Notes from Serochat
Honest, practical writing on building AI customer support that answers from your own docs — and knows when to hand off to a human.
Can an AI chatbot cite its sources?
Some can, and some only appear to. How a chatbot cites its sources, what verifies the footnote, and how to test the citations yourself in five minutes.
What is a grounded AI chatbot?
A grounded AI chatbot answers from a defined set of documents, cites its source, and admits when it cannot help. The definition, the mechanics, the limits.
How to train a chatbot on your own data
You probably should not train anything. The practical route to a chatbot that answers from your own data: choose documents, connect them, test, fill the gaps.
Why AI chatbots make things up
Hallucination is not a glitch. It is what a language model does when it has no source. Why it happens, what it costs, and what actually reduces it.
A small-business guide to AI customer support chatbots
What a small business should ask of an AI chatbot for customer support: where answers come from, what the pricing counts — and when not to buy at all.
Cut repeat support tickets without hiring more agents
Most repeat support tickets are a content problem, not a staffing one. A practical routine for finding them, fixing them at the source, and keeping them gone.
How much does an AI chatbot cost?
Ours is £49 or £149 a month, and the pricing model matters more than the number. The arithmetic, the overage rules, and the costs no price page lists.
Chatbot data privacy: the questions to ask
Who sees your documents and your visitors’ messages when you add a chatbot, which permission scopes matter, and the six questions to ask any vendor.
What is a RAG chatbot, minus the code
RAG means the bot looks things up before it answers. What retrieval-augmented generation does, what goes wrong with it, and what it is not — no code.
What ticket deflection actually measures
Ticket deflection counts the conversations that avoided a human. It cannot tell a customer helped from one abandoned — so here is what to check instead.
Signs you don’t need a chatbot yet
We sell one, so read this sceptically. Four situations where buying an AI chatbot now would be a waste, and what to do instead until they change.
Why 25 chosen pages beat a site crawl
Serochat indexes 25 pages you pick instead of crawling your site. Why a wider index makes worse answers, and how to choose the pages worth quoting.
Your customers don’t read the FAQ
They never read it, and they never will. The FAQ is usually right; the delivery is broken — and a doc-grounded bot is the documentation answering back.
What “grounded in your docs” really means
“Grounded” is not a buzzword. It is a concrete promise about where every answer comes from — and a way to check it.
Connect your docs in an afternoon, not a quarter
A practical look at getting a doc-grounded support agent live: what to connect first, and how to keep answers accurate as your content changes.
Writing help-centre articles your AI agent can actually use
A few small habits make your existing documentation far easier for an AI support agent to answer from accurately.
Why we built Serochat to say “I don’t know”
Most support bots bluff when they are unsure. We think a good answer includes admitting when there is not one — and handing the conversation to a human.
When should support hand off to a human?
A confident wrong answer costs more than a polite “let me get someone.” Here is how we think about the handoff line.
A 10-minute setup checklist for your support agent
The shortest path from sign-up to a working, doc-grounded agent on your site.
Why source links beat confidence scores
A percentage tells your customer nothing they can act on. A link to the source tells them everything.
Keeping answers fresh: a re-sync routine that works
Documentation changes. Here is a light habit for keeping your agent’s answers current without a heavy process.
Q&A pairs: the fastest fix for a wrong answer
When the agent gets something wrong, a short question-and-answer pair is often the quickest, most precise correction.
Measuring an AI support agent without vanity metrics
Deflection rate looks great on a slide and tells you almost nothing. Here is what we watch instead.
Match your support widget to your brand in two minutes
The chat widget should look like part of your site, not a bolt-on. Theming it takes about two minutes.
Connecting Google Docs, Notion, and your help centre
A practical tour of the sources you can connect, and what each one is best for.