test mode. It allows you and several colleagues to test
MCP using a phone number provided by us. Simply
log into MCP
and follow the instructions on your dashboard.
You should now be able to send messages and see tickets being opened in MCP! Feel free to play around and explore MCP at this point—there's nothing you can break.
Now that everything is ready, you should be able to view all open
right here, on the Fact checking platform. This list is a
little different from the one in MCP, and with good reason: It's
intended to help in keeping an overview of incoming user questions.
You can search, sort and filter this list.
As end users write new messages, the tickets will be updated. This can sometimes take some minutes—that delay is caused by the nature of messaging apps, but nothing you should worry about.
If you click on a ticket in the list, you will be shown the main
ticket interface. It might look a little intimidating at first, but
provides you with all tools you need to quickly respond to the end
user. On the left side of the screen, you'll find the chat history.
It looks like a WhatsApp conversation, and it works in the same way
for the most part, too.
Next to the chat history, you'll find the available answers, ongoing research tasks this ticket might be part of, and a detailed meta-data analysis of the entire ticket. Read on to discover all parts of the ticket interface!
The chat history shows all messages in a ticket, in chronological order, with the newest at the bottom: Exactly what you're used to from messaging apps. Unless the ticket is closed, you will also see a chat input field you can use to send free-form messages to the end user.
Note that with WhatsApp, there's a catch here: You can only respond to end users in a 24-hour window after the last user initiated message, unless you are using a separately billed notification message.
Notifications are based on a template that may contain placeholders. When you send a notification, the placeholders can be replaced by actual values, say, the end user's name. If you have any questions on this topic and how templates will be used for fact checkers, please contact us directly.
If one or more messages contain a question that you don't currently know the answer to, you'll want to start a research for it (or add them to an existing research!). To do so, click the round checkbox left of the message to select it. As at least one message is selected, the selection actions will appear in the header bar of the chat history. Here, you can choose whether you want to add the messages to an existing research or create a new one.
The answers panel shows all answers we deemed as related to the
end user's inquiry. Now, how do we deem something as
related? Before addressing this question, we'll need to consider
what answers actually are: Finalized research tasks. As soon as a
research is successfully reviewed, its internal status changes
and it becomes available for usage in tickets.
As new tickets roll in, we automatically analyze their content:
- Are there any attachments in it?
- Does the message text contain any URLs?
- Does it contain certain keywords?
Depending on the answers to these questions, we compile a list of
answers that have been used for other tickets with similar
properties or research tasks that seem to match the topic.
This algorithm learns over time and adapts to the type of tickets you receive and how you typically respond.
Still, it's perfectly possible the answer you have in mind might not be shown or the results don't really match. In that case, you may use the top-right search bar to search for answers manually. The results will include any reviewed research that matches the search term.
Below each answer, you'll find two buttons:
Copy to chat and Automate.
Using the copy button, you can insert the answer into the chat
area: This gives you the opportunity to modify the answer if
necessary. The automate button allows you to configure an answer
to be sent automatically whenever the same image or link is
recognized—in any new conversations after sending—and
close the ticket, without any human interaction required.
This turned out to be tremendously helpful in traffic spikes for hot topics or news updates!
The research panel shows you all currently ongoing
research—that is, research that hasn't been reviewed
yet—any message from this ticket is included in.
This is intended to show you that you or one of your colleagues
is already working on a question, so you don't end up doing the
same work all over again!
Speaking of this situation: As soon as an individual message is being researched, a helpful reminder will be shown below the chat bubble in the chat history. Additionally, the only available chat action when such a message is selected is Show research.
The analysis panel shows all data we could extract from the message: Keywords, Links (URLs) and Media files. These three data types form the basis for all automated ranking decisions our platform makes. As more data comes in, the algorithms “learn“ and become more versatile.
The keyword section lists any prominent words we could extract from the message. This will disqualify any stop words—that is, words that are so common they occur in practically any sentence and don't carry any real meaning, such as the, is or at, for example. Additionally, numbers and short words with two or three characters are filtered out.
The links section shows previews for any links found in the
ticket: Thee previews contain any meta data we can find for the
link, a preview image if available and a summary. This works
similar to the way Facebook or Twitter show previews.
Before actually clicking a link, the platform will show you the full URL you are going to visit and expects confirmation before redirecting you. As links can be sent by anyone, they may contain viruses or distasteful content. Please be careful before visiting any user-supplied web pages!
The media section shows any files sent as attachments in the ticket. This works well for images, video or audio files as well as PDF documents—all of which can be shown directly in the browser. Simply click on media files to open the full file in a new browser tab.
Research tasks represent the workflow for fact checking.
Oh, snap! This section is not ready yet! We're still writing this one up. In the mean time, you can either explore the interface yourself or ask us directly.