The AI Chat Widget is a floating assistant you can drop onto any website. It answers visitor questions from your knowledge base and live inventory, then sends visitors to book. Conversations flow into your Inbox, where a human can take over at any time, and questions the assistant could not answer are collected for you to fill in. The widget, the AI phone, and inbound-text answering all share the same personality and knowledge, so setting it up once configures every channel.
Everything on this page is configured per location on the AI Chat Widget settings page. If your account has more than one location, use the location selector at the top to choose which one you are editing.
Turning the widget on
The widget stays hidden on your site until you enable it. On the settings page:
- Flip the AI Chat Widget switch to Live.
- Copy the embed snippet (see Install on your website) and paste it into your site.
Turning the switch back to Off hides the widget everywhere it is installed without removing the snippet, so you can pause it any time.
Appearance
Control how the launcher and chat panel look.
| Setting | Description |
|---|---|
| Brand color | The primary color for the header, launcher, and visitor message bubbles |
| Header text color | The text/icon color that sits on top of the brand color |
| Position | Bottom right or bottom left of the page |
| Theme | Light or dark chat panel |
| Launcher icon | Pick from a set of emoji, or upload your own image (square works best, under 2 MB) |
Style
Extra styling controls for a polished, on-brand look.
| Setting | Options |
|---|---|
| Accent color | The end color of the header gradient |
| Header style | Gradient or solid |
| Widget size | Compact, standard, or large |
| Launcher shape | Circle, rounded, or pill |
| Bubble corners | Rounded or sharp |
| Font | System, rounded, serif, or mono |
| Header subtitle | A small line under the title, e.g. "Typically replies in a few minutes" |
| Input placeholder | The greyed-out prompt text in the message box |
| Assistant avatar | Show or hide the small avatar next to assistant messages |
| Launcher pulse | A subtle pulse animation on the closed launcher |
Content
Set the words visitors see first.
| Setting | Description |
|---|---|
| Assistant name | Shown in the chat header. Defaults to your location name if left blank |
| Welcome message | The first thing visitors see when they open the chat |
| Suggested questions | Up to six one-tap starter questions that appear before the visitor types anything |
AI and knowledge
This section controls how the assistant thinks and where it gets its facts.
The assistant name, personality, business context, and knowledge base are shared with your AI phone. Editing them here updates your phone assistant too, so both channels stay consistent.
| Setting | Description |
|---|---|
| Personality and tone | How the assistant should talk, e.g. "Friendly and upbeat, like a local guide." Leave blank for a helpful default |
| Business context | Anything the AI should know: policies, FAQs, what makes you special. It will never invent facts beyond what you provide |
| Knowledge bases | Which knowledge bases this chat agent can draw on. Leave nothing selected to use this location's bases plus shared ones |
Answering inbound texts with AI
The same assistant can answer texts sent to your location's phone number, not just website chats.
- Answer inbound texts with AI turns this on. It handles both brand-new texters with no booking and questions from existing customers about their booking (threading the reply into that booking and staying quiet if a staff member is already replying).
- Auto-reply confidence sets how sure the assistant must be before it sends a text answer on its own:
- Cautious only sends very sure answers.
- Balanced is the recommended middle ground.
- Eager answers more and escalates less.
When the assistant is not confident enough, the customer gets a quick acknowledgment and the conversation is flagged for a human in your Inbox. Every text conversation shows up in the Inbox and you can take over at any time.
How answers are grounded
The assistant does not guess. It answers from a knowledge base built from sources you provide, plus your live inventory and availability.
- Your website, FAQs, and uploaded documents are indexed once and kept in sync. When a visitor asks a question, the assistant searches your knowledge for the most relevant passages and answers from them.
- Knowledge bases are account-level. A base can be scoped to a single location or left shared across all of them.
- When you create your first base (or open the Knowledge tab in the Inbox for a location that has none), one is created automatically and seeded with your configured website, which is scraped for you so the assistant works out of the box.
Knowledge sources
Manage sources from Inbox > Knowledge. Each base can hold any mix of:
| Source type | Use it for |
|---|---|
| Q&A | A single question and its answer |
| Text / policy | A block of prose such as a cancellation policy or FAQ paragraph |
| Website page | A URL that RentalTide scrapes and re-scrapes |
| Upload file | A document (text is extracted automatically); files up to 15 MB |
| Bulk import | Paste a whole FAQ or Q&A block, or a structured list, to create many at once |
Every source is chunked and embedded when you save it, so the assistant can find it by meaning rather than exact wording. Editing a source's content re-indexes it automatically, and you can force a re-index at any time.
You can preview what the assistant would find before going live. The knowledge manager includes a search test: type a visitor question and see which sources come back and how strong each match is.
Knowledge gaps
When the assistant searches your knowledge and the best match is weak, it records that question as a knowledge gap. This capture happens quietly in the background and never changes the answer the visitor sees.
- An hourly job clusters near-duplicate questions together into a single gap with a canonical question, a category, the different phrasings people used, and a times-asked count.
- Categories come from a fixed set (Booking and Reservations, Cancellations and Refunds, Pricing and Fees, Age and License Requirements, Waivers and Forms, Check-in and Logistics, Equipment and Specs, Policies and Rules, Hours and Location, Safety and Insurance, Add-ons and Extras, and Other) so similar questions group cleanly.
Triaging gaps
Open Inbox > Knowledge Gaps. Filter by Needs answers, Answered, or Archived, and for each gap you can:
| Action | What it does |
|---|---|
| Add answer | Write the answer once. It is saved as a Q&A into the knowledge base you choose and the assistant can use it next time |
| Archive | Dismiss a gap you do not want to answer |
| Reopen | Move an answered or archived gap back to "needs answers" |
Expand a gap to see every way people phrased the question, ranked by how often each phrasing came up. The most-asked unanswered gaps also surface in your daily briefing so recurring blind spots do not get lost.
The Inbox
Conversations from the widget (and AI-answered texts) live in your Inbox, alongside leads and knowledge tools. The tabs are:
| Tab | Purpose |
|---|---|
| Conversations | Live AI chats with the option for a human to take over |
| Opportunities | Captured leads who have not booked yet |
| Forms | Operator-built intake forms |
| Knowledge | The knowledge-base manager (sources feed every AI channel) |
| Knowledge Gaps | Triage for questions the assistant could not answer |
| Settings | Inbox configuration |
Human takeover
Open a live conversation to read the transcript in real time. When you send a reply, the assistant is automatically paused so your message never races an AI answer. If the visitor still has the widget open, your reply is delivered instantly in their chat; if they have closed it and shared a phone number, the reply is sent as a text instead, and their text replies thread back into the same conversation. Hand the conversation back to the assistant when you are done.
Converting a chat to text
From a conversation you can also convert to text message. RentalTide texts the visitor from your business number. If the number matches an existing booking, the whole conversation continues in your Messages inbox and future replies thread there; otherwise it is sent as a one-off text.
Stats
Once conversations start coming in, the settings page shows a quick summary for the selected location:
| Metric | Meaning |
|---|---|
| Conversations | Total conversations for this location |
| Today | Conversations started today |
| Handed to human | Share of conversations the assistant handed off |
| Avg messages | Average number of messages per conversation |
Live preview
A working preview sits next to the settings form. It is not a mockup, it is a real test console: it uses your unsaved edits and your saved knowledge, so you can ask it questions and see exactly how the live widget will answer before you publish anything.
Install on your website
Once the widget is enabled, copy the one-line snippet from the Install on your website section and paste it just before the closing </body> tag on any page. The snippet is a single script tag that carries your location ID, so there is nothing else to configure:
The page fills in your location automatically, so use the Copy button rather than typing it by hand. The widget works on any site, including WordPress, Squarespace, Wix, Shopify, and plain HTML.
Widget does not appear on my site -- Confirm the switch is set to Live and that the snippet is pasted before </body>. The widget stays hidden until it is enabled.
Assistant says it does not know something -- Add the missing fact as a knowledge source, or answer it from the Knowledge Gaps tab. The assistant only answers from what you have provided.
A new source is not being used -- Sources are indexed when you save them. If a large source was just added, give it a moment, or use the re-index action to force it.
Answers are not on-brand -- Adjust Personality and tone and Business context under AI and knowledge. Remember these are shared with your AI phone.