Hollo AI
Hollo AI answers your users first, in seconds and at any hour, in their own language. It only answers from what you give it, and it hands the conversation to you when it isn't sure. This page covers how to set it up, write good knowledge, and check that it's working.
There's no code for this page. Everything happens in your app's AI settings in the dashboard.
Add knowledge
In your app's AI settings, add sources:
- Pages and docs: paste URLs, such as your help centre, FAQ page or docs site. Hollo fetches them and keeps them in sync, so when you update a page, the AI's answers follow.
- Text and FAQs: paste answers to the questions you get most, as plain text or question-and-answer pairs. Good for things that aren't written down anywhere public.
Write instructions
Instructions shape how the AI talks and what it avoids. For example:
Be friendly and brief. Never promise refunds or release dates. If someone reports a bug, ask for their device and app version, then hand the conversation to the team.
Mark topics for humans
List the topics the AI should never handle itself, like refunds, account recovery or legal questions. When one comes up, the AI hands over straight away.
Test it
From your test device, ask the questions you get most often, in your own words. Check each answer and the sources it used in the inbox. If an answer is wrong or vague, fix the source, not the instructions.
Turn it on
The AI is switched on or off per app. Turn it on when you're happy with the answers. While it's off, every message goes straight to your inbox.
Turning the AI off
The AI is optional. Some teams like to answer every message themselves, and that's a fine way to use Hollo.
- Where: in the dashboard, open your app's AI settings and use the Hollo AI switch. Each app has its own switch, so you can use the AI in one app and not another.
- New apps start with the AI off. Nothing is answered by the AI until you switch it on.
- What changes when it's off: new messages go straight to your inbox and your notifications, users see your usual "we'll get back to you" welcome text, and no AI replies are counted against your plan.
- Your knowledge and instructions are kept. Switch the AI back on and it picks up where it left off.
- Conversations already in progress stay in your inbox. Turning the AI off never deletes messages.
You can also step in on a single conversation without switching anything off: Take over in the inbox stops the AI for that conversation only.
Writing knowledge that works
The AI is only as good as its sources. A few habits help a lot:
- One topic per entry. "How to restore a purchase" works better than a long page covering billing, accounts and restores at once.
- Write the way users ask. Users say "I paid but it still shows the free version", not "entitlement sync failure". Use their words in your FAQ.
- Include the steps. If the answer is "go to Settings → Account → Restore purchases", write exactly that, with the real names of your screens.
- Say what you don't do. If your app has no Android version or no family sharing, say so. Otherwise the AI has nothing to answer with and hands off.
- Keep it current. Remove old pricing and retired features. The AI can't tell an old page from a new one.
Writing instructions
Instructions are for tone and boundaries, not facts. Facts go in your knowledge, where they're easier to keep up to date.
| Good for instructions | Better in knowledge |
|---|---|
| "Keep answers under four sentences." | Your prices and what each plan includes. |
| "Use the user's first name if we have it." | How to export data from your app. |
| "Never promise a fix date." | Known issues and their workarounds. |
| "For bug reports, ask for steps to reproduce." | Supported devices and iOS versions. |
When it hands off
The AI passes the conversation to you when:
- it can't find the answer in your sources,
- the user asks for a person, or
- the topic is one you've marked for humans, like refunds.
The user is told a person will follow up. The conversation is marked Needs you in your inbox, and you get a notification. Once you reply, the conversation is yours: the AI stays out of it until it's closed.
Handing off is the right result for a question the AI can't answer well. A short, honest "a person will get back to you" is better than a confident wrong answer.
What users see
- AI replies are clearly labelled as coming from AI, so users always know who they're talking to.
- The AI replies in the language the user writes in, even if your sources are in English.
- If the user asks for a person, they get one. The AI doesn't argue.
Sources on every answer
Every AI answer records which sources it used. In the inbox you can see them under the answer, so you can tell at a glance why it said what it said. When an answer is wrong, the source list shows you what to fix.
Resolved by AI
A conversation counts as resolved by AI when it's closed without anyone from your team replying in it. Your dashboard shows how many conversations the AI resolved, so you can see what it's saving you. The inbox also separates conversations resolved by the AI from those resolved by your team.
AI replies and your plan
Each message the AI sends counts as one AI reply. Starter includes 500 a month and Growth includes 2,000. If you need more, top up any time: $10 for 1,000 replies. See Early access.
Good to know
- Your inbox replies aren't knowledge. The AI only learns from the sources you add. If you answer the same question by hand often, add the answer as a source.
- Read a sample of answers. Look over some AI answers each week, especially in your first month. It's the quickest way to find gaps in your sources.
- Sending cards is coming later. The AI will be able to offer a review request or an offer when it fits the conversation.