Table of Contents
- Key takeaways
- What is personalized AI customer support?
- Why does personalization matter in customer service?
- The four levels of AI personalization
- Level 1: Content
- Level 2: Page
- Level 3: Person
- Level 4: Live account
- Which AI solution fits each level?
- The stack: which tool does which job
- How to choose your personalization level
- Worked example: sizing level 4 for 1,000 chats a month
- What data should personalized AI use?
- Does personalized AI need a BAA or DPA?
- How to set up personalized AI customer service
- Frequently asked questions
- Basics
- What are personalized support solutions powered by AI?
- Is a personalized chatbot the same as an AI agent?
- Can AI personalize answers for anonymous visitors?
- Data and privacy
- Is it safe to give an AI customer data?
- Does SiteGPT remember customers between conversations?
- Setup and tools
- Do you need a developer to personalize AI customer service?
- Can SiteGPT look up an order?
- Which AI solution is best for Shopify personalization?
- Sources
Created by
Do not index
Created time
Oct 16, 2023 12:57 AM
Personalized AI customer support solutions tailor each answer to the visitor's question, page, identity, or account, and the right solution depends on how much of that data the AI can safely reach.
TL;DR
- AI personalization in customer service has four levels: content, page, person, and live account.
- Each level needs more data and more setup. Most questions need less than teams assume, so count your own transcripts first.
- Verify identity before the AI uses any account data. Fin requires it, and SiteGPT offers it on Growth and above.
- Live lookups and account changes need an AI agent connected to your systems, such as Fin, Zendesk, or Gorgias. SiteGPT covers the first three levels and hands the rest to a person.
- Personal data raises the compliance stakes. A HIPAA BAA and a signed DPA at SiteGPT are Enterprise only.
Key takeaways
Level | What the AI knows | Example answer | Data it needs |
1. Content | Your pages and documents | "Returns are free within 30 days." | A help center, website, or files |
2. Page | Where the visitor is | "This shoe runs half a size small." | Page URL, title, or product |
3. Person | Who the customer is | "Your Pro plan includes this feature." | Verified identity and account facts |
4. Live account | Current records | "Order 4417 shipped today." | An API connection and permissions |
What is personalized AI customer support?
Personalized AI customer support is automated service that changes its answer based on what it knows about the customer and their situation. A generic bot gives every visitor the same reply. A personalized one uses context such as the current page, the customer's plan, or a live order record.
Personalization and accuracy are separate problems. The AI first needs correct content, which retrieval-augmented generation provides. For the broader question of where AI works in service at all, see what AI customer support handles well and where it fails.
Why does personalization matter in customer service?
Personalization matters because customers notice generic answers and stop using the channel. McKinsey's 2021 research found that 71% of consumers expect personalized interactions and 76% get frustrated when they do not receive them.
Gartner's 2026 survey of 3,566 customers points the same way. Gartner lists generic information among the past experiences that push customers to other channels. Only 27% would try a chatbot again after a negative experience.
The four levels of AI personalization
The four levels of AI personalization are content, page, person, and live account. Each level adds data, setup work, and risk, so start at the lowest level that answers the question.
Level 1: Content
Level 1 means the AI answers from your own content and matches the reply to the question asked. It needs no customer data, and it works for anonymous visitors.
Language also sits at this level. SiteGPT lists 95+ languages. Its docs say the AI usually answers in the language the visitor writes in, without a guarantee, and that you can set a language in the instructions.
Level 2: Page
Level 2 means the AI knows what the visitor is looking at. A question like "is this in stock in blue?" has no answer without the page.
In SiteGPT, the Enable Page Context Awareness setting sends the page URL, title, and description to the chatbot. It does not send the page text. A developer can also pass plain facts, such as the product name, through the widget SDK.
Level 3: Person
Level 3 means the AI knows which customer it is talking to and uses stable facts about them. Typical facts are plan, account type, company, and region.
This level starts with identity verification. Your server signs the visitor's identity, so a visitor cannot claim to be someone else. Fin can only reference customer data for identity-verified customers. In SiteGPT, identity verification needs the Growth plan or above and a backend that keeps the API key secret.
Level 4: Live account
Level 4 means the AI reads current records and can change them. Examples are an order lookup, an address change, or a cancellation.
This level needs an AI agent connected to your systems through an API. It carries the most risk, because a wrong action costs money. Fin warns that if a data connector fails and no fallback is set, it does not automatically say it lacks the information. Gorgias turns on customer confirmation automatically for irreversible actions such as cancelling an order.
Which AI solution fits each level?
The right AI solution depends on the highest level your questions need. The four options below are ordered for the main scenario in this guide, which is personalizing with customer data. Fin comes first because it covers all four levels on any system with an API.
# | Solution | Best for | Levels | Main limit |
1 | Live account answers across any system with an API | 1-4 | Customer data needs identity verification, and connectors need fallback rules | |
2 | Personalized answers from your content, without changing your helpdesk | 1-3 | No built-in order lookups or account changes | |
3 | Zendesk AI agents | Teams that already run Zendesk | 1-4 | Get user info works with the Ticket Received event only |
4 | Gorgias AI Agent | Shopify stores that want order changes automated | 1, 4 | A Shopify cancellation does not pass to 3PL or fulfillment apps |
For a ranked comparison with prices, see the guide to choosing a customer support chatbot service for your website. For the pricing units and first-year costs, see how to choose a customer service AI chatbot solution.
The stack: which tool does which job
Most teams run personalization as layers, with a different tool at each one. SiteGPT sits in the answer layer next to your helpdesk or live chat tool.
Layer | Job | Tool |
Answer layer | Answer from your content, using page and customer context | |
System of record | Hold tickets, customer records, and orders | Zendesk, your CRM, or Shopify |
Action layer | Read live records and change them | Fin, Zendesk AI agents, or Gorgias |
People | Handle exceptions, disputes, and upset customers | Your team, with the transcript |
Gartner found that 87% of customers say access to a human agent is essential when companies use GenAI in customer service. Keep the people layer visible at every level.
How to choose your personalization level
To choose a level, read 100 recent conversations and tag each one with the lowest level that would have answered it. Then use this matrix.
If you need... | Pick... | Why |
Accurate answers for anonymous visitors | Level 1 | No customer data, so little privacy risk |
Product or pricing help on specific pages | Level 2 | Page context removes the "which product?" question |
Answers that differ by plan or account type | Level 3 | Stable facts change rarely and need no live connection |
Order status or account changes at high volume | Level 4 | Only a connected agent can read or change records |
Refunds, disputes, or anything that commits money | A person, or a fixed flow | The cost of a wrong answer is too high |
Worked example: sizing level 4 for 1,000 chats a month
This example uses a hypothetical subscription business with 1,000 chats a month. The numbers are illustrative. Replace them with your own tags.
Level needed | Share of 100 tagged chats | Chats per month |
1. Content | 52% | 520 |
2. Page | 14% | 140 |
3. Person | 21% | 210 |
4. Live account | 13% | 130 |
Levels 1 to 3 cover 87% of chats, or 870 a month, with no live connection. The other 130 chats go to a person. At 5 minutes each, that is 650 minutes, or about 10.8 hours a month. At $25 an hour, the cost is about $271 a month.
Compare that $271 with the cost of building and maintaining a live connection. If the level 4 share grows, the connection starts to pay for itself. The customer service automation guide covers the full ROI math.
What data should personalized AI use?
Personalized AI should use the least customer data that answers the question. GDPR Article 5 calls this data minimisation: personal data must be adequate, relevant, and limited to what is necessary for the purpose.
Follow these rules at every level:
- Pass facts the visitor could already see. SiteGPT's context runs in the visitor's browser, so its docs say never to put secrets there.
- Verify before you personalize. Anyone can type an email address into a chat.
- Set a failure rule. Tell the AI what to say when a lookup fails, so it never invents account details.
- Keep regulated data out of the first rollout. Test with plan and account type before you touch health or financial records.
Does personalized AI need a BAA or DPA?
Personalized AI needs a signed BAA if it handles protected health information for a covered entity, and a DPA is the standard GDPR contract for personal data. At SiteGPT, both are Enterprise only. Starter, Growth, and Scale do not include them.
A HIPAA workspace also narrows the product. Chat integrations are not available, conversations stay in the chat on your website, and you must keep PHI out until SiteGPT confirms in writing that HIPAA processing is on. See SiteGPT's HIPAA program and the guides to HIPAA compliant AI chatbots and GDPR compliant chatbot platforms.
How to set up personalized AI customer service
To set up personalized AI customer service, build one level at a time and test each level before you add the next.
- Tag 100 recent conversations by level.
- Connect your content and fix any gaps the tags show.
- Turn on page context and test questions on your top 10 pages.
- Add identity verification, then pass one account fact, such as the plan.
- Set the handoff rules and confirm the agent receives the transcript.
- Review the results weekly, and add level 4 only if the volume justifies it.
SiteGPT covers steps 2 to 5 for levels 1 to 3. Growth costs $129 per month billed monthly, or $79 per month billed yearly ($948 a year). Yearly billing saves $600 a year, or 39%. Every self-serve plan has a 7-day free trial that requires a credit card.
Frequently asked questions
Basics
What are personalized support solutions powered by AI?
Personalized support solutions powered by AI are tools that adapt service answers to the customer. They use the visitor's question, current page, verified identity, or live account records. Chatbots, AI agents, and helpdesk add-ons can all do this at different levels.
Is a personalized chatbot the same as an AI agent?
A personalized chatbot answers with customer context, and an AI agent can also take actions in your systems. The line is level 4. See customer service AI use cases for how the two differ in practice.
Can AI personalize answers for anonymous visitors?
AI can personalize for anonymous visitors using the question, the language, and the current page. It cannot use account data until the visitor's identity is verified.
Data and privacy
Is it safe to give an AI customer data?
Giving an AI customer data is safe only with verified identity, the smallest useful set of fields, and a contract that matches the data. Start with facts such as plan or account type.
Does SiteGPT remember customers between conversations?
SiteGPT treats a verified visitor as the same person by email address, and a different person on the same browser starts fresh. Context set through the widget SDK is not saved, so your page must send it again on each page load.
Setup and tools
Do you need a developer to personalize AI customer service?
Levels 1 and 2 need no developer in SiteGPT, because page context is a setting. Passing custom context and verifying identity need a developer, and level 4 needs API work on any platform.
Can SiteGPT look up an order?
SiteGPT has no built-in order lookup. Your page can pass a visitor's order status as a plain fact through the widget context, and anything that changes an order should go to a person or a connected AI agent.
Which AI solution is best for Shopify personalization?
Gorgias AI Agent is the closest fit for order changes, because it works with Shopify orders directly once connected. A knowledge-trained chatbot such as SiteGPT fits product and policy questions.
Sources
Checked on October 30, 2026.
Written by
