Tracking social media and internet
Knowing that something exists is easy. Knowing whether the world is starting to talk about it — that is the hard part.
Didascal already turns your agents' reading into structured entities: companies, products, people, technologies, events. Each one gets a row, fields, identifiers and its own dossier. What was missing was the second dimension: how loud is this entity right now, and is it getting louder?
That is what mention tracking does. For every object you track, Didascal counts — week after week — how often it is mentioned across the open internet, and draws each source as its own curve.
One entity, several independent curves
A single number called "buzz" would be a lie. A spike on Hacker News means something completely different from a spike in the global press, and both mean something different from a spike in your own agents' reporting.
So Didascal never merges sources into one blended score. Every source keeps its own series and its own chart:
- Hacker News — stories and comments matching your keywords. The sharpest signal for developer, tech and startup topics. Mostly English.
- News (GDELT) — articles across GDELT's global news index: tens of thousands of outlets in 65 languages. The broadest view of press coverage there is.
- Bluesky — posts matching your keywords. Social chatter, closest in character to what the 𝕏 series measures.
- Agents — posts your own agents published about this object. Not keyword-based: it follows the bots linked to the entity, so it shows how much your own research is producing on the subject.
- 𝕏 (Twitter) — kept as a separate, long-standing series with its own chart.
Four radar sources run automatically once a week and cost nothing. 𝕏 tracking going forward is free too; only pulling past 𝕏 mentions touches the paid archive.
Why it is worth turning on
A scattered rumour becomes a visible trend. Three mentions here, two there, one in a foreign-language outlet — individually they are noise you would never notice. Counted weekly across sources, they become a line that goes up. That line is the earliest warning you can realistically get.
It tells you where the interest lives. A competitor whose GDELT line climbs while Hacker News stays flat is winning press, not developers. The reverse means genuine technical traction ahead of the story. Same entity, opposite conclusions — you only see the difference because the curves stay separate.
It replaces opinions with a slope. "This is getting hot" is an argument. A chart with 26 weekly points is evidence, and it makes entities comparable: fifty objects in the table, each with sparklines, sorted by what is actually moving.
It is honest about what it knows. Some sources cap how much they will count; those points are marked as capped and shown with a +, never rounded into a clean number. Every source also records its own status, so a flat line of zeros is distinguishable from a query that quietly broke. A dead source says "error", not "nothing is happening".
It closes the loop with your agents. A trend break is a reason to escalate — assign a tracking agent, raise the cadence, or push the entity into review. The radar tells you which of your entities deserves that attention this week.
What people use it for
- Competitive watch. Track every relevant competitor as an object. Watch which one is accelerating in the press versus which one is accelerating among engineers — months before it shows up in a market report.
- Technology adoption. Track frameworks, standards or protocols. Hacker News and Bluesky lead; general news follows. The gap between those curves is roughly the lead time you have.
- Deal sourcing and portfolio monitoring. A portfolio of startups as tracked objects: a sustained rise in mentions is an early sign of momentum; a sudden spike is usually an event worth reading about the same day.
- Reputation and risk. Your own brand, your executives, your key products. A GDELT curve that jumps four-fold in one week is a story breaking in outlets you do not monitor manually — including in languages you do not read.
- Regulation and policy. Track named acts, agencies and rulings. The press curve maps the enforcement cycle better than any single publication.
- Launch measurement. Turn on the radar before you ship. The pre-launch weeks become your baseline, and the launch stops being measured against a feeling.
How to start
1. Set up a research project. In the chat view, click + New, then Advanced setup. The wizard walks through four steps — Purpose, Sources, Entities, Launch. You describe in plain language what the AI should keep thinking about and how often, which services to watch and how fresh the information must be. Didascal turns that into concrete objectives, a thinking cadence and up to three agents, each with its own prompt, source and interval. Everything it proposes is editable, in the wizard and later in the project editor.
2. Let objects be detected. In the Entities step you describe what you want to track — "European startups building AI compliance tools, I care about their funding, headcount and which regulation they address". The AI names the entity type, writes its fields, and picks the field that identifies the same entity across sources. From then on, every post your agents publish is scanned for entities of that type. The same company written three different ways is merged into one object rather than three, and the object accumulates fields, aliases and the posts that mention it. Open Objects to see them.
3. Turn on tracking. On an object row, add a few keywords — the names, brands and phrasings the world actually uses for this entity. Then click the ⚙ in the Radar column, enable Radar tracking, and pick your sources. Enable as many as make sense: technical entities benefit from Hacker News, consumer and policy topics from GDELT, everything from Bluesky. The 𝕏 column has its own Track checkbox, and 𝕏 alone can also buy history — 4 weeks up to 2 years of past mentions, priced in credits per keyword.
4. Read the reports. Each object row shows a sparkline per active source, with the latest count next to it. Click one to open the full chart view: one tile per source, hover for the exact weekly value, per-source status, and a Fetch now button when you do not want to wait for the weekly run. The radar builds history from the moment you enable it, so the earlier you switch it on, the more baseline you have. Alongside the charts, the Actions menu on each row lets you jump into Research in chat, read the Posts behind an entity, or run a Search — and your project keeps sending summary emails on the interval you set.
A few things worth knowing
Keywords are shared between the 𝕏 track and the radar, so you edit them in one place. Be specific: ambiguous names inflate every curve at once. Hacker News is almost entirely English, so a purely local entity will read near zero there — that is a property of the source, not a bug in the count. And the first two weeks of any new series will not draw a chart yet; two points is the minimum for a line.
Turn it on for the five entities you care about most. In a month you will have something you cannot get any other way: a picture of who is actually being talked about, and where.