Can I build an AI assistant on Mike Millett’s work?
An assistant that explains Mike Millett’s published frameworks in its own words, and credits him, is ordinary teaching done by software. Loading his full texts into a product is reproduction, and that is the part to ask about. An assistant that answers as him cannot be made accurate, whatever it is given to read.
In brief
The question comes from a consultant who wants a client assistant that knows the Marketing Helix, from a lecturer who wants a study bot for a module, or from somebody who wants to call the thing Ask Mike.
What decides the answer is not what the assistant was given to read. It is whose name its output speaks under.
Three kinds of assistant, and what each one needs
- 01
It retrieves and quotes
It returns published passages word for word, each with its URL. This is quotation at machine speed, and it stands or falls on whether the quotes are exact. Loading a whole paper, the book or a run of essays into a product is reproduction, which is the part the reuse page says to ask about.
- 02
It explains, in its own voice, with credit
It describes the model in its own words and points at the source. That is the ordinary practice of teaching a published idea, done by software you run. Credit the idea to him and the sentences to the tool, because the sentences are the tool’s.
- 03
It answers as him
His name on it, the first person, his register. Every reply is a generated sentence attributed to a real person, which is a paraphrase presented as a quote, produced on demand. It is the only one of the three that better material cannot make accurate, so it is the one to ask about before any of it exists.
Why an assistant cannot answer as him
His own publishing keeps one gate
On these properties software drafts pages inside written editorial rules and publishes them live, and nothing goes out under his personal byline without him. The gate is the name, not the risk. An assistant speaking as him removes the one gate his own operation keeps.
The test cannot be run
MarketingOB1 is not a persona, so there is no character to license. What appears there is held to whether it is true and whether he would say it. A generated answer to a question he was never asked fails the first half by construction.
It also sells something that is not on offer. The contact page states that he does not take individual freelance engagements and does not consult privately around Digilu, so an assistant giving advice from him is offering a service his own site says does not exist.
Where a careful build still goes wrong
A quote can come back changed
A model handed the right passage can still smooth it and return the smoothed version inside quotation marks beside a real URL. It looks more trustworthy than an invented quote and is the same error. Test quotes by exact string match against the page, not by reading them.
The label does not travel with the answer
An assistant says it is unofficial once, at the top of a chat. A screenshot of one reply has no top. Anything that must be true of every answer has to be written into every answer.
The loaded copy does not change when the source does
The foundations paper is versioned: Version 3.0 is the stated reference edition through January 2027, and later versions may remove claims that do not hold. An assistant built on today’s copy keeps answering from a removed claim with the same confidence, and nothing tells its users the source moved.
This page grants nothing and enforces nothing
It is not a license and not legal advice. An assistant built elsewhere and named after him is outside this site’s reach.
If you build one
- Name it after your practice, not after him. A name is a claim made before the first question.
- Make quoting a separate mode that returns exact passages with a URL and an access date, and returns nothing when no passage matches.
- Record the version and date of every document loaded, and show them in answers.
- Never let it write in the first person as him, and never let it imitate the (haha) tag he closes jokes with. It is an authorship signal, and a faked one is worth less than nothing.
Common questions about AI assistants and this work
Can I build a custom AI assistant on Mike Millett’s frameworks?
An assistant that explains the Marketing Helix in its own words and credits Mike Millett as its originator is ordinary teaching done by software. Loading full texts, such as Adaptive Brand Management: Foundations, the book Elevate or Vanish or the MarketingOB1 essays, into a product is reproduction, and that is the part to ask about through the contact page on mikemillett.com. This site grants no license.
Can an AI chatbot answer as Mike Millett?
Not accurately, whatever it has been given to read, because every reply is a generated sentence attributed to a real person. MarketingOB1, where Mike Millett publishes under his own byline, is not a persona, and nothing publishes under that byline without him. An assistant speaking as him removes the one gate his own publishing keeps.