Smarter intranet search: when your Govintra WordPress theme gets a brain
I’ve built many intranets — large government departments, small charities, and everything in between. And one problem always comes back: search that doesn’t really help.
You type, hit Enter, and get a list of results. Maybe one of them is right. That’s not intelligent — that’s just fast indexing.
So I decided to go further.
I gave my Govintranet installation a brain.
Why an intranet needs a brain
Whether you’re a major public-sector department or a two-person digital team, your intranet’s greatest value lies in the “how-do-I?” guidance — the everyday answers your people need fast.
But even a perfect article library is only useful if staff can find the right piece at the right moment. Traditional search can’t interpret meaning; it only matches words.
As proof of concept, I developed a new custom plugin that works with AI and Govintranet — it lets your intranet understand your question and bring back the most relevant answers, written in plain language, complete with links to the original content.
How it works
The proof of concept plugin combines three technologies:
- Govintranet: content is initially encoded and indexed in bulk, then future additions, amendments and removals are processed in realtime.
-
Pinecone: a secure vector database that stores tiny mathematical representations of your intranet’s text content with options to store in Amazon or Google cloud.
-
OpenAI’s ChatGPT: a language model that interprets a user’s question, searches Pinecone for context, and generates a coherent answer with source links.
Each task, page, or event on your intranet is broken into small chunks, converted to embeddings, and indexed in Pinecone. When a user asks a question, the system compares it against those embeddings, fetches the closest matches, and feeds them into ChatGPT — which replies with a short, clear explanation and references.
It’s still your intranet content — just smarter.
You can see a working proof of concept based on just guidance content here:
👉 https://demo.govintra.net/how-do-i/
Handling shortcodes and structured content
One of the biggest challenges with intranet content is that guidance pages often rely on embedded tables, process lists, and data grids — especially when using TablePress for structured information such as chart of accounts, salary scales and other tabular data.
I built the indexing engine to recognise and safely render specific shortcodes — like TablePress tables — during the embedding process. This means the system expands table data into readable text, preserving structure using tabs and line breaks so the embeddings accurately represent what users actually see on the page.
In short, the assistant can understand and retrieve answers from content stored inside tables — something traditional search would miss entirely.
For example, if your list of departmental response rates for the annual people survey sits inside a TablePress shortcode, the AI assistant can read, embed, and respond using that information directly.
A system that doesn’t make things up
AI models are powerful, but they can also be overconfident — producing answers that sound right but aren’t based on real data.
To prevent that, the Govintra AI assistant uses a strict system instruction that tells the model exactly how to behave.
We explicitly tell it:
“Use only the context provided. If the answer isn’t in the context, say you don’t know.”
This ensures that all responses come only from verified intranet content, not from general internet knowledge or assumptions. It’s a safeguard against misinformation — essential when staff rely on the intranet for policy, HR, and operational guidance.
We can also influence the tone and style of the replies.
For example, we can make the assistant:
-
sound formal and procedural for HR or compliance topics,
-
use friendly, conversational phrasing for staff FAQs,
-
or even localise tone and idioms for multilingual sites.
This makes the assistant feel native to the organisation — not like an external chatbot dropped into the site.
Built for GovIntra’s content model
Govintra’s task hierarchy (parent steps and subpages) is indexed as a single knowledge unit.
That means the assistant understands full processes rather than isolated steps.
We’ve extended the logic for pages and events, adding event metadata (start/end dates, times) so users can ask things like:
“When’s the next content editors’ workshop?”
“Show me the next training courses for new employees”
“Tell me the history of the department”
and get direct, linked answers from verified content.
Features we’re testing
-
Automatic language detection — detects the language of the question and replies accordingly (because English is not always a first language).
-
Keyword and taxonomy metadata — uses categories, tags, and keywords to refine matches.
- Role-based answers — context is filtered by role/team to give more personalised answers
-
Recency weighting — for time-sensitive news or updates, newer content can rank higher.
-
Transparent sources — every answer lists the exact intranet posts used, improving trust and auditability.
- Processing documents — even the best intranets rely on PDF and Word documents, spreadsheets and presentations, so these could be indexed too.
A bespoke proof of concept
This plugin is a bespoke development by Agento Digital, built as a proof of concept to explore how intranet content can be indexed and queried through vector search and generative AI.
It currently integrates with Pinecone and OpenAI’s ChatGPT APIs. Further work is planned to make it compatible with other vector store providers (for example AWS OpenSearch Serverless, Weaviate, or Azure Cognitive Search) and to offer configurable LLM processing pipelines.
If you are interested in working with Agento Digital to give your intranet a brain, please get in touch:
Why this matters
Intranet search has been stuck for years.
AI doesn’t just make it faster — it makes it smarter.
-
Staff get answers, not a list of links.
-
Editors can write once and let the system re-use the knowledge contextually.
-
Organisations can scale knowledge-sharing without reinventing FAQ pages.
This new plugin demonstrates how turning the intranet from a static reference into an active participant — a conversational knowledge assistant built on your own trusted content. As I’ve always said: content is king. This plugin works with existing Govintranet content; tasks and guides, pages and events. No changes are needed to the way you publish content.
The future
Our aim is not to replace editors or experts — but to remove friction.
The result?
An intranet that doesn’t just file knowledge away — it uses it.
Govintra can work with AI and your intranet can have a brain.