AI chatbots and assistants for agencies

A white-label AI chatbot is an on-site or messaging assistant that we build and your agency delivers under its own brand. The client sees your agency’s work, and we never contact them.

What a white-label AI chatbot should and should not do

It should answer from the client’s own content and admit when it cannot. It should not guess a price, a policy or a delivery date.

We build it on retrieval. The client’s pages, product data, policies and help articles are indexed, and each answer is drawn from those sources with a link back.

Tone rules keep it in the client’s voice. Refusal rules cover what it must not answer, such as legal or medical advice, or a price the content does not state.

Edge cases matter more than the demo. Visitors try to make it say odd things, pages contradict each other, content goes out of date and people paste personal details into the chat. We test for each before launch and set a schedule for refreshing the index.

Our plugin or a custom build

For many WordPress clients our own plugin is the quicker route. A custom build fits when knowledge lives in several systems, the site is not WordPress, or the agent must carry only your brand today.

Situation	Best fit	Why
A WordPress site selling services or WooCommerce products	The free SoftVolt AI Voice Agent plugin	Voice and text chat, bookings, hand-offs and a leads inbox already exist. Speech runs in the visitor’s browser, so voice adds no cost.
The agent must show only your agency’s or the client’s brand now	A custom build	White label and agency mode are planned for the plugin’s Pro add-on, which is not available yet.
Knowledge sits in PDFs, a help desk or several systems	A custom assistant on the Claude or OpenAI API with vector search	Retrieval works across every source, with citations, inside a widget built in Next.js.
The client already uses a live chat tool	An integration into that tool	Their team keeps one inbox, and the assistant hands over inside it.

Measuring whether it works

Every assistant ships with an evaluation set: real visitor queries with the answers the client signed off. We re-run it whenever the content, prompt or model changes.

The monthly review shows what the assistant answered, what it refused and what it passed to a person. Unanswered queries become content tasks, which gives your team a list of pages worth writing.

We do not promise a deflection rate before launch. We agree what to measure at scoping and report it plainly.

Pricing, data and the wider picture

A custom assistant gets a written scope and a fixed price within two business days of the brief. Model usage is billed at cost or runs on an API account in your name or your client’s. See the current plans on our rates page.

Chat transcripts can hold personal data, so we name the model provider in the data processing agreement. For UK and EU visitors we work under the UK IDTA or EU Standard Contractual Clauses, as our security page sets out.

Assistants often sit inside a wider process, such as routing leads to a CRM. That work is covered by our white-label AI automation and workflow automation services.