Answers from your own material
A chatbot that tells you when it failed - and gets better because of it.
Most bots answer what they can and quietly drop the rest. Ask flags the conversation instead: the question it could not answer, the visitor who got annoyed, the person who asked to be called back. Each one lands in an inbox as a case, and each one is a thing you can fix.
No card. 200 answers a month on the free plan, and every feature works.
What it looks like on your site
Three businesses, the same machinery. Pick the one that looks like yours - the example changes, the product does not.
The conversation
Do you do same-day callouts in Bristol, and roughly what does it cost?
Same-day callouts cover Bristol and the surrounding postcodes when you book before 11am. The callout is charged at the standard rate plus parts; the exact figure depends on the job, so the team confirms it before starting.
Every answer is built from your own material and cites what it used.
What it knows
Your service list, your coverage area and your pricing policy - crawled from your own pages on the first run, then corrected by hand where the site was vague.
A paragraph of instructions, entries in Ask, or a read-only API of yours. Most people start with a crawl of their own site.
The case it opened
"Do you take on commercial contracts?" - nothing in your material says either way, so the bot could not answer and flagged it. That is one entry away from being fixed.
Not a support ticket. A thing the bot got wrong, with the transcript attached.
How it works
Put a bot up, find out where it failed, fix it.
-
Put a bot on your site
Give Ask your address and it reads your site, drafts the instructions and the entries, and shows you both before anything goes live. One script tag, or the WordPress plugin.
-
It flags what went wrong
Four states, detected as the conversation happens: the bot could not answer, the visitor asked for a follow-up, the visitor was unhappy, the bot had stopped answering. States accumulate into one case per conversation.
-
You fix the bot from the flags
Write an entry straight from the failed question, or edit the instructions. Before you publish, the new version is replayed against the questions that previously failed, so you see what you fixed and what you broke.
Before you publish
"11 of your 14 unanswered questions are now answered. 2 got worse."
That is the sentence you get before a new set of instructions goes live, not a week after. Ask keeps the questions your bot failed, replays them against the draft, and shows you the before and after side by side. Publishing is one click. So is rolling back.
The inbox
Every failure has a transcript attached.
A case is one conversation that needs your attention: what the visitor asked, what the bot said, which turn the state was detected at, and whatever they left behind - an email address, a name, a callback request. Filter it, resolve it, or turn it into content. There are no SLAs and no assignment, because this is feedback, not a helpdesk.
WordPress
On WordPress this takes about two minutes.
Install the plugin, connect or create an account, let it read your site. No theme file to edit and no snippet to paste. The plugin also puts the cases in wp-admin, so you can read and resolve them without leaving WordPress.
Trust
What this actually puts on your site.
The widget is open source
The code that runs in your visitors' browsers is public and readable. You can audit exactly what it does and restyle it. The server is not open source and there's no self-hosting option - worth saying plainly rather than letting you find out.
The AI notice cannot be removed
Every conversation says it is talking to an AI. You can word it yourself; you cannot switch it off, on any plan.
One embed cannot run up a bill
When the included usage is gone the bot stops answering. Warnings arrive at 80% and 95%, there is a daily spend cap on top, and a top-up restores service immediately.
Where answers are produced
Answers come from third-party model providers reached through an aggregation service that picks a provider per request. The full set an answer might reach is published as a sub-processor annex.
Plans
Billed per answer. Never per seat.
An answer is one reply from the bot, whatever it had to do to produce it. Staff accounts, cases and inbox usage are not metered at all.
Free
Free
200 credits a month, Basic.
Every feature works. Limited by capacity, not by capability.
Starter
$19 a month
2,000 credits a month, Basic and medium.
Growth
$49 a month
5,000 credits a month, Basic, medium and advanced.
Pro
$99 a month
10,000 credits a month, Basic, medium and advanced.
Business
$249 a month
25,000 credits a month, Basic, medium and advanced.