World Model Readiness
Engraved functions instrument

For your company

AI in the Functions

Module · When the brand ships on autopilot

The AI-in-Marketing Check

AI can write a month of posts before lunch, and that is exactly the problem. The bottleneck moves from production to judgement: what is on brand, what is true, what is worth publishing at all. This module checks the five controls that keep AI a force multiplier instead of a slop machine: brand guardrails the model can follow, a human gate before publish, a way to catch generic filler, consistency across channels, and outcomes you can actually measure.

Question 1 of 5 · Brand rules are usable

Does your AI have brand guardrails it can actually follow?

A model does not know your voice, your claims policy or your no-go words unless you tell it, every time. Guardrails written as a reusable brief, style guide and banned-claims list are the difference between content that sounds like you and content that sounds like everyone.

Question 2 of 5 · A human approves publish

Does a person review AI content before it reaches customers?

AI is confidently wrong at scale: fabricated stats, off-brand claims, tone that misreads the moment. A named approval step before anything goes live is the gate between a fast draft and a public mistake. Which channels have that gate is a decision you make, or one made for you the first time something slips.

Question 3 of 5 · Slop gets caught

Can you tell the difference between good AI content and generic filler?

The danger is not obvious errors, it is competent mediocrity: on-topic, grammatical, and utterly forgettable. Volume without a quality bar floods your channels with content that says nothing and trains your audience to scroll past you.

Question 4 of 5 · Channels stay consistent

Does your brand sound like one company across every AI-assisted channel?

Different people, tools and prompts drift into different voices: the ads sound bold, the emails timid, the social posts like a stranger. Without a shared source of truth the AI amplifies the fragmentation, one channel at a time.

Question 5 of 5 · Outcomes are measured

Do you know whether AI content actually performs, or just that there is more of it?

More output feels like progress and proves nothing. The question that matters is whether the AI-assisted content converts, engages and sells as well as what it replaced. Without that comparison you are scaling activity, not results.

For the statistics · one click each

Three questions for the public picture

These do not affect your score. They feed the anonymised, aggregated statistics; groups under 8 respondents are never shown.

What share of your marketing content is now AI-assisted?

None
Under a quarter
A quarter to half
More than half
We do not track it

Does a human approve AI content before it is published?

Always
Only for high-stakes pieces
Sometimes
It publishes automatically
No AI content yet

How does AI-assisted content perform against what it replaced?

Measurably better
About the same
Measurably worse
We do not measure it
Too early to tell

Your context

Used to calibrate the report. Company size and sector remain in the anonymized dataset; your email does not.

What the five levels look like

Every dimension in this assessment is scored 1 to 5. This is what the levels mean, dimension by dimension. The graded report diagnoses where your own answers land and what to do about it.

Brand rules are usable

  1. 1Nothing written
  2. 2Vague voice notes
  3. 3Style guide exists
  4. 4Brief plus rules
  5. 5Enforced in tooling

At the low end: Without written brand rules the model defaults to the average of the internet, and so does your brand. Write a one-page voice brief and a banned-claims list this week; it is the cheapest guardrail you have. What good looks like: Guardrails enforced in the tooling mean the brand holds even when the intern is driving. Review them as the voice evolves; a frozen brief slowly drifts out of date.

A human approves publish

  1. 1Auto-published
  2. 2Spot-checked rarely
  3. 3Reviewed when remembered
  4. 4Review before publish
  5. 5Tiered by risk

At the low end: AI content going live unread is a correction, a screenshot and an apology waiting to happen. Put one human approval step in front of every published channel before the next campaign runs. What good looks like: Risk-tiered review lets low-stakes copy move fast while claims and campaigns wait for a human. Keep the tiers honest; the cost of a bad claim does not shrink because you were busy.

Slop gets caught

  1. 1Volume is the metric
  2. 2No quality bar
  3. 3Editors push back
  4. 4Explicit standard
  5. 5Standard plus measurement

At the low end: When output volume is the goal, slop is the result and nobody notices until engagement dies. Define what good looks like for your brand and reject anything that only clears the grammar bar. What good looks like: A written standard paired with engagement data tells you whether the bar is real or just aspirational. Feed the signal back into your prompts; slop is a fixable input problem.

Channels stay consistent

  1. 1Every channel differs
  2. 2Drifts by author
  3. 3Loose alignment
  4. 4Shared brief
  5. 5One governed voice

At the low end: A brand that sounds like five companies reads as none. Point every channel at one shared voice brief so the AI works from the same source instead of inventing its own. What good looks like: One governed voice across channels is what makes AI scale the brand instead of diluting it. Audit the channels quarterly; drift creeps back in wherever a new tool or hire appears.

Outcomes are measured

  1. 1Only count volume
  2. 2Vanity metrics
  3. 3Some tracking
  4. 4Performance compared
  5. 5Attributed to outcomes

At the low end: Counting how much you published tells you nothing about whether it worked. Start comparing the performance of AI-assisted content against your old baseline on one real metric that ties to revenue. What good looks like: Content attributed to real outcomes lets you scale what works and kill what does not. Keep the loop tight; the model only improves if the results actually reach it.