Answers your customers can find without asking.
A help centre they can search, article suggestions inside the chat, and an AI Agent that answers from exactly the same articles — and tells them which one it used.
- Your branding
- Custom domain
- Trains the AI Agent
Write it once. It answers in four places.
Most teams keep the refund policy in a help centre, a chatbot script, a macro and somebody's head — and three of those are out of date. In Chatdrill there is one article, and everything else reads from it.
- No separate bot training. Publishing the article is the training.
- Fix an answer by editing a paragraph — the correction reaches the AI, the help centre and the widget at once.
- Every AI answer cites its article, so you can always trace a bad answer back to the text that caused it.
- Unpublish and it disappears everywhere, including from the AI's sources.
One article — “Returns & Refunds”
Published · English, German, French · updated 2 days ago
The AI Agent
Answers around the clock, quoting the article and linking to it. No separate training step.
Your public help centre
Searchable, categorised and branded as yours — on your own domain if you want one.
Suggestions inside the chat
The widget offers the article as somebody types, before the question ever reaches a person.
Your agents' replies
Search the same articles from the composer and drop the relevant paragraph into an answer.
How can we help?
Getting started
12 articles
- Create your account
- Install the widget
- Invite your team
Orders & shipping
24 articles
- Track an order
- Change a delivery address
- Shipping times by country
Billing
18 articles
- Returns & refunds
- Download an invoice
- Change your plan
A help centre that looks like the rest of your website.
Customers should not be able to tell where your site ends and your help centre begins. Yours runs on your domain, in your colours, with your navigation.
- Your domain — point a CNAME at it and the help centre lives on your subdomain.
- Your branding: logo, colours, fonts, header and footer links, and custom CSS if you need the last five percent.
- Findable by search engines with clean URLs, meta descriptions and a sitemap.
- The chat widget on every page, so the self-service route always has a human exit.
Built for people who write articles between answering chats.
The editor gets out of the way, the structure is obvious, and nothing goes live until you say so.
Write it like a document
Headings, lists, tables, images, callouts, code blocks and links. Paste from a doc and the formatting survives.
Structure people can follow
Group articles into collections, order them by hand, and give each one a short description that reads like a shelf label.
Finds it by the wrong words too
Matches on titles, body text and keywords you add, so a customer searching “money back” still lands on “Returns”.
One article, many languages
Link translations to the same article so the help centre and the widget serve the visitor's language automatically.
Written for your team only
Escalation paths, refund limits and the things you would never publish. Visible in the inbox, never in the help centre.
Nothing publishes by accident
Keep an article in draft, share it for review, and publish when it is right. Every version is kept.
The answer arrives before the question is sent.
As a visitor types, the widget looks for a matching article and offers it. Plenty of people read it and close the window — which is the cheapest resolution there is.
- Suggested as they type, matched on the question rather than on a keyword list you maintain.
- Read without leaving the chat — the article opens inside the widget, and the conversation is still there behind it.
- “Still need help” is always visible. Self-service is an offer, never a wall.
- Agents search the same library from the composer and paste the paragraph that answers it.
The most valuable report is what people searched for and never found.
Views tell you which articles are working. Failed searches tell you which articles do not exist yet — and they arrive already sorted by how many people wanted them.
Suggested outline. Pro-rata rules, when access ends, and what happens to the data afterwards.
84 people looked for this and found nothing, so 84 people either opened a chat or left annoyed. Chatdrill drafts the article outline from the conversations that followed those searches, and once you publish it, the AI Agent starts answering from it — no separate bot training step.
Suggested outline. Where invoices live, how far back they go, and how to add a VAT number retroactively.
Billing questions cluster at the start of a financial year, so this one arrives as a spike rather than a trickle. Publishing it once removes a seasonal queue you would otherwise staff for every January.
Suggested outline. A direct yes or no near the top, rather than a pricing page that implies it.
A weak match is worse than no match: the search returns something, the reader skims it, and they still open a chat. The report shows which article was returned and how many readers voted it unhelpful, so you know exactly which paragraph to rewrite.
Suggested outline. The exact steps, the retention window, and what deletion does not cover.
Deletion questions are the ones customers escalate fastest when they cannot self-serve, and in several jurisdictions the answer is a compliance obligation rather than a nicety. This one is worth writing the week it appears.
Suggested outline. What is cached, what queues, and what genuinely needs a connection.
The existing article answers a neighbouring question, which is why it ranks at all. Adding the specific wording people searched for — not a keyword list, the actual phrasing — is usually a five-minute edit rather than a new article.
Open any row to see what Chatdrill does with a failed search. In the product these become article drafts you can edit and publish — start free and the report fills itself from your own traffic.
Handoffs become a writing list
Every question the AI could not answer is grouped by topic. That list is the fastest route to a knowledge base that covers what customers actually ask.
Helpful votes with context
Readers can say whether an article helped and add a sentence explaining why not. Low scores show up next to the search terms that led people there.
Everything else the knowledge base does.
Custom domain
Serve the help centre from help.yourcompany.com with your logo, colours and typography.
Article performance
Views, helpful votes, and how many conversations each article prevented.
Questions about running a help centre.
No. The AI Agent can also learn from your website and uploaded files. But articles give you control the other sources cannot: you decide the exact wording, you can correct an answer by editing one paragraph, and you can see which article an answer came from.
Publishing re-indexes the article, and the AI answers from the new version shortly afterwards — usually within a few minutes. If you need to stop an answer immediately, unpublishing the article removes it from the AI's sources as well as the help centre.
Yes. Internal articles are visible to your team inside the inbox and can be used by the AI when it drafts a reply for an agent, but they are never published and never quoted directly to a customer. Refund limits and escalation rules usually live there.
You write a version per language and link them as translations of the same article. The help centre serves the visitor's language and offers a switcher, and the widget suggests the right version. The AI Agent can also answer in the visitor's language even when only one translation exists.
Yes. Point a CNAME at your help centre and it runs on your subdomain with your logo, colours, fonts, footer links and custom CSS if you need it. It should look like part of your website, because to a customer it is.
Yes. Import from HTML or Markdown files, or create articles through the API if your documentation already lives in a repository. Categories and internal links are preserved, and you can review everything as drafts before anything goes live.
Write the answer once. Let it work everywhere.
A help centre, in-chat suggestions and an AI Agent that all read from the same articles.
No credit card required.