Adding AI Chatbots to Client Sites: A Guide for Freelancers and Agencies

For a freelancer or small agency, an AI chatbot is one of the easiest things to add to a client site: one script tag, an hour of content work, and the client gets something that answers their visitors at 2 a.m. The hard part isn't the install. It's deciding who owns the account, who reads the conversations, and who keeps the answers up to date after launch.

I build ChatBento, which installs on Framer, Webflow and WordPress with one line, so it fits naturally into client work. This is the workflow I'd use, from the first client call to the monthly check-in.

Should the client own the account, or should you?

Decide this before you set anything up. Both work, and the right answer depends on how involved you want to stay.

Option 1: the client has their own account. You set it up, then hand over the login (or set it up while sharing their screen).

• The client pays the vendor directly, so you're not chasing reimbursements.
• Leads and data belong to the client from day one, which keeps privacy responsibilities clear: the client is the business collecting the data.
• If the relationship ends, nothing needs migrating.
• Downside: you need the client's help for every change, unless they give you access.

Option 2: you run one account with a chatbot per client. You manage everything and include it in a monthly fee.

• One dashboard for all your clients' bots.
• You control quality: you can fix a wrong answer the same day.
• You become responsible for forwarding leads and for the data handling, and the client needs to know who processes their visitors' data.
• If a client leaves, you have to hand over or rebuild their bot.

A useful rule: if you're selling a one-off build, use option 1. If you're selling an ongoing care plan, option 2 can make sense, as long as your contract says who handles leads and data.

With ChatBento, each plan includes a set number of chatbots: one on Free and Hobby, three on Standard ($30 a month) and ten on Pro ($100 a month). For one client, the free plan or Hobby ($8 a month) is usually where to start. If you run bots for several clients from one account, check that the plan's monthly message allowance covers all of their traffic together.

Step 1: Gather the content in the kickoff call

The chatbot answers from the client's own content, so most of the setup is a content job, and it's best done while you're already asking the client questions for the site build. Add these to your kickoff questionnaire:

1. What are the ten questions customers ask most, by email, phone or DM?
2. What are your prices, and what's not included?
3. Shipping, booking, cancellation and refund rules, including the exceptions.
4. What don't you do? ("We don't ship outside the UK." "We don't do weddings.")
5. When someone needs a person, what should happen: an email, a booking link, a phone number?
6. Who should receive leads from the chat, and how fast will they reply?

Questions 4 and 5 are the ones clients forget. A bot that doesn't know what the business doesn't do is the one that says yes to everything.

Step 2: Build the bot

Once the site has content on it:

• Point the bot at the site so it reads the pages. Do this after the copy is final, or update it after launch if the copy changes.
• Add the answers from the kickoff call as Q&As. These matter more than the pages, because they cover what the site doesn't say.
• Match the brand. Set the accent color, avatar (the client's logo works well), greeting and suggested questions. A greeting that names the business and what the bot can help with ("Questions about bookings or prices? Ask me.") gets more replies than a generic hello.
• Pick the corner. Bottom-right is the default. Use bottom-left if the client's design already has a cookie banner or a floating button there.

Step 3: Install it on the client's platform

The snippet is one line. Where it goes depends on the builder:

Step-by-step guides for these and other platforms are on the integrations page.

One thing that catches people out on client work: staging domains. If you limit the bot to allowed websites, which you should, so nobody can embed the client's bot elsewhere, add both the client's custom domain and the staging address (*.framer.website, *.webflow.io) while you test. Otherwise it will work on one and not the other, and you'll lose half an hour wondering why.

Step 4: Test like the client's customer

Before you hand it over, ask it:

1. The ten questions from the kickoff call, worded the way a customer would write them
2. Three questions the site doesn't answer, to check it says so instead of guessing
3. One request for a human, to check the handoff and that the lead arrives in the right inbox
4. One off-topic question, to check it politely declines

Show the client the results. Watching their own bot answer their own questions correctly is the moment the value clicks for most clients.

Step 5: Hand it over properly

A chatbot fails quietly after launch when nobody is responsible for it. At handover, agree in writing:

• Who receives leads, and the reply time the client is committing to.
• Who reads the conversations, and how often. Weekly for the first month is a good start.
• Who updates the content when prices, services or policies change.
• That the client's privacy notice mentions the chatbot. It's the client's website and the client's visitors, so it's their notice. I covered what it should say in Website Chatbots and GDPR.

Also show the client that the chat says it's an AI assistant, and why that matters. ChatBento's widget labels itself "AI assistant" in the header; if you write the greeting, keep it honest too. The EU rules are summed up in Does Your Website Chatbot Need an AI Disclosure?

Turning it into a care-plan service

If you already sell maintenance plans, the chatbot gives you something concrete to do each month:

• Read the month's conversations and list the questions the bot couldn't answer.
• Add the missing answers as Q&As, and fix anything it got wrong.
• Update content after price or service changes.
• Send the client a short report: how many conversations, how many leads, the top questions, and what you improved.

That last item is what makes the service visible. Clients rarely log in to read transcripts themselves, but they'll read a one-page summary of what their customers asked.

How you price setup and care is your call, and it depends on your market and how much content work each client needs. Price the content work and the monthly review, not the script tag. The install takes minutes. The value is in the answers.

Mistakes to avoid on client sites

• Launching before the copy is final. The bot learns the old text and quotes it.
• Skipping the "what we don't do" answers. That's where invented promises come from.
• Forgetting staging domains in the allowed websites list.
• Leads going to your inbox, not the client's, with nobody forwarding them.
• No owner after launch. Decide who reads the conversations before you hand over.

A quick checklist per client

• Account ownership decided and written into the contract
• Kickoff questions answered, including what the business doesn't do
• Bot trained on the final site, plus Q&As
• Brand color, avatar, greeting and corner set
• Custom domain and staging domain in allowed websites
• Ten real questions, three unanswerable ones and a handoff tested
• Lead recipient and reply time agreed
• Client's privacy notice mentions the chatbot
• Monthly review scheduled

If most of your client sites are built in Framer or Webflow, the Framer guide and Webflow guide show exactly where the snippet goes, and the pricing page lists how many chatbots each plan includes.