AI: Personalizing Choices to Shrink Decision Time

Why AI changes the game for Hick's Law

Until 2022, narrowing the number of options shown to each visitor required expensive manual work: marketing segmentation, A/B testing, custom landing pages. Past a hundred segments or so, the work became economically infeasible.

With modern LLMs and recommendation models, you can now:

• Detect in real time what interests a visitor
• Reduce dynamically the number of visible options
• Reorder options by individual relevance
• Personalize the wording of each option

A website that would show each visitor the right pair of options (their own, not someone else's) would convert 2 to 3× better than average. That is the goal of this chapter.

Detecting intent: from anonymous visitor to personalization

Before personalizing you have to infer intent. Multiple signals can be exploited:

AI aggregates these signals and infers an intent score by segment. Low-intent segments get a minimalist page (1 CTA, 0 distractions). High-intent segments get an expert page (detailed options).

The "less is more, but adaptive" pattern

Instead of showing 12 options to everyone, or 3 options to everyone, AI makes it possible to show 3 options picked from 12 — those that maximize click probability for this visitor.

graph LR
    A[12 possible options] --> B{Scoring model}
    B -->|Visitor A| C[Options 2, 5, 9]
    B -->|Visitor B| D[Options 1, 4, 11]
    B -->|Visitor C| E[Options 3, 7, 12]

Each visitor therefore stays in the optimal Hick zone (3 options), while the system as a whole covers all 12 segments. The classic "simplicity vs. exhaustiveness" trade-off disappears.

Pragmatic implementation in 4 steps

Step 1: map the option inventory

List, without taboo, every option present on your site: plans, features, resources, CTAs, demo modules, etc. You usually end up with 30 to 80 entries. That is normal.

Step 2: tag each option by relevant profile

For each option, attribute 1 to 3 target-profile tags. Examples:

• "Pro plan" → freelancer, agency
• "API module" → technical CTO
• "Marketing Director testimonial" → mainstream CMO

Step 3: build the scoring engine

SCORE(option, visitor) = Σ weighting(option_tag ∩ visitor_tag)

Simple rules are enough to start. Later, a supervised ML model (XGBoost, logistic regression) takes over.

Step 4: display the top 3 scoring options

That's it. No sorcery. The magic comes from the Hick × personalization synergy.

LLM prompt to generate scoring automatically

If you do not have a data team, a LLM can do the job — in JSON, instantly:

You are a web personalization expert.

Here is the full list of options offered on my website:
[PASTE THE LIST]

Here are the visitor profiles I want to target:
[FREELANCER / SMB / ENTERPRISE / DEVELOPER / MARKETER / EXECUTIVE]

For each profile, pick the 3 priority options to display,
with a justification per choice.

Response format:

{
  "freelancer": {
    "priority_options": ["...", "...", "..."],
    "justification": "..."
  },
  ...
}

At the end of the answer, add 3 metrics to monitor to
validate the personalization is effective.

This prompt produces a profile × option matrix in 10 seconds, directly usable by a front-end developer.

Personalize the wording, not only the list

A step often forgotten: rewrite each option to resonate with its target profile. Example:

The option is the same. The friction is not. A LLM can produce these variants:

You are a B2B copywriter.

Here is one option of my product:
[NAME]: [DESCRIPTION]

Produce 5 reformulations, one per profile:
1. Freelancer
2. SMB (5-50 employees)
3. Mid-market sales director
4. Enterprise CTO
5. CFO

Constraints: one sentence each, max 15 words, no marketing jargon.
Stay factual and concrete.

AI use in sales chatbots

Traditional chatbots used to show an 8-button menu. Classic Hick mistake. The new LLM-based chatbots have learned the lesson:

1. They ask one open question: "What are you trying to automate?"
2. They interpret the answer
3. They propose 2 binary paths: "More about sending emails or collecting signatures?"
4. They converge to the right demo or the right sales contact

This design cuts T from 30+ seconds to under 10 seconds per interaction. And the qualification rate jumps from 15% to 40%.

Personalizing pricing: ethics and limits

AI technically allows displaying a personalized price to each visitor. Legally that borders on discrimination (EU: DSA, DMA; France: DGCCRF). Ethically it is very contestable.

Recommendation: personalize options and wording, not the price. If a discount logic exists, trigger it by a deliberate visitor action (promo code, newsletter sign-up, quote request form) — never silently.

Use case: a Hick × AI optimized SaaS funnel

A SaaS publisher in financial tools (target: freelancers and SMBs) rebuilt its funnel along Hick + AI lines:

Overall result: site → trial conversion multiplied by 2.3.

Watching for drift: when AI narrows too much

Watch out for the trap: if AI filters too aggressively, it can lock a visitor into options that do not match their real need (filter-bubble effect). Three safeguards:

1. Always leave a "See all options" entry accessible
2. Log clicks on "See all": a rate above 15% signals poor scoring
3. Retrain the model regularly (monthly for high-traffic sites)

Auditing your site with AI: turnkey prompt

You are a UX auditor expert in Hick's Law.

Here is a textual capture of the page [URL]:
[PASTE CONTENT OR SECTION DESCRIPTION]

1. Count the number of first-level options (CTAs, menu links,
   plan choices).
2. Estimate decision time T for an average visitor (b = 0.17).
3. Sort options into 3 buckets:
   - Essential (must keep)
   - Secondary (move to a sub-menu)
   - Noise (remove)
4. Propose a reduced version with 3 visible CTAs maximum.
5. For each kept CTA, suggest an optimized wording in under 5 words.

Format: Markdown table + 3-sentence verdict.

Summary

AI is the first tool to let you satisfy two seemingly contradictory requirements at once: cover a large number of profiles, and show each visitor only 3 options. Dynamic personalization of choices (and of their wording) halves decision time and multiplies conversion accordingly. The tools are within reach of any startup thanks to LLMs, with no dedicated data team. The next chapter examines how to weave these principles into the product itself and the entrepreneurial strategy.