For its first CTV campaign, Jennifer Aniston’s DTC haircare brand LolaVie had a few non-negotiables. The campaign had to be simple. It had to demonstrate measurable impact. And it had to be full-funnel.

LolaVie used Roku Ads Manager to test and optimize creatives — reaching millions of potential customers at all stages of their purchase journeys. Roku Ads Manager helped the brand convey LolaVie’s playful voice while helping drive omnichannel sales across both ecommerce and retail touchpoints.

The campaign included an Action Ad overlay that let viewers shop directly from their TVs by clicking OK on their Roku remote. This guided them to the website to buy LolaVie products.

Discover how Roku Ads Manager helped LolaVie drive big sales and customer growth with self-serve TV ads.

The DTC beauty category is crowded. To break through, Jennifer Anniston’s brand LolaVie, worked with Roku Ads Manager to easily set up, test, and optimize CTV ad creatives. The campaign helped drive a big lift in sales and customer growth, helping LolaVie break through in the crowded beauty category.

Learn more.

Formal sentences. Clean structure. Technically correct. Zero personality.

You will likely read the output and immediately start rewriting. Every time. The AI will save you time on research but cost you time on editing.

To ensure your AI sounds like you, and not every other AI, you will need to give it a brain.

Think of your Jarvis as hiring a new employee. Even if you were to hire someone with tons of experience and formal education, you would still need to train them on how you operate. You would still need to invest in them, by spending time with them making sure they know your vision, your value, your differentiation, and so much more. Especially if you were hiring them to be your Chief of Staff. In order for them to perform at this role, they need to almost be hard-wired to your brain so they can help take as much off of your plate as possible.

While that is not possible with a human, it 100% is possible with Jarvis.

This is how that happened.

People blame the model. They say Claude is too formal, or ChatGPT writes like a press release, or the AI just cannot match their voice.

That is not the problem.

The problem is context.

AI generates text by predicting what should come next based on everything it knows. When you give it no specific context about you, it predicts based on patterns from the internet. The internet averages out to corporate, cautious, and verbose.

Give it specific context about you, and the prediction changes. It is not the model that is generic. It is what you fed it.

A knowledge base is structured context. It is the collection of files you give Jarvis before it writes anything. Voice rules, writing examples, framework definitions, terminology preferences, anti-patterns to avoid. All of it in plain text.

Every time Jarvis generates content or answers a question in your name, it reads the relevant files first. Then it writes with that context active.

The output reflects what the AI knows about you. A thin knowledge base produces thin output. A rich knowledge base produces output you would actually send.

Not every piece of context is equally important. These five files do the most work.

Voice guide. How you write. Sentence length patterns. Vocabulary preferences. Tone descriptors. What you never say. This is the single most important file. Without it, everything else is generic regardless of topic.

Exemplars. Examples of your actual writing that you would use as a benchmark. Five to ten pieces that represent you at your best. The AI reads these as "this is the target." The more specific and distinctive your examples, the better the calibration.

Anti-patterns. What you actively avoid. Clichés you hate. Phrases that feel off-brand. Structures that do not match how you think. Naming the anti-patterns is as important as describing the voice. The AI will use common patterns unless you explicitly exclude them.

Terminology. The specific words and phrases you use consistently. How you refer to concepts, frameworks, stages, and roles. If you call it a "champion" instead of a "sponsor," and "deal score" instead of "probability," that goes here.

Frameworks. Your proprietary ways of thinking about problems. How you structure advice. What mental models you apply. These give the AI your intellectual fingerprint, not just your stylistic one.

Five files. Build these before you build anything else.

Having the files is step one. Making sure the right files get loaded at the right time is step two.

Not every tool needs every file. A deal scoring tool does not need your LinkedIn post exemplars. A content drafting tool does not need your pricing framework.

Routing means: when tool X runs, load files A, B, and C. When tool Y runs, load files B and D.

In practice this looks like a simple map. Each tool or category of tools has a list of knowledge base files associated with it. Before the tool executes, those files get read and included in the prompt context.

This keeps your prompts focused. Loading everything every time wastes tokens and can dilute the output with irrelevant context. Loading the right subset produces cleaner, more relevant results.

The starting point is to load your voice guide and the relevant exemplars for every content-generating tool. Then layer in framework or terminology files as needed.

Here is what most people miss about knowledge bases.

The value compounds.

Every time Jarvis generates a post that misses the mark, I add the correction to the anti-patterns file. Every time I find a piece of writing that nails the voice, I add it to the exemplars. Every time I coin a new term or framework, it goes into the terminology file.

The files get richer over time. The output gets better over time. Automatically.

Six months in, Jarvis knows my voice better than I can articulate it consciously. Because the knowledge base has captured hundreds of small calibrations that I made in the moment and would never have written down otherwise.

A prompt library gives you good output on day one. A knowledge base gives you better output every week.

That is the compounding effect. It does not happen with prompting. It only happens with a persistent, updated knowledge base.

Do not start by trying to document everything.

Start with your voice guide.

Open a text file. Answer these questions:

That document is your voice guide. It does not need to be long. Three paragraphs is enough to start.

Then write or find five examples of your best writing. Emails, posts, newsletter issues, anything. Save them in one file.

That is your starting knowledge base. Two files. Ten minutes to create. The output shift is immediate.

Build from there. Add the anti-patterns file after your first week of using it. Add terminology as you catch inconsistencies. Add frameworks when you have a system worth documenting.

The knowledge base is never finished. It just gets better.

Paid subscribers get the templates below. Fill-in-the-blank versions of all five file types, plus the context routing pattern for wiring your knowledge base into your tools.

Talk next week.

Quickstart for paid subscribers | sellingwithai.vip

Fill-in-the-blank templates for all five knowledge base files. Plus the context routing pattern for loading the right files at the right time.

These are starting points, not finished products. The goal is to get something useful in place this week, not to build the perfect system.