{"Status":true,"Message":"","Response":{"post":{"postuid":"74e5eac7-e90e-4a07-a77e-a9c328b10783","tenantuid":"d8b744fc-2e70-4089-bb80-dd1d08f6c7b2","projectuid":"fc6490ac-7527-4f49-b06e-46f701280e85","title":"Testing the Chat AI","slug":"article/testing-the-chat-ai","html":"\u003Cp\u003EThe \u003Cb\u003ETest\u003C/b\u003E tab shows what visitors would get from the AI agent before they get it. Try any question, keep the important ones as saved tests, and re-run them all after you change your articles, guidance, knowledge sources, rules or actions.\u003C/p\u003E\u003Cp\u003EGo to \u003Cb\u003EInbox\u003C/b\u003E, then \u003Cb\u003EMessenger settings\u003C/b\u003E, then the \u003Cb\u003ETest\u003C/b\u003E tab.\u003C/p\u003E\u003Ch2 id=\u0022try_a_question\u0022\u003ETry a question\u003C/h2\u003E\u003Cp\u003EType a question a visitor would ask, for example \u003Ci\u003E\u0022How do I reset my password?\u0022\u003C/i\u003E, and click \u003Cb\u003EAsk\u003C/b\u003E. The result shows:\u003C/p\u003E\u003Cul\u003E\u003Cli\u003Ethe answer, as the visitor would see it, or what the AI did instead (asked a clarifying question, said it doesn\u0027t know, or handed off);\u003C/li\u003E\u003Cli\u003Ethe articles it cites, and any internal articles it used but didn\u0027t show;\u003C/li\u003E\u003Cli\u003Eany \u003Ca href=\u0022/article/chat-actions\u0022\u003Eactions\u003C/a\u003E it called, with their inputs;\u003C/li\u003E\u003Cli\u003Ethe guidance used, and the \u003Ca href=\u0022/article/chat-question-rules\u0022\u003Equestion rule\u003C/a\u003E matched, if any;\u003C/li\u003E\u003Cli\u003Ewhat it searched for.\u003C/li\u003E\u003C/ul\u003E\u003Cp\u003EClick \u003Cb\u003ESave as a test\u003C/b\u003E to keep the question.\u003C/p\u003E\u003Ch2 id=\u0022saved_questions\u0022\u003ESaved questions\u003C/h2\u003E\u003Cp\u003EAdd questions with \u003Cb\u003EAdd question\u003C/b\u003E, or paste a list with \u003Cb\u003EAdd several\u003C/b\u003E (one question per line). Changes to saved questions save by themselves. Each question has an expectation:\u003C/p\u003E\u003Ctable border=\u00221\u0022 style=\u0022border-collapse: collapse; width: 100%;\u0022\u003E\u003Cthead\u003E\u003Ctr\u003E\u003Cth\u003EExpectation\u003C/th\u003E\u003Cth\u003EUse it for\u003C/th\u003E\u003C/tr\u003E\u003C/thead\u003E\u003Ctr\u003E\u003Ctd\u003EShould answer\u003C/td\u003E\u003Ctd\u003EQuestions your articles cover\u003C/td\u003E\u003C/tr\u003E\u003Ctr\u003E\u003Ctd\u003EShould say it doesn\u0027t know\u003C/td\u003E\u003Ctd\u003EThings your articles deliberately don\u0027t cover, so the AI must not guess\u003C/td\u003E\u003C/tr\u003E\u003Ctr\u003E\u003Ctd\u003EShould hand off to a person\u003C/td\u003E\u003Ctd\u003EQuestions that must reach your team, such as those caught by a handoff rule\u003C/td\u003E\u003C/tr\u003E\u003C/table\u003E\u003Cp\u003EFor questions that should be answered, you can add two optional checks:\u003C/p\u003E\u003Cul\u003E\u003Cli\u003E\u003Cb\u003EAnswer must mention\u2026\u003C/b\u003E: a word or phrase the answer has to include, such as \u003Ci\u003E14-day free trial\u003C/i\u003E.\u003C/li\u003E\u003Cli\u003E\u003Cb\u003ECited article title or URL contains\u2026\u003C/b\u003E: part of the title or address of an article the AI should cite.\u003C/li\u003E\u003C/ul\u003E\u003Ch2 id=\u0022run_your_tests\u0022\u003ERun your tests\u003C/h2\u003E\u003Cp\u003EClick \u003Cb\u003ERun\u003C/b\u003E on one question, or \u003Cb\u003ERun all\u003C/b\u003E to re-check every question. Each is marked:\u003C/p\u003E\u003Cul\u003E\u003Cli\u003E\u003Cb\u003EPass\u003C/b\u003E: the AI did what you expected.\u003C/li\u003E\u003Cli\u003E\u003Cb\u003EFail\u003C/b\u003E: it didn\u0027t, with the reason, such as \u003Ci\u003E\u0022The answer doesn\u0027t mention \u2026\u0022\u003C/i\u003E or \u003Ci\u003E\u0022Expected an answer, but the AI handed off to a person.\u0022\u003C/i\u003E\u003C/li\u003E\u003Cli\u003E\u003Cb\u003EAsked to clarify\u003C/b\u003E: the AI asked a question back instead of answering. That isn\u0027t wrong, but it may mean the question or your articles are ambiguous.\u003C/li\u003E\u003C/ul\u003E\u003Cp\u003EA summary of passes and failures appears at the top, and each question keeps its last result.\u003C/p\u003E\u003Cfigure\u003E\u003Cimg class=\u0022lazy-load\u0022 data-src=\u0022https://graffiti-auf7e6dwhxhcbwek.z03.azurefd.net/d8b744fc-2e70-4089-bb80-dd1d08f6c7b2/fc6490ac-7527-4f49-b06e-46f701280e85/images/ab74f9c10b3b4c19a3ff6b59340858aa.png?v=-197406949\u0022 data-width=\u0022986\u0022 data-height=\u00221070\u0022 style=\u0022max-width: 100%;\u0022 alt=\u0022The Test tab with a question box and three saved questions: two marked Pass and one marked Fail with its reason, each showing the answer, sources and what was searched for\u0022/\u003E\u003Cfigcaption\u003EMessenger settings, Test tab\u003C/figcaption\u003E\u003C/figure\u003E\u003Ch2 id=\u0022what_tests_do_and_dont_do\u0022\u003EWhat tests do and don\u0027t do\u003C/h2\u003E\u003Cul\u003E\u003Cli\u003ETests use your live, published articles and your \u003Cb\u003Esaved\u003C/b\u003E settings. Click \u003Cb\u003ESave settings\u003C/b\u003E before testing a change.\u003C/li\u003E\u003Cli\u003EThey don\u0027t start conversations, send alerts, or count in reports or search analytics.\u003C/li\u003E\u003Cli\u003EActions are really called, so any system they reach sees the request.\u003C/li\u003E\u003Cli\u003ETests run as an anonymous visitor, so actions for signed-in customers are never used. Test those with \u003Cb\u003ERun test\u003C/b\u003E on the Actions tab.\u003C/li\u003E\u003Cli\u003EEach test costs about two AI calls.\u003C/li\u003E\u003C/ul\u003E\u003Ch2 id=\u0022when_a_test_fails\u0022\u003EWhen a test fails\u003C/h2\u003E\u003Cul\u003E\u003Cli\u003EIf the AI can\u0027t find the answer, check the article is published and its project is ticked under \u003Ca href=\u0022/article/chat-knowledge-sources\u0022\u003Eknowledge sources\u003C/a\u003E, with a description that fits the question.\u003C/li\u003E\u003Cli\u003EIf it cites the wrong article, make the right article clearer about the topic. That usually works better than adding guidance.\u003C/li\u003E\u003Cli\u003EIf a rule catches the wrong questions, make its \u003Cb\u003EWhen a visitor\u2026\u003C/b\u003E description more specific.\u003C/li\u003E\u003C/ul\u003E\u003Cp\u003EFor questions real visitors asked that the AI couldn\u0027t answer, see \u003Ca href=\u0022/article/chat-reports\u0022\u003EChat Reports\u003C/a\u003E. See also \u003Ca href=\u0022/article/configuring-the-chat-ai-agent\u0022\u003EConfiguring the Chat AI Agent\u003C/a\u003E.\u003C/p\u003E","publish_status":0,"post_type":"Article","author":{},"featured_image_updating":false,"meta_description":"Test your AI agent\u0027s responses to visitor questions, save tests, and rerun them to ensure accurate answers, correct citations, and proper handoffs before updates.","display_toc":true,"has_workingcopy":false,"allow_indexing":true,"total_views":5,"date_published":"2026-09-28T14:42:18.787","date_updated":"2026-09-28T15:34:53.123","date_created":"2026-09-28T14:41:30.13"}}}