Find out what your AI agent says when someone pushes it.
A paid audit of your website chatbot, AI receptionist or support bot. We test it the way a hostile or confused customer would, and hand back a scored report with every failure shown.
Seven standards, four of them critical.
This is the battery our own agents run before they are switched on for a client. Here it runs against yours.
The four critical standards
Regulated advice: does it stay out of legal, medical or financial advice. Complaint handover: does a complaint reach a person. Human request: does "let me talk to someone" get you someone. No instruction-following: can a customer talk it out of its rules.
A failure on any of these is a critical failure in the report.
Three more standards
The battery covers seven standards in total. The remaining three are scored in the same report, with the same level of detail on each failure.
The fixed hostile questions are supplemented with questions built from your own FAQ and deliberately roughened, the way real people type.
Canada set
Extra checks on privacy, consent and handover behaviour for agents that talk to Canadian customers: what it says about personal information, how it treats consent, and whether it hands over to a person when it should.
Quebec French set
The same style of checks, written and run in Quebec French, because customers in Quebec expect to be served in French.
A report you can act on.
A scored report
A score for each standard, with every failure shown: the exact question we asked and the exact reply your agent gave.
A plain-language summary
One page for the owner: what is fine, what is not, and what to fix first. No jargon.
One re-test within 30 days
Fix what the report found, then we run the battery again and show you what changed.
For agencies and vendors
If you build agents for clients, the same report is evidence of what you delivered, and the re-test shows the fixes worked.
You authorise it. We test it. You get the report.
You authorise the test
You confirm in writing that we may test the agent, and tell us which one.
We connect
For a chatbot we connect to its chat endpoint. For an AI receptionist we test the line. Phone lines are tested by a person, not robocalled.
Report in about 5 business days
The scored report and the plain-language summary, then one re-test within 30 days.
A test, not a verdict.
The audit is not legal advice and it is not a certification. It shows how your agent behaved on the day we tested it, against the questions we asked. A clean result does not promise the agent will never fail, and it does not mean you comply with any law. For legal questions, speak to a lawyer.
Fixed prices, one agent at a time.
Agent Audit: $1,500 USD, one-off. Includes the hostile-test battery, the scored written report with transcripts, a risk ranking and one re-test within 30 days.
Canadian-law and Quebec French test sets, add-on: $500 USD.
Monthly monitoring: $299 USD a month. We re-run the battery every month.
Prices are shown in US dollars, or Canadian dollars if you are in Canada.