> For the complete documentation index, see [llms.txt](https://docs.guestway.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.guestway.io/guestway-platform/ai-center-draft.md).
# AI Center (DRAFT)
{% hint style="warning" %}
**Do not publish yet.** The AI Center is not reachable by customers on production today.
* The navigation item is hidden on production.
* Access requires knowledge-base read permission.
* It is unavailable in Hub mode.
* The Dashboard currently shows a coming-soon placeholder on production and staging-prod.
Hold this as an unpublished page until the production gate lifts. Before publishing, re-check the **BETA** badge on Auto-Pilot and whether **Send threshold** has shipped.
{% endhint %}
The **AI Center** is where you build and maintain the knowledge Guestway's AI draws on when it answers guests. Everything the AI knows about your properties — the answers you've curated, your saved replies, your guidebooks, your internal procedures — lives here, and this is where you check whether it's good enough to trust.
Open it from **AI Center** in the left sidebar navigation.
{% hint style="info" %}
Access to the AI Center depends on your role having **knowledge base read** permission. If you don't see it in the sidebar, ask an admin to check your role. It's also unavailable while you're working in a [Hub](https://docs.guestway.io/settings/personal-settings/hubs) — switch to a single organization first.
{% endhint %}
## Overview
The AI Center is organised into sections, each answering a different question:
| Section | What it's for |
| ----------------------- | ----------------------------------------------------------------------- |
| **Add Knowledge** | Grow your knowledge base — teach the AI what guests need to know. |
| **Browse** | Search and explore everything already in your knowledge base. |
| **Suggestions** | Review knowledge gaps and improvements the AI has discovered for you. |
| **Playground** | Test how the AI would respond, before it ever reaches a guest. |
| **Auto-Pilot** *(Beta)* | Configure automatic replies — which topics, which properties, how fast. |
| **Dashboard** | Knowledge base health and Auto-Pilot activity. *Coming soon.* |
A sensible first pass: add a handful of Q\&As for your most repetitive guest questions, check them in **Playground**, then work through **Suggestions** as the AI surfaces what's missing.
`[VISUAL: the AI Center with the section navigation visible]`
## Knowledge scope — where an entry applies
Every knowledge entry carries a **scope**: the properties it's true for. Scope is the difference between "check-in is at 3pm" being helpful and being wrong.
You have two options:
| Scope | Use it when |
| ----------------------- | ----------------------------------------------------------------------------------------------------------------- |
| **Always available** | The entry is true across your whole portfolio — cancellation policy, company contact details, payment methods. |
| **Specific properties** | The entry is only true somewhere. Bind it to any combination of **complexes**, **multi-units**, and **listings**. |
Binding at a **complex** or **multi-unit** level cascades to the listings inside it, so a building handbook attaches once rather than per unit.
{% hint style="warning" %}
**Scope is the single most important field on an entry.** An entry marked **Always available** that's only true for one building will be quoted confidently at guests staying somewhere else. When in doubt, scope narrowly — you can always widen it later.
{% endhint %}
## Add Knowledge
Open **Add Knowledge** and pick the type that matches what you have. There are four.
| Type | What it is | Guests see it? |
| --------------------- | ---------------------------------------------------------------------------------------------------------------- | -------------- |
| **Add Q\&A** | A question guests ask often, with the answer you want sent. Live as soon as you save. | Via AI replies |
| **Add Saved Reply** | A reusable message template your team can send as-is. Auto-Pilot quotes it when a guest message is a good match. | Via AI replies |
| **Guest guidebook** | A guest-facing property guide — welcome info, house rules, local tips. PDF upload or rich text. | Yes, directly |
| **Internal document** | Operator reference only — SOPs, playbooks, internal notes. | **Never** |
{% hint style="info" %}
**Internal documents are never shown to guests**, but the AI still reads them. That makes them the right home for procedure that shapes a correct answer without being quotable — escalation rules, supplier contacts, how you handle a lockout at 2am.
{% endhint %}
{% stepper %}
{% step %}
### Choose a type
Select one of the four cards. The form below changes to match.
{% endstep %}
{% step %}
### Let the AI draft it from your history *(optional)*
Q\&A, saved reply and guidebook entries can be drafted from your own conversation history instead of written from scratch.
1. Enter a topic in the prompt field — for example *Parking*, *check-in codes*, *WiFi password*.
2. Guestway scans your conversation history for how your team has answered that topic before, and fills in the fields.
3. **Review every field and edit before saving.** The draft is a starting point built from past messages, not a finished answer.
`[VISUAL: the Add Knowledge prompt field with a drafted Q&A below it]`
{% endstep %}
{% step %}
### Fill in the content
Write or refine the entry. For guidebooks and internal documents you can either **upload a PDF** or write **rich text**.
{% endstep %}
{% step %}
### Set the scope
Choose **Always available**, or select the complexes, multi-units and listings the entry applies to. See [Knowledge scope](#knowledge-scope-where-an-entry-applies) above.
{% endstep %}
{% step %}
### Save
Q\&A entries are live as soon as you save. `[confirm: whether guidebook and internal-document entries need any additional processing time after upload before the AI can use them]`
{% endstep %}
{% endstepper %}
## Browse
**Browse** is the full inventory of your knowledge base — use it to find an entry, check what you already have before adding something, and spot the entries that are pulling their weight.
### Filtering by type
Tabs across the top narrow the list by entry type:
**All** · **Q\&A** · **Saved replies** · **Guidebooks** · **Internal docs**
Each tab shows a count, so you can see the shape of your knowledge base at a glance — a portfolio with 200 guidebooks and 3 Q\&As will answer very differently from the reverse.
### Searching
Use the search field (*"Search questions, replies, guides…"*) to search across entry content. When nothing matches you'll see **No matching entries** rather than an empty list.
### Sorting
The **Sort** control offers six orders:
| Sort | Shows you |
| -------------------- | ------------------------------------------------------------------- |
| **Most used** | The entries the AI reaches for most. This is the default. |
| **Recently updated** | What's changed lately — useful when auditing after a policy change. |
| **Recently created** | Newest additions first. |
| **Most dismissed** | Entries that were retrieved but then dropped rather than used. |
| **Most upvotes** | Entries your team has marked as good answers. |
| **Most downvotes** | Entries your team has marked as bad answers. |
{% hint style="success" %}
**💡 Academy Tip — read "Most downvotes" and "Most dismissed" together.** They're your two highest-value lists. A frequently-dismissed entry is being retrieved for questions it doesn't actually answer, which usually means the wording is too broad or the scope is too wide. A downvoted entry is being sent and is wrong. Fix these before adding anything new.
{% endhint %}
### Filtering further
Beyond type, you can narrow by:
* **Properties** — show only entries scoped to a given complex, multi-unit or listing.
* **Topics** — show only entries covering a given guest-message topic.
### Opening an entry
Click a row to open the detail panel, which shows the full content, its scope, its topics, and its usage figures. Edit it from there. `[confirm: exact set of actions available in the detail panel — edit, delete, duplicate?]`
`[VISUAL: the Browse list with type tabs, sort control, and an entry detail panel open]`
## Suggestions
**Suggestions** is where the AI tells you what your knowledge base is missing. Rather than auditing it yourself, you work a review queue.
Suggestions are grouped into three kinds:
| Kind | What it means | What to do |
| ------------------ | ---------------------------------------------------------------------------- | -------------------------------------------------------- |
| **Knowledge gaps** | A guest asked something your knowledge base couldn't answer. | Add the missing entry — this is the highest-value queue. |
| **Discovery** | The AI spotted knowledge worth capturing from how your team already answers. | Review and save if it's right. |
| **Conflicts** | Two entries disagree with each other. | Decide which is correct and fix or remove the other. |
Two views split the work:
* **Open** — suggestions still waiting on you. When it's empty you'll see **You're all caught up**.
* **Reviewed** — what you've already actioned, so you can see the history. Empty state: **Nothing reviewed yet**.
{% hint style="warning" %}
**Conflicts are worth jumping the queue for.** When two entries contradict each other, the AI may quote either one — so the same guest question can get two different answers depending on the phrasing. A conflict is a live inconsistency, not a backlog item.
{% endhint %}
Knowledge gaps can also be raised straight from a conversation in the [Unified AI Inbox](https://docs.guestway.io/guestway-platform/unified-ai-inbox) when the AI couldn't answer — so the queue fills from real guest messages, not guesswork. `[confirm: exact label and placement of the "add knowledge" prompt shown on an unanswerable inbox conversation]`
`[VISUAL: the Suggestions queue showing the Knowledge gaps tab with open suggestions]`
## Playground
**Playground** lets you dry-run a guest message against your knowledge base and see how the AI would reply — without a guest involved.
{% stepper %}
{% step %}
### Start a new conversation
Open **Playground** and start a **New conversation**.
{% endstep %}
{% step %}
### Write a guest message
Type a guest message the way a guest would actually write it — misspellings, no punctuation, half a sentence.
{% endstep %}
{% step %}
### Review the reply
Read the reply, and check which knowledge it drew on.
{% endstep %}
{% endstepper %}
Use it to answer *"would we get this right?"* before you find out from a guest. It's also the fastest way to confirm a new entry works: add it, then ask the question it's meant to answer three different ways.
{% hint style="success" %}
**💡 Academy Tip — test the phrasing guests actually use, not the phrasing you wrote.** Your Q\&A might be titled *"What time is check-in?"* while guests write *"when can we get in"*. Playground is where you find that mismatch, and the fix is usually to widen the question wording rather than to add a second entry.
{% endhint %}
`[VISUAL: a Playground conversation showing a dry-run reply and the knowledge it used]`
## Auto-Pilot
{% hint style="warning" %}
**Auto-Pilot is in Beta.** It's labelled **BETA** in the platform. Roll it out on a small set of properties with **Testing mode** on before letting it reply to guests unsupervised.
{% endhint %}
**Auto-Pilot** sends replies to guests automatically, using your knowledge base. You control which topics it handles, which properties it covers, and how quickly it responds.
You can build more than one Auto-Pilot configuration and scope each to different properties — a city-centre portfolio and a set of rural cottages rarely want the same behaviour. One config can be the **default**.
{% stepper %}
{% step %}
### Name the configuration
Give it a name you'll recognise in the overview list.
{% endstep %}
{% step %}
### Set Enabled and Testing mode
| Control | What it does |
| ---------------- | ---------------------------------------------------------------------------------------- |
| **Enabled** | When off, Auto-Pilot does not send replies for this config. |
| **Testing mode** | Replies that would have been sent automatically arrive as **drafts for review** instead. |
{% hint style="warning" %}
**Start every configuration in Testing mode.** You get the full behaviour — same topics, same knowledge, same timing — but every reply lands as a draft in the [Unified AI Inbox](https://docs.guestway.io/guestway-platform/unified-ai-inbox) for a human to approve. Run it for a week, read what it *would* have sent, and only then turn Testing mode off.
{% endhint %}
**Testing mode** only appears once **Enabled** is on.
{% endstep %}
{% step %}
### Set the response delay
**Response delay** is how long Auto-Pilot waits before it replies — so you can take over first, and so replies don't feel robotically instant.
* Minimum **10 seconds**, maximum **15 minutes**.
* Default **10 seconds**.
* Enter the value in **seconds** or **minutes**.
{% hint style="success" %}
**💡 Academy Tip — a longer delay is a feature, not a compromise.** A two-minute delay gives whoever is watching the inbox time to step in on anything sensitive, and makes the reply read as considered rather than automated. Instant replies are only worth it for genuinely trivial questions.
{% endhint %}
{% endstep %}
{% step %}
### Choose the allowed categories
Under **Allowed categories**, pick the guest-message topics Auto-Pilot is permitted to answer. Anything outside your selection is left for a human.
Topics are grouped, and each shows how much of your guest message volume it represents — so you can see what you'd actually be automating. Compare against **Industry averages** or **Your averages** to judge whether a category is worth turning on.
{% hint style="warning" %}
**Be deliberate about which categories you allow.** Start with topics that are high-volume and low-risk — Wi-Fi, parking, check-in times. Keep anything touching money, cancellations, complaints or access problems with a human until you've watched Auto-Pilot handle the easy categories well.
{% endhint %}
{% endstep %}
{% step %}
### Set the scope
Choose which complexes, multi-units and listings this configuration applies to.
{% endstep %}
{% step %}
### Save
The overview list shows each configuration with its scope, its allowed topics, whether it's enabled, whether it's in testing mode, and its **coverage** — how much of your guest message volume it's set up to handle.
`[VISUAL: the Auto-Pilot editor showing Enabled, Testing mode, Response delay and Allowed categories]`
`[VISUAL: the Auto-Pilot overview list with coverage bars]`
{% endstep %}
{% endstepper %}
## Dashboard
An overview of knowledge base health and Auto-Pilot activity. **Coming soon** — the section appears in the navigation but isn't available yet.
## Tips & Best Practices
{% hint style="success" %}
**💡 Fix before you add.** Work **Suggestions** and the **Most downvoted** / **Most dismissed** lists in Browse before writing new entries. A knowledge base with 40 accurate entries outperforms one with 400 where a tenth contradict each other.
{% endhint %}
{% hint style="success" %}
**💡 Scope narrowly, widen later.** An over-scoped entry is worse than a missing one: a missing entry produces a knowledge gap you'll see in Suggestions, while a wrongly-scoped entry produces a confident wrong answer you won't.
{% endhint %}
{% hint style="success" %}
**💡 Write answers, not documents.** The AI quotes what you give it. A Q\&A that answers the question in two sentences will be reused cleanly; the same information buried in page four of a PDF often won't surface at all.
{% endhint %}
{% hint style="success" %}
**💡 Put procedure in internal documents.** Anything that shapes the right answer but shouldn't be quoted at a guest — escalation paths, supplier contacts, pricing latitude — belongs in an internal document.
{% endhint %}
{% hint style="success" %}
**💡 Re-test after a policy change.** When check-in times, pet rules or cancellation terms change, use **Recently updated** in Browse to confirm every affected entry was caught, then re-ask the question in **Playground**.
{% endhint %}
## FAQ
Why can't I see the AI Center in my sidebar?
Your role needs knowledge-base read permission. It's also hidden while you're working in a Hub — switch to a single organization.
Does the AI use my guidebooks even if I don't add them here?
Guidebooks created under [Properties → Guidebooks](https://docs.guestway.io/guestway-platform/properties/guidebooks) have a **Used by AI** setting that controls whether the AI may read them. `[confirm: whether guidebooks created in Properties appear in AI Center → Browse automatically, or whether the two are separate collections]`
What's the difference between a Q&A and a saved reply?
A **Q\&A** is a question-and-answer pair for the AI to draw on. A **saved reply** is a message template your team sends as-is — and which Auto-Pilot may quote when it closely matches what the guest asked. See [Saved Replies](https://docs.guestway.io/guestway-platform/properties/saved-replies).
Will Auto-Pilot reply to everything?
No. It only answers topics you've allowed under **Allowed categories**, only for properties in its scope, and only when your knowledge base actually supports an answer. Everything else waits for a human.
Can I see what Auto-Pilot would have said without it sending?
Two ways: turn on **Testing mode** so live replies arrive as drafts, or use **Playground** to dry-run a message yourself.
Do internal documents ever reach guests?
No. They're operator-reference only. The AI reads them to inform an answer, but their content isn't shown to guests.
## Related pages
* [Unified AI Inbox](https://docs.guestway.io/guestway-platform/unified-ai-inbox) — where AI drafts and Auto-Pilot replies appear, and where knowledge gaps are raised from real conversations.
* [Saved Replies](https://docs.guestway.io/guestway-platform/properties/saved-replies) — the reply templates that also feed the knowledge base.
* [Guidebooks](https://docs.guestway.io/guestway-platform/properties/guidebooks) — guest-facing property guides and their **Used by AI** setting.
* [Users](https://docs.guestway.io/settings/organization-settings/users) — roles and the permission that controls AI Center access.
* [Hubs](https://docs.guestway.io/settings/personal-settings/hubs) — the multi-organization view where the AI Center is unavailable.
## Open items for the author
Resolve these against the live app before publishing:
1. `[confirm]` Whether guidebooks and internal documents need processing time after upload before the AI can use them.
2. `[confirm]` The full set of actions in the Browse entry detail panel (edit / delete / duplicate).
3. `[confirm]` The exact label and placement of the knowledge-gap prompt shown on an unanswerable inbox conversation.
4. `[confirm]` Whether guidebooks created under Properties appear in AI Center → Browse automatically, or whether the two are separate collections. **This one matters most** — it determines whether the Guidebooks article needs a cross-reference, and it's the question a customer will hit first.
5. Re-check the **BETA** badge on Auto-Pilot and the absence of **Send threshold** at publish time.
6. Capture the six `[VISUAL: …]` screenshots.
---
# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.
## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.
Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:
```
GET https://docs.guestway.io/guestway-platform/ai-center-draft.md?ask=&goal=
```
`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.
Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.