A marketing plan can fail even when the idea is sound. They fail because the plan carried a quiet cargo of assumptions the team had stopped noticing. This guide sets out a working method for putting evidence and assumption into two visible columns before you commit budget, headcount or time.

Why the split matters before anything else

Advertising and public marketing statements are held to a truthfulness standard. The US Federal Trade Commission states plainly that under the FTC Act advertising must be truthful and non-deceptive, advertisers must have evidence to back up their claims, and advertisements cannot be unfair (FTC, Advertising FAQ's: A Guide for Small Business). The same guidance says a company has to have a "reasonable basis" for a claim before an ad runs, meaning objective evidence that supports the claim, and that the kind of evidence required depends on the claim (FTC).

That is the marketing-planning version of a simple discipline: the words you intend to say in market should be anchored to something you could show someone. If they cannot be, they belong in the assumptions column, not the evidence column.

What counts as evidence

Evidence, for planning purposes, is a statement you could hand to a skeptical colleague and defend without saying "well, we think so." Useful tests:

  • Does it exist outside your head? A document, dataset, recorded interview, contract term or published figure you can point to.
  • Is it specific to the claim? The FTC notes that the level of proof needed depends on the claim, and that the statement in the ad sets the bar at least as high as what it says — for example, a claim like "two out of three doctors recommend" must be supported by a reliable survey to that effect (FTC).
  • Would a customer accept it? The FTC cautions that testimonials from satisfied customers usually are not sufficient to support a claim requiring objective evaluation, and that a money-back guarantee is not a substitute for substantiation (FTC).

For AI-assisted work there is a second layer. NIST describes the AI Risk Management Framework (AI RMF 1.0) as intended for voluntary use and to improve the ability to incorporate trustworthiness considerations into the design, development, use and evaluation of AI products, services and systems (NIST, AI RMF). NIST also released a Generative AI Profile, NIST AI 600-1, which can help organizations identify unique risks posed by generative AI and proposes actions for generative AI risk management (NIST); the profile is described by NIST as centering on 13 risks and more than 400 actions (NIST AIRC, Technical Reports). None of this certifies an approach, but it does supply a vocabulary for asking which of your plan's premises are genuinely tested and which are projections.

What belongs in the assumptions column

An assumption is a belief you are relying on that has not yet been tested. Typical marketing examples: that a segment will care about a message; that a channel will reach the intended audience; that a conversion rate will hold; that an AI tool's output will be accurate enough for public use without human review. State each as a testable sentence, not a feeling. "Prospects will respond to a cost-saving framing" is an assumption. "We have three customer interviews that mention cost" is closer to evidence, provided the interviews actually say what you claim.

The two-column exercise

Take one proposed initiative and write two lists side by side.

  1. Evidence. What do we actually know, and where is it recorded?
  2. Assumptions. What must be true for this to work, and what would change our mind?

Then add the practical constraints — cost, owner, time available — and a single decision question, such as "should we run this in the next quarter or defer it?" The point of the exercise is not to score options but to make the gaps legible. A plan whose assumptions outnumber its evidence is not automatically bad; it is a plan that must be tested cheaply before it is funded heavily.

Hidden claims in AI-assisted work

AI-assisted plans often smuggle in claims without anyone noticing. A generated draft may assert a result, an audience size or a trend that nobody verified. Google's own guidance to content creators is useful as a check: its automated ranking systems are described as designed to prioritize helpful, reliable information created to benefit people (Google Search Central, Creating helpful, reliable, people-first content). The same documentation warns that fabricating creator profiles — AI-generated headshots, made-up names or false credentials — is a form of deception and a signal of a low-quality page (Google). It also notes that AI or automation disclosures are useful where someone might reasonably ask how content was created (Google).

This is guidance about a search engine's own systems, not an independent audit of any plan. The inference we draw is only this: a brief that quietly presents AI-generated assertions as findings is a brief that has contaminated its own evidence column.

Putting it into practice

The private Decision Brief worksheet is built for exactly this split. It sets up to three proposed initiatives side by side using only what you type in — evidence, assumptions, cost, ownership, time constraints and the decision question — produces no score, promises no outcome, and runs no external models at visitor runtime. It exports a private Markdown brief that leaves your unknowns visible rather than smoothing them away.

This essay sits alongside three others: start with a decision, compare initiatives without fake precision, and set a useful review cadence. Read the about page for what this publication is and is not.

A closing check

Before the brief goes to the people who control budget, ask of every line: could we show this, or are we taking it on trust? The FTC's reminder that its staff cannot clear ads in advance (FTC) is a good final caution. No one will review your reasoning for you. Keeping evidence and assumption separate is how you make that review possible.