- New
Visual Insights: pictures are now data
Visual Insights no longer just shows the pictures from your feed — it reads them. See how images spread between sources, what's inside them, which ones look AI-generated, what the screenshots say, and where the picture contradicts the caption.
Up to a third of what your theme collects is pictures, stories and videos — and until now they were invisible to analytics. The new Visual Insights page treats every image as a data source. It works with the same date range and Filters as the Materials list.
Visual stream and mood
Seven numbers at the top, each a click away from the matching pictures: images, videos, reused images, text-in-image, AI-looking, recurring faces and tone mismatches. The chart below shows how many pictures arrived per day and what tone the picture itself carries — alarming, solemn, neutral or positive — independently of the caption.
Image cloud and facets
The top pictures sized by the audience that saw them, and four facets you can drill into: format (photo, screenshot, meme, infographic, poster, document…), what's in the picture (police, crowd, fire, accident, official meeting…), visual tone and network.
Propagation: who published first, and how far it went
Identical and near-identical pictures are grouped automatically. For each group you see the origin (first publisher), how many copies appeared across how many sources and networks, the spread in hours, the speed in copies per hour, and how many copies were edited (cropped, watermarked, re-captioned). This is how a rumour or a fake travels — text monitoring never sees it.
Text layer: what the screenshots say
Many channels post nothing but screenshots, scanned letters and banners. Text is now recognised inside pictures, summarised as keywords and listed with the source and reach — including a "text-only images" bucket.
Authenticity and brands
Visual share of voice: how often your theme keywords appear as a logo in pictures. Watermarks and handles printed on pictures reveal where a picture really came from. AI-looking pictures flags images that look generated, rendered or manipulated — review them before reacting.
People and places
How many people are in the pictures (none, one, a few, a crowd), recurring faces — the same person across several materials and sources (visual similarity only, no identification) — plus recognisable landmarks and the language of signage.
Video
Videos with transcripts, on-screen text recognised from keyframes, the tone of the keyframes, and the most-viewed videos of the period.
Gallery and picture details
Every facet, cluster or face filters the gallery. Opening a picture shows everything we know about it: what's in it, format, tone, AI likelihood, logo presence, script and language of the text, watermark, reuse count, palette, recognised text and transcript.
Analysis runs automatically for new materials. For older periods press Analyze all — the page fills in as the pictures are processed.