HOW TO GET AI TO RECOMMEND YOUR PRODUCT?

WORKFLOW, PROMPTS, AND INSIGHTS

NEW YEAR, NEW TRENDS. ARE YOUR PRODUCT PAGES READY FOR 2026?

As we enter 2026, a crucial part of staying-on-top-of-things will be optimising for “Buyer Bots”. That’s right. AI is now your audience too. Sure it might not need shoes and cars, but it’s chatting to your consumer who is fed up with clicking away your banners and going through 4759450983204 landing pages to get to your product.

The marketing playbook you’ve been using so far is obsolete. You want to focus on optimising your PDPs (Product Description Pages) with the goal of having it picked up by LLMs and AI agents. Let’s break this down in simple human language.

WHAT ARE BUYER/SHOPPING BOTS?

AI-powered bots that help customers find products, compare prices, offer discounts, track stock, and even facilitate checkout on a website or through chat. Think AI agents but also LLMs recommending your product to someone who just asked ChatGPT “What are the best running shoes for running a marathon?”

WHAT DO BUYER BOTS LOOK FOR?

Bots are programmed to read the underlying code and structured data of web pages rather than interpreting the visual layout or lifestyle content that influences human shoppers. This allows them to process vast amounts of product information very quickly. Storytelling is great, but bots don’t give a shit. They target specific keywords, benefits and usage situations. They assess credibility and authority based on rigid standards of professionalism. Your tone of voice and brand personality have little influence here.

HOW DO I OPTIMISE FOR BUYER BOTS?

It starts with providing the right input [literally, always!]. You wanna imagine how someone might search for your product. What problem does your product solve? What is it used for? In what situations? What makes it special? Work these into your description copy and visuals. You wanna be detailed, thorough, and you wanna give your PDP lots of TLC (tender love and care).

HOW AM I SUPPOSED TO DO THAT AT SCALE?

Fair question, honestly. If you’re a major clothing brand, for example, you probably have millions of PDPs. How the hell are you gonna optimise each one for buyer bots? There’s an easy way to do this at scale using automated workflows with software like Make.com or Microsoft Power Automate.

This will help you elevate your PDPs and optimise them for buyer bots or LLMs. Remember: Good input leads to good output. Invest time into providing a liberal amount of information about the product for best results. I’ve based on this on an adidas running shoe, as an example, but it can be customised to many different types of products.

The Workflow

STEP 1 — TRIGGER

Product Manager Technical Intake

A Product Manager creates or updates a product record in a central Product Knowledge Database.

Required Input:

• Product category & subcategory
• Materials (upper / midsole / outsole)
• Physical specs (weight, size, geometry)
• Embedded technologies (plates, foams)
• Links to studio product images

Condition: Status set to “Ready for AI Enrichment”

STEP 2 — VALIDATION

IF / THEN: Input Validation

Checks for missing fields, vague entries, or inconsistent geometry data.

PASS
Proceed to Step 3

FAIL
Go to Feedback Loop

STEP 2a — LOOP

Validation Feedback Loop

Surfaces in-platform feedback to the PM explaining exactly what is missing and why it matters for AI understanding. The workflow pauses until data is corrected.

STEP 3

Technical Canonicalisation

Standardises units, normalises material names, and calculates derived values.

System Prompt

You are converting raw technical product input into a canonical technical profile. Your task is to: 1. Normalise units and terminology. 2. Standardise material and technology names. 3. Calculate derived values where possible (e.g., drop from stack heights). 4. Preserve uncertainty explicitly where data is incomplete. Rules: - Do NOT add new information. - If a value is derived, mark it as derived. - If a value is uncertain, include a confidence indicator. Output a structured object with: - normalised_specs - derived_specs (with calculation notes) - technology_list (standardised names) - confidence_notes

STEP 4

Tech-to-Advantage Analysis

Analyses functional effects, runner benefits, and trade-offs for each spec.

System Prompt

You are analysing technical product specifications to determine functional advantages. For EACH material, technology, or physical spec: 1. Describe the primary functional effect. 2. Translate that effect into a runner-level advantage. 3. Identify any known or likely trade-offs. Rules: - Base reasoning only on established biomechanical or performance principles. - Do NOT exaggerate benefits. - If a relationship is uncertain, state the uncertainty. - Do NOT reference brand marketing language. Output an array where each entry includes: - spec_or_technology - functional_effect - runner_benefit - trade_offs - confidence_level (high / medium / low)

STEP 5

Intended Use & Performance Goal Inference

Evaluates combined advantages to infer primary use, secondary uses, and non-recommended uses.

System Prompt

You are determining intended use cases for a performance product based on its technical advantages. Your task is to: 1. Identify the primary intended use. 2. Identify secondary acceptable uses. 3. Explicitly list non-recommended uses. Consider: - Weight - Geometry - Propulsion technologies - Cushioning characteristics - Trade-offs identified earlier Rules: - Do NOT assume a broad audience. - Be conservative if confidence is low. - Clearly distinguish between “optimised for” and “can be used for.” Output: - primary_use_case - secondary_use_cases - non_recommended_uses - reasoning_summary - confidence_level

STEP 6

Environment & Situation Mapping

Maps uses to real-world situations (e.g., Road vs Track, Competitive vs Recreational).

System Prompt

You are mapping intended product uses to real-world situations. For each identified use case: 1. Describe typical environments. 2. Describe typical usage situations. 3. Identify contextual constraints (pace, terrain, runner type). Rules: - Use concrete situations, not abstract descriptions. - Avoid lifestyle or emotional language. - Keep descriptions relevant to buyer decision-making. Output an array of: - use_case - environment - situation_description - constraints

STEP 7

Knowledge-Base Pairing Intelligence

Queries a shared Product & Outfit Knowledge Base to suggest contextual apparel, gear, and footwear rotation pairings.

STEP 8

Buyer-Query Anticipation & Recommendation Logic

Models how this product should surface in AI chats and under what conditions it should (or should not) be recommended.

System Prompt

You are preparing recommendation logic for AI shopping and advisory systems. Your task is to: 1. Identify likely buyer questions or queries. 2. Determine when this product should be recommended. 3. Determine when it should NOT be recommended. Rules: - Use natural language query patterns. - Avoid keyword-style phrasing. - Include conditional logic where appropriate. Output: - query_type - example_query - recommendation_rationale - recommendation_conditions - rejection_conditions

STEP 9

Bot-Friendly Product Description Assembly

Assembles all structured intelligence into a machine-legible, human-readable description.

System Prompt

You are generating a bot-friendly product intelligence description. Your task is to synthesise all prior analysis into a structured description that: - Explains what the product does - Explains who it is for - Explains when it should be used - Explains trade-offs clearly Rules: - Prioritise clarity over persuasion. - Use explicit cause-and-effect reasoning. - Avoid marketing superlatives. - Structure content so it can be parsed by AI systems. Output sections: - concise_summary - key_benefits (linked to specs) - ideal_use_cases - environments_and_situations - pairing_suggestions - limitations_and_trade_offs

STEP 10

Studio Image Semantic Tagging

Existing studio images are classified by angle and detail, then linked to specific benefits.

STEP 11

In-Situ Image Generation

Generates lifestyle imagery aligned with usage situations and goals.

STEP 12

Human Review & Approval Gate

A human reviewer verifies accuracy and claims.

APPROVED
Proceed

REJECTED
Return to Step

OUTCOME: ELEVATED PDP

Distribution & Activation

Approved intelligence is published to PDPs, AI shopping assistants, retail feeds, and image libraries.

---

WELCOME TO 2026

With this workflow (first draft) you’re ready to step into 2026 and smash targets. Ideally, you’d want to eventually train a team of AI agents who will carry out these tasks more efficiently and using less AI tokens. This will help you handle a bigger volume, decrease generation time, and use less energy. Training AI agents will take time though.

Need help building or implementing a similar workflow for your organisation?

Book a call Email me

Sharing is caring...