AI and the Creation of Distinctive Assets

The paradox of generative AI in the face of Von Restorff

A major irony of the moment: generative AI produces, by construction, average content. Large language models are trained to minimize perplexity — to reproduce the most probable sentence. But the most probable sentence is precisely the dominant pattern, which is the opposite of the Von Restorff effect.

The consequence is that salespeople and marketers who simply prompt "write me a prospecting email" generate invisibility at scale. The more they send, the less they are remembered.

The bet of this chapter is the inverse: use AI not to produce probable content, but to map the dominant pattern and force the break in dimensions a single human would struggle to explore systematically.

Method 1: automated dominant-pattern mapping

Before any production, you need to understand what is saturating the market. AI is excellent at digesting a large corpus of competitor communications and extracting regularities.

You are a semantic analyst. Below are 30 homepage texts from competing 
software products in the [industry] vertical:

[paste the 30 texts]

Your mission:
1. Identify the 10 most frequent words and expressions (token frequency).
2. Spot the 5 recurring promises (jobs to be done as worded).
3. List the 5 dominant syntactic structures (e.g., "The best 
   platform for X", "X simplified by Y", etc.).
4. Identify the 3 dominant emotional registers (technical, heroic, 
   collegial, etc.).
5. Identify the 3 notable absences: angles, words or emotions that 
   no competitor uses.

For each absence, propose a hypothesis explaining why it is absent 
(taboo, low effectiveness, unexplored, etc.) and assess its 
potential as a Von Restorff break vector.

Typical outcome: you obtain a list of empty semantic territories where the break is still available. The creative job shifts from "come up with a hook" to "occupy a vacant memory territory."

Method 2: the option generator via inverted constraints

Most prompts ask the model to "be creative." That is a weak instruction. True breaks come from counter-intuitive constraints that force the model out of the statistical mean.

Generate 15 prospecting email openers for [solution] toward 
[persona], following these constraints (anti-average forcing):

- No opener starts with "Hello" or "Hi"
- No opener contains the words: "solution", "innovative", 
  "leader", "support", "delighted to"
- No opener mentions my company name in the first sentence
- 5 openers must start with a question
- 5 openers must start with a counter-intuitive statement
- 5 openers must start with an unexpected numerical fact
- No opener exceeds 35 words
- All must remain professional and credible

For each one, indicate:
- The Von Restorff vector activated
- Perceived risk (1-10)
- Predicted memorability (1-10)

This anti-prompt technique is probably the best use of an LLM for commercial creativity: do not ask for creativity, forbid the mean.

Method 3: distinctive visual generation

Image models (Midjourney, FLUX, Stable Diffusion) also have dominant patterns. A naïve prompt like "business meeting professional, photo realistic" yields the same bland corporate look as every stock library. For a Von Restorff visual:

Visual prompt for [context]:

Style: documentary reportage, natural light, film-grain, 
deliberate imperfections (slightly off focus, vignetting)
Composition: broken rule of thirds, off-center subject, deliberate 
60% negative space
Dominant color: muted monochrome palette (petrol blue, ochre, 
sepia) — avoid corporate saturation
Subject: [concrete, unusual scene that breaks the visual expected 
in the industry]
Anti-references: avoid LinkedIn-style renders (smiles, suits, 
handshakes, screens)
Format: 3:2, 4K

The dominant pattern of marketing visuals is the clean, centered, glossy render. The break runs through: raw, off-centered, matte. This does not work in every sector (luxury goods stay clean) — hence the importance of always charting the pattern first.

Method 4: Von Restorff naming

Your product or company name is the first Von Restorff element a prospect encounters. A weak name dilutes everything else.

Generation prompt:

You are a branding consultant. My product is: [description].
My sector is: [industry]. My 5 competitors are: [list].

Generate 30 names following these Von Restorff constraints:
- The name must not belong to the industry's lexical field
- The name must be pronounceable in both English and French
- The name should be available as .com and trademark-feasible
- 10 names inspired by scientific or anatomical words
- 10 names inspired by everyday objects unrelated to the field
- 10 names purely invented but phonetically memorable

For each name:
- Etymology or inspiration
- Distinctiveness score (1-10)
- Interpretive risk (misunderstanding)
- Approximate .com availability (to be verified manually)

Note: final usage and availability checks remain manual. AI is an option generator, not a legal validator.

Method 5: the "Von Restorff Score" for each deliverable

Before publishing, submit each deliverable to an AI evaluation that simulates the Von Restorff effect in its real context.

Here is my content: [paste the final version]
Here are 9 competing pieces of content seen in the same flow by 
the same persona: [paste the 9 pieces]

Play the role of a target reader scanning these 10 pieces in 
8 seconds.

1. Which content does your eye land on first? Why?
2. Which content holds attention beyond 3 seconds? Why?
3. Which one would you come back to if you could only revisit one tomorrow?
4. Is my content (number [X]) memorable? On which Von Restorff vector?
5. If not, what minimal change (1 sentence, 1 word, 1 visual) 
   would push it out of the pack?

Reason explicitly about contrast with the 9 others.

Method 6: AI to detect over-saturation

A real risk of applying Von Restorff at scale: the break itself becomes a pattern. When 200 marketers all use the same "no-hello" prompt, that new format becomes the dominant pattern and loses its effect. AI can monitor this drift.

Here are 50 prospecting emails my persona received in the last 30 days.

Analyze:
1. Are there "pattern breaks" that have become patterns themselves 
   (meta-saturation)?
2. Which formats were distinctive 6 months ago and are no longer?
3. Which vectors remain underexplored and therefore high-potential?

Give me a 3-point strategic report on the state of the attentional 
market for [persona].

Case study: the micro-SaaS that went from 0 to $4K MRR in 60 days

A real anonymized case observed in 2024. A B2B micro-SaaS for ticket management, launched into a market dominated by Zendesk, Intercom, and Freshdesk. Von Restorff strategy applied systematically with AI assistance:

AI did not replace the creative work — it industrialized contextual analysis and option generation via anti-prompt.

Ethical limits: the break in service of value

A necessary warning: the Von Restorff effect can be hijacked into a manipulation tool. A mediocre product with a shocking package gets remembered, but its broken promise destroys the brand long-term ("bad buzz" effect).

A break unsupported by the product experience consumes trust. Long-term, it is the consistency of peak + end (see Peak-End Rule) that turns captured attention into durable positive memory.

Summary

Generative AI defaults to producing average — and therefore invisible — content. To leverage the Von Restorff effect, the use must be inverted: chart dominant patterns at scale, force breaks through inverted constraints (anti-prompt), evaluate every deliverable inside its competitive context, and monitor for the saturation of the breaks themselves. The next chapter scales these principles up to the entrepreneur level: positioning, naming, packaging, pricing, and product architecture.