Feature
From a paragraph to a live catalog.
Most bots start as a wall of prompt text and stay that way. Ask has three levels, and you can move up one at a time without rebuilding anything: instructions, entries, and a read-only API of your own.
The three levels
Level 1
Instructions
A paragraph or two: what you do, who you serve, what to say when you don't know. Enough for a small site, and it's what the crawl drafts for you on the first run.
Level 2
Entries
Structured records held in Ask - a service, a product, a policy, a page. Each has a title, a body, a category and typed attributes. This is where most sites end up.
Level 3
A connected API
Your own read-only endpoint, called during the answer for the things that change hourly: stock, price, availability. The bot never writes anything back.
How content gets in
- Crawl your site. The onboarding path. Ask reads your pages and proposes entries; you pick which ones to keep before any of it is live.
- Write one by hand. Usually straight from a case, where the entry editor opens pre-filled with the question the bot could not answer.
- Bulk import. Upload a file, preview what will change, and undo the whole import if it went wrong.
- A synced source. A sitemap or a feed, re-read on a schedule. Changes are held for review rather than applied silently, and a sync that fails tells you.
- MCP. Point your own AI tooling at your workspace's content.
Why typed attributes matter
An entry is not just text. Give it a price as a number and a "has photo" as a boolean, and the bot can answer "which of these is cheaper" and "do you have one with X" - the two question shapes that a text-only bot answers confidently and wrongly. Attributes are also what the analytics use to tell you which questions your catalog cannot answer at all.
Citations on every answer
When an answer comes from your content, it says which entries it used, and the citation is built by the server from what was actually retrieved. The bot does not get to write its own links - a model inventing a plausible URL is the most common way a chatbot embarrasses its owner.