Legal

AI Policy

Last updated: September 2026

How to read this

Lemonwater advises on AI. We also use AI as a tool in that work. This page says what that means, where we have built in limits, and where responsibility sits if something is wrong. Read it with the Privacy Policy and Terms.

This is Lemonwater, LLC’s company-wide AI policy — it governs how we use automated systems in our own work, including AI consulting engagements. It applies alongside, and does not replace, the Privacy Policy and Terms of Service. Premix, our audio software, has its own supplementary policy at /premix/ai-policy, since it runs a different kind of automated system on different material. This page is the one that governs everywhere else.

1. Our position

AI is a tool. Like any tool, the way it is used determines the outcome. We are not here to make you dependent on it. We are here to make you fluent in it — including when the honest answer is not to use it.

We do not hide behind “the algorithm did it.” Advice we give is ours. If we use a model to draft, search, or summarize, we still own the recommendation. You still own the decision in your business.

2. What we will not do

  • Train on your work. Client materials are not used to train, fine-tune, or evaluate a model we ship.
  • Send your data out by default. We do not paste your files into a third-party generative-AI product unless the engagement writing says that will happen, and you can still say no.
  • Sell you a platform. We do not take a cut of token usage. If a vendor is the right fit, we will say so; we are not their channel.
  • Pretend a demo is a contract. Tools often fail to perform as advertised. That gap is usually the first thing we look at.

3. Voice input and LLMs

Speaking into a model is not the same as pasting a paragraph, and it is not the same as repairing a field recording. A voice note can identify a person. A transcript can still be sensitive. We do not treat either as free context for a chatbot.

  • Consulting. We do not paste your recordings or dictated notes into a third-party LLM unless the engagement writing says that will happen, and you can still say no. If we use a model to draft, we still own the recommendation.
  • No always-on capture. We do not run a listening assistant on your calls or set. Guidance, when we give it, is initiated by you or by us in the engagement — not by a device that overheard you.
  • Premix is not a voice tutor. Production audio is not sent to an LLM, and Premix has no voiceprint technology. See the Premix AI Policy.

4. Human oversight

Automated output can omit, invent, or skew. We treat it as a draft. You should too. We will not tell you a system is “safe” in the abstract. We will tell you what it touches, what it cannot see, and who is still accountable.

5. If you use Premix

Premix is Lemonwater software, not a consulting engagement, and it runs its own automated processing on material you upload. That is covered by a separate, more specific policy — read it at Premix’s AI Policy — since the commitments that matter there (the shipped model does not learn from your jobs; a job-local adapter may match this packet; the Mixer License is a client, not a host) are different in kind from a consulting engagement.

6. Changes

If we add a vendor that runs inference on your materials off our machines, or if we start using voice input with an LLM for guidance, this page and Privacy will be updated together, and we will say so before that happens on an engagement.

Questions: info@lemonwater.io. This page is a public description of how we work. It is not a warranty that every recommendation is error-free.