A chatbot that only knows your documentation
Which is exactly the point. It answers from your published articles, shows the source, and passes the conversation to a human the moment it is out of its depth.
- Cited answers
- Clean handoff
- Deflection reporting
What is a documentation chatbot?
A documentation chatbot is a conversational interface to a knowledge base. Instead of searching and reading, the visitor asks a question and gets an answer assembled from published articles. In TheDocs the chatbot uses the same grounded retrieval as AI search, cites its sources, respects the reader's permissions, and hands the conversation to a human - with the transcript - whenever it cannot answer or the visitor asks for one.
- Answers only from your published articles, with citations.
- Escalates to a support form, live chat, email or a helpdesk ticket.
- Carries the transcript into the handoff so nobody repeats themselves.
- Embeds via script tag, product component, helpdesk app or API.
- Reports resolution rate, handoff rate and unanswered questions.
- Matches your brand colours, position and opening message.
A conversation, end to end
-
01
Visitor asks
In their own words, on your marketing site, inside your product, or from a help article they were already reading.
-
02
The bot grounds the answer
Retrieval runs against the articles this visitor is permitted to see, then the answer is written from those passages with sources attached.
-
03
It resolves or it escalates
If the answer is there, the visitor is done. If not, the bot says so and offers the handoff you configured - never a plausible-sounding guess.
-
04
You see what happened
Every conversation is logged with outcome. Unanswered questions cluster into a content backlog; resolved ones count towards deflection.
The parts most chatbot projects get wrong
Knowing when to stop
The failure mode that damages trust is not silence, it is a fluent wrong answer. The escalation path is designed first.
- A confidence floor below which the bot declines rather than answers - tunable per project.
- Topic exclusions: refuse to answer on pricing, legal or security topics and route them to a human.
- An always-visible 'talk to a person' control, so nobody is trapped in a loop.
- The handoff carries the full transcript, the articles consulted, and the visitor's context.
Embedding it where people already are
A widget that only exists on the help centre reaches the visitors who already found the help centre.
- Script tag: one line on any page, with configuration attributes for colour, position and opening message.
- In-product: a component that passes the signed-in user's context so answers can use their entitled documentation.
- Helpdesk: runs beside an agent as a suggestion panel, drafting a reply with citations for the agent to edit.
- API: build it into a channel we do not support natively - WhatsApp, Slack, an IVR.
The agent-assist mode is often the quickest win: agents answer faster and the articles they reach for tell you which ones matter.
Measuring deflection honestly
Deflection numbers are easy to inflate. These are defined so they survive scrutiny from a support director.
- Resolved: the visitor got an answer, did not escalate, and did not open a ticket within 24 hours.
- Escalated: the bot or the visitor chose a human.
- Abandoned: the visitor left mid-conversation - counted separately, never as a resolution.
- Deflection estimate is comparable against your pre-launch ticket baseline, not against a vendor benchmark.
Related features
Questions people ask before they start
Point a chatbot at your existing docs and see what it cannot answer.
Start free for 14 days. No credit card, no setup fee, and your content is yours to export at any time.
Questions first? Email sales@thedocs.in or call +91 8585953085.