Cleaning and normalizing
Each incoming item is normalized into a standard shape before anything else happens — title, publication date, URL, source, body text, firm, analysts, and tags all get mapped onto the same fields regardless of which firm or feed it came from. That's what makes content from very different sources searchable and filterable side by side.
Content older than a set cut-off is not re-processed, so historical items stay stable.
For content you upload yourself, we extract the text from the file or URL at the point of upload, and generate a title and summary from it.
Social posts are additionally checked for professional relevance and language.
De-duplicating
The same article often reaches us more than once — a feed re-publishes it, several sources carry it, or a firm revises it after publication. You should only ever see it once.
Repeat deliveries are recognised and ignored. They don't create a second card in your results.
Genuine revisions update the item you already have rather than appearing alongside it, so the version in ARchitect stays current.
Publication dates and firm names are kept consistent across sources, so content from one firm groups together even when sources format it differently.
This applies to the content ARInsights provides. Content you upload isn't currently checked against what's already in Premium Content, so it's possible to upload something we already hold — if you spot a duplicate you've added, you can easily remove it.
Enriching
Once an item is clean and confirmed new, we add the context that makes it useful:
Research and blogs — AI extracts the topics covered, the key takeaways, and sentiment. Analyst names from the source are matched against known analyst records.
Quotes — AI identifies the named speakers, their firms, and the specific quotes attributed to them, then matches each speaker to a known analyst.
LinkedIn and X posts — checked for professional relevance and matched to the analyst via their profile or handle.
Reviews — normalised into a consistent structure. Reviews do not go through AI enrichment or analyst matching.
Firm names are normalised at publication, so content from the same firm groups together even when sources spell it differently.
Everything published is then indexed for keyword search across titles, summaries, body text, and metadata — which is why you can find content that was never tagged with the right topic. Published items are also checked against your auto-updating collections and added to any they match, and notifications are sent to anyone subscribed.
Read part 1: What is included in Premium Content?