Two years after the CrowdTangle shutdown, AI monitors fill part of the gap

The CrowdTangle shutdown on August 14, 2024 split one group of users into two. Researchers with an academic or nonprofit affiliation got a path to Meta Content Library. Almost everyone else, including many reporters and fact-checkers, got nothing from Meta. Two years on, part of that gap is being filled by a different kind of tool: AI monitors that read public posts and decide for themselves what matches a question. This piece looks at what the shutdown changed, what the AI approach does well, and where it still falls short.

What Meta replaced, and what it did not

Meta's transparency site now says only that CrowdTangle is no longer available, and points to Meta Content Library and the Content Library API. Those tools are free, with no fees for access or computation, and they cover public Facebook Pages, groups and events, Instagram business and creator accounts, and Threads profiles with 100 or more followers.

Access is the structural change. Applicants must be affiliated with an academic institution or a nonprofit whose main purpose is scientific or public interest research. They apply through Meta Research Tools Manager, and the Secure Data Access Center reviews each application independently. CrowdTangle, by contrast, had worked with approved partners such as newsrooms. The new arrangement serves research well and daily newsroom monitoring poorly.

Two designs for watching public conversation

CrowdTangle followed a design common to listening tools: build lists of accounts or keyword searches, collect everything they produce, and sort it by engagement. The human does the judging.

AI monitors flip the order. The user writes a rule in plain language, and a model reads each new item and decides whether it fits. AyeWatch, one of the tools that positions itself as a CrowdTangle alternative, is a clear example. A monitor can be as short as "claims that the city water supply is unsafe, on Instagram, X and TikTok", with a second sentence such as "alert on new claims and official corrections, skip reposts of claims already covered".

That second sentence is the interesting part from an AI point of view. Keyword systems cannot easily express "skip reposts of claims already covered", because the words in a repost are the same as in the original. A model judging meaning can at least attempt it.

What the AI design gains

Less sorting. Because the model filters before alerting, the user sees fewer, more relevant items. Each AyeWatch alert also includes a short AI-written summary with its sources, which changes the job from reading a feed to checking a claim.

Wider coverage than Meta-only tools. AyeWatch watches news sites, blogs, forums, government pages and academic publications, which it counts at more than a billion sources, plus specific web pages and accounts or searches on X, Reddit, YouTube, LinkedIn, Instagram and TikTok. CrowdTangle was centred on Meta's platforms.

Open access. There is no application. A Free Preview gives three runs across up to three topics with no card, Pro costs $9 a month with hourly checks, and Pro+ at $19 a month adds an ASAP mode for breaking stories.

Machine-readable output. Alerts can go to Slack, Discord, email or a signed webhook, and an MCP endpoint lets AI assistants list and manage monitors. That makes the monitor a component in a larger newsroom or research workflow, not just a dashboard.

What it still cannot do

The limits are real and worth stating plainly.

• No Facebook. AyeWatch does not cover the platform CrowdTangle was best known for.
• No engagement data. It does not report reach, likes or shares, so it cannot answer "how far did this spread", only "did this appear".
• No archive. It reports new items from the moment a monitor starts, rather than offering historical exports.
• Judgment is delegated. A meaning-based filter decides what you never see. Users should spot-check by reading the raw platform now and then, especially early on.

Enterprise tools such as NewsWhip, which combines web and social feeds with engagement data and prediction, cover the engagement question, but they are sold through demos with no public price.

Where this leaves public-interest monitoring

The CrowdTangle shutdown did not end public content tracking. It divided it. Researchers who qualify have Meta's own data. Large organisations can buy engagement analytics. Everyone else now relies on tools that trade engagement numbers for broad coverage and AI judgment. For many newsroom tasks, such as catching a new false claim early or seeing an official correction as soon as it is posted, that trade works. AyeWatch's own comparison with CrowdTangle is direct about it: it covers the alerting job, not Facebook data or historical exports.