AI Content Strategy for Ecommerce: 10 Steps to Win Product Citations and AI-Referred Traffic

Intro

Crawl-data analysis shows GPTBot now makes 3.6 times more requests to websites than Googlebot, and Google itself acknowledges a growing share of visitors arriving from AI systems. The implications for ecommerce are direct: buyers increasingly ask ChatGPT and Perplexity for product recommendations, category comparisons, and "best X for Y" answers — and the brand cited in that response captures the referral without a single click.

This is why an AI content strategy for ecommerce is no longer optional. This guide walks through the complete process: optimizing product pages for AI answers, building category content that earns citations, structuring FAQ pages that answer engines trust, and measuring AI-referred traffic with tools that track visibility across Google, ChatGPT, Perplexity, Gemini, and Copilot. Alef's position as an AI visibility engine provides direct vantage on which ecommerce content patterns actually earn citations — patterns detailed in its analysis of what an AI visibility engine measures and how AI crawlers differ from Googlebot.

Expect 20–30 hours for initial setup, requiring intermediate SEO or marketing skill, a published store, analytics access, and a documented list of buyer questions.

When You Need an AI Content Strategy for Ecommerce

The trigger scenario is now familiar to many ecommerce teams: a prospective buyer asks ChatGPT or Perplexity for the best standing desk under $500, and the answer lists three competitors — none of which is the brand in question. The purchase decision begins and ends within that response, before a single Google search result is ever consulted. The loss is silent, instantaneous, and invisible to traditional rank tracking.

Several concrete signals indicate this is already happening. Organic click-through rates decline as AI Overviews absorb navigational and product-research queries. Competitor brand names appear consistently in AI-generated answers while the brand's do not. And in analytics platforms, a small but growing slice of referral traffic arrives with source labels such as chat.openai.com or perplexity.ai — visitors who bypassed search entirely.

Ecommerce is uniquely exposed to this shift because product research constitutes a high-intent phase where answer engines consolidate comparison, specification, and pricing information into a single response. The model becomes the new storefront, and brands absent from its output forfeit consideration at the moment of maximum purchase intent.

Single-engine checks provide an incomplete picture. Each answer engine draws from different source sets and ranking methodologies, so visibility must be measured across ChatGPT, Perplexity, Gemini, and Copilot simultaneously. A brand cited by one engine may be entirely absent from another. Alef's analysis of AI-referred traffic patterns for ecommerce quantifies the revenue stakes, while its comparison of AI search visibility versus Google rankings demonstrates why traditional position tracking no longer suffices.

Steps: Build Your Ecommerce AI Content Strategy

Executing an AI content strategy for ecommerce requires a methodical sequence of actions, each building on the previous one. The following eight steps move from diagnosis to deployment, giving online stores a repeatable process for earning citations in ChatGPT, Perplexity, and Gemini. Time investment for the full sequence ranges from 40 to 60 hours for a store with 50–200 products, depending on the depth of the existing content library and the number of priority pages identified in Step 3. No advanced technical skill is required beyond basic familiarity with a content management system and access to Google Search Console.

1. Audit current AI visibility across answer engines

The first step is establishing a baseline. Without knowing whether ChatGPT, Perplexity, and Gemini currently cite the brand, any optimization effort is guesswork. The audit requires running the store's 5–10 highest-intent buyer prompts through each of the three major answer engines and recording the results systematically.

High-intent buyer prompts are questions a shopper asks when close to a purchase decision. For a store selling ergonomic office chairs, these include "best ergonomic chair for lower back pain," "Herman Miller Aeron vs Steelcase Gesture," and "what is the best chair for an 8-hour workday." For a skincare brand, the equivalents are "best moisturizer for sensitive skin," "ceramide vs hyaluronic acid for dry skin," and "is this brand cruelty-free."

For each prompt, run the query in ChatGPT, Perplexity, and Gemini, then record one of three outcomes: the brand is cited by name, its content is cited without the brand name, or it is completely absent. The distinction matters. A citation by name indicates strong entity recognition; a content citation without the brand name suggests the material is useful but the entity is not well established.

The same prompts should then be run against 3–4 direct competitors. This benchmarking reveals the competitive gap. If a competitor appears in six of ten responses and the store appears in zero, the content gap is clear. If both appear in three, the differentiator becomes content quality and freshness rather than mere existence.

Alef's visibility monitoring across ChatGPT, Perplexity, and Gemini can automate this tracking over time, converting what would otherwise be a manual weekly exercise into a continuous measurement of citation share. The audit output should be a simple table: prompt, engine, brand presence, competitor presence, and the source URL the engine cited. This document becomes the reference point for measuring progress after each subsequent step.

2. Map buyer questions to the purchase funnel

Once the baseline is established, the next task is understanding how shoppers phrase questions to AI assistants versus how they type into Google. The phrasing difference is substantial and directly affects content structure.

Conversational queries directed at AI assistants typically run 7–10 words, while Google searches average 3–4 words. A shopper types "best running shoes" into Google but asks ChatGPT "what are the best running shoes for flat feet with arch support?" The AI version contains context — the foot condition and the desired feature — that changes which content answers the question.

The mapping exercise separates prompts into three funnel stages:

• Discovery prompts — broad category questions: "what is the difference between LED and OLED monitors," "how do I choose a coffee grinder." These map to educational content and category pages.
• Comparison prompts — mid-funnel evaluations: "Sony WH-1000XM5 vs Bose QuietComfort Ultra," "is a mechanical keyboard worth it for typing." These map to comparison pages and spec tables.
• Purchase-intent prompts — decision-finalizing questions: "is [brand] worth the price," "does [product] come with a warranty," "where can I buy [product] with free shipping." These map to product pages, FAQ blocks, and policy pages.

For each of the store's top 20 products, the exercise involves documenting three prompts per funnel stage. The documentation should capture the exact wording shoppers use, not the wording the brand would prefer. Real phrasing comes from customer service emails, live chat transcripts, and the "People also ask" boxes on Google search results pages for the store's category terms.

The output is a question map: a spreadsheet where each row is a product or category, each column is a funnel stage, and each cell contains the exact conversational prompts shoppers use. This map drives content prioritization in Step 3 and FAQ development in Step 6. The questions themselves also reveal gaps — if shoppers consistently ask about shipping times and the store has no content answering that question, the map has identified a citation opportunity.

3. Prioritize product pages for answer readiness

Not every product page deserves equal optimization effort. The stores that win AI citations focus on their highest-margin, highest-traffic, or most competitively differentiated products first. A store with 500 SKUs should not rewrite all 500 product descriptions in a single pass; it should identify the 20–30 products that drive 80 percent of revenue and prioritize those.

For each priority product, the page needs a concise, entity-rich summary in the first 100–150 words. This opening block is what answer engines extract when they cite a product page. It must state, in plain language an AI can parse: what the product is, its category, its key specifications, and its differentiators.

A weak opening reads like marketing copy: "Introducing the Aurora Desk, the ultimate workspace solution for modern professionals who demand style and functionality in equal measure." An answer-ready opening reads like a spec sheet written in prose: "The Aurora Desk is a standing desk in the electric height-adjustable desk category. It has a dual-motor lift system with a 28- to 48-inch height range, a 350-pound weight capacity, and a bamboo surface. It differs from the company's Motion Desk in its integrated cable management and memory presets for four height settings."

The second version works because it names the category ("standing desk," "electric height-adjustable desk"), provides extractable specifications (height range, weight capacity, material), and draws a clean comparison against a sibling product. Answer engines can parse these attributes and reproduce them in a response without editorializing.

The summary block should also include the product's entity relationships: the brand name, the product line, and the category in consistent phrasing. If the brand is "Alef Office" and the product line is "Aurora," those exact names should appear in the summary. Inconsistency — sometimes "Alef," sometimes "Alef Office," sometimes "the company" — weakens entity recognition.

The remaining product page content — detailed specifications, shipping information, warranty terms, and customer reviews — supports the summary. Answer engines that cite the page for a purchase-intent question will pull from these sections. The goal is ensuring every question a shopper asks about the product has an answer somewhere on the page, in plain text that a crawler can access.

4. Add structured data for machine-readable product attributes

Structured data is the layer that makes product information unambiguous for crawlers like GPTBot and PerplexityBot. While the prose summary in Step 3 helps human readers and language models, schema markup gives machines explicit signals about what each piece of content means.

The schema.org vocabulary provides several types relevant to ecommerce product pages. The Product schema documentation defines properties for name, brand, category, description, sku, offers, and aggregateRating. Implementing Product schema requires embedding JSON-LD in the page's head section with these properties populated from the store's product database.

Four schema types matter most for AI citation:

• Product — the core entity definition: name, brand, image, description, sku, offers with price and availability.
• AggregateRating — the average review score and review count, which answer engines frequently cite when responding to "is this product worth it" prompts.
• FAQPage — question-and-answer pairs displayed as structured data, which enables rich results and gives answer engines pre-formatted content to cite.
• BreadcrumbList — the category hierarchy (Home > Office Chairs > Ergonomic > Aurora Desk), which reinforces entity relationships.

Implementation requires either a schema plugin for the store's platform or custom JSON-LD injected via the theme. Most ecommerce platforms — Shopify, WooCommerce, BigCommerce — have plugins that generate Product and BreadcrumbList schema automatically. FAQPage schema typically requires manual setup because it depends on the FAQ content created in Step 6.

Verification is the non-negotiable final action. Google's Rich Results Test documentation describes how to validate that the markup parses correctly and qualifies for rich results. The test accepts a URL or code snippet and reports errors, warnings, and detected rich result types. A page with invalid schema is worse than no schema, because it signals to crawlers that the structured data cannot be trusted.

The crawl behavior difference is measurable. Search Engine Journal's analysis of ChatGPT vs Googlebot crawl data shows that AI crawlers prioritize pages with clear structure and machine-readable content. Pages with proper schema are more likely to be crawled thoroughly and cited accurately, because the crawler can extract attributes without guessing.

5. Build category content that answers comparison prompts

Category pages are the natural home for comparison content, and comparison content is the format answer engines most frequently cite for mid-funnel questions. When a shopper asks "Sony vs Bose noise cancelling headphones," the AI response typically synthesizes information from a comparison article, a spec table, or a review roundup. Stores that publish this content in their category structure position themselves as the source.

The two content formats that earn citations are "X vs Y" pages and "best Z for [use case]" pages. Both require a specific structure to be citable:

"X vs Y" pages compare two specific products or brands. The page should open with a direct answer — "For most buyers, the Sony WH-1000XM5 is the better choice because of its superior noise cancellation, while the Bose QuietComfort Ultra wins on comfort for extended wear" — followed by a spec comparison table. The table format matters: answer engines extract tabular data reliably and reproduce it in responses. Columns should include price, weight, battery life, noise cancellation rating, and unique features. The page should close with clear recommendations for specific use cases: "Choose the Sony for air travel, the Bose for office use."

"Best Z for [use case]" pages rank products within a category for a specific scenario. The page should state the ranking criteria upfront, then present each product with its specs, price, and why it earned its position. The format "Best Running Shoes for Flat Feet" with five products ranked and explained gives answer engines a structured list to cite.

Category hubs tie these pages together. A store selling audio equipment should have a category page for "Headphones" that links to "Sony WH-1000XM5 vs Bose QuietComfort Ultra," "Best Noise Cancelling Headphones for Air Travel," and "Best Headphones for Phone Calls." The hub establishes the category entity, and the comparison pages answer specific prompts.

The internal linking structure matters as much as the content itself. Comparison pages should link to the product pages they discuss, using the product name as anchor text. This passes entity signals and gives answer engines a clear path from the comparison answer to the product page where a purchase can occur.

6. Create FAQ pages that earn citations

FAQ content is the most direct way to answer the purchase-intent prompts documented in Step 2. When a shopper asks an AI assistant "does [brand] offer free returns," the assistant needs a source that answers that exact question. A well-structured FAQ page provides that source.

The format rules for citable FAQ content are specific. Each question should be answered in 40–60 words — long enough to be substantive, short enough to be extracted verbatim. The answer should be a complete sentence or two that directly addresses the question without requiring additional context.

The phrasing of the question must match how shoppers actually ask. The question map from Step 2 provides this phrasing. If shoppers ask "how long does shipping take," the FAQ question should be "How long does shipping take?" not "What is your shipping timeline?" The exact-match phrasing signals to answer engines that this content directly addresses the prompt.

Each FAQ block should answer one question only. Combining two questions into one block ("How long does shipping take and what does it cost?") forces the answer engine to parse and separate, reducing the likelihood of a clean citation. One question per block, 40–60 words per answer, exact conversational phrasing.

FAQPage schema markup is the technical requirement that makes this content machine-readable. The schema defines a Question entity with the acceptedAnswer property. Implementing it requires adding JSON-LD to the FAQ page with each question-answer pair structured according to the schema.org FAQPage specification. The markup enables rich results in Google and gives AI crawlers a parseable structure.

The placement of FAQ content matters. A dedicated FAQ page for the store's policies (shipping, returns, warranty) answers the most common purchase-intent prompts. Product-specific FAQs — "Does the Aurora Desk support dual monitors?" — belong on the product page itself, in the section following the spec table. This placement ensures the FAQ content is contextually linked to the product entity it answers questions about.

7. Publish original data and first-party research

Original data gives answer engines a citable reason to choose the store over competitors. When ChatGPT responds to "what percentage of online shoppers abandon their cart," it cites sources with actual statistics. Stores that publish unique survey data, usage statistics, or industry research position themselves as authoritative sources that answer engines prefer.

The data does not need to be academic-grade research. An ecommerce store has access to first-party data that no other source possesses: customer usage patterns, product performance metrics, category trends from sales data, and survey responses from its own customer base. Publishing this data in a structured format creates content that cannot be found anywhere else.

Examples of citable first-party research for an ecommerce store:

• A survey of 1,000 customers about their buying preferences, with the methodology stated and the raw percentages published.
• Usage statistics from connected products — a smart home store publishing "how the average customer uses smart thermostats" based on anonymized device data.
• Category trend analysis based on the store's sales data — "the 10 fastest-growing kitchen gadget categories in 2025" with year-over-year percentages.

The format for citable data content follows a consistent pattern: a headline stating the finding, an opening paragraph with the key statistic and its source, a methodology section explaining how the data was collected, and tables or charts presenting the full dataset. Answer engines cite the statistic and the source URL, giving the store a citation and the accompanying referral traffic.

Original data also strengthens the store's E-E-A-T profile — Experience, Expertise, Authoritativeness, and Trustworthiness. Answer engines evaluate source reliability when deciding what to cite. A store with published research and transparent methodology ranks higher as a trustworthy source than a store with only marketing content.

The data content should live on the store's domain, not on third-party platforms. A blog post or research page on the store's own site ensures the citation points back to the store. The content should be linked from relevant category and product pages, reinforcing the connection between the data and the products it relates to.

8. Strengthen entity clarity and internal linking

The final step ties the content architecture together through consistent entity representation and a clear internal linking structure. Answer engines build knowledge graphs from the relationships they detect between entities — brands, products, categories, and attributes. Inconsistent naming weakens these relationships.

Entity clarity requires consistent naming across the entire site. The brand should always be referred to by its full name, the product line by its exact name, and categories by standardized terms. If the store sells "standing desks" and "height-adjustable desks" interchangeably, answer engines may treat them as two separate categories. The store should choose one canonical term per entity and use it consistently across product pages, category pages, FAQ content, and blog posts.

Internal linking reinforces entity relationships by creating paths between related content. The structure should follow a hub-and-spoke model:

• Category hubs (e.g., "Ergonomic Office Chairs") link to all product pages in the category and to relevant comparison and FAQ content.
• Product pages link to their category hub, sibling products, and product-specific FAQs.
• Comparison pages link to the products they compare and to the category hub.
• FAQ pages link to the products and policies they address.

Each link should use descriptive anchor text that names the destination entity. "Compare the Aurora Desk with the Motion Desk" is a functional anchor; "click here" is not. The anchor text itself reinforces the entity relationship.

The XML sitemap requires a final audit. Every page that deserves citation — product pages, category hubs, comparison pages, FAQ pages, and research content — must appear in the sitemap. Pages excluded from the sitemap are less likely to be crawled by AI crawlers, which often begin their crawl from the sitemap file. The sitemap should be submitted through Google Search Console and kept current as new content is published.

The verification for this step is a crawl test. Running the store's key pages through a crawler or using the URL inspection tool in Google Search Console reveals whether AI crawlers can access and parse the content. If GPTBot and PerplexityBot are blocked by robots.txt or the pages return errors, the entire content strategy fails at the delivery layer. The robots.txt file should explicitly allow AI crawlers, and the sitemap should be referenced in the file for discoverability.

The process for optimizing content for AI search engines extends beyond these eight steps into ongoing measurement and iteration. Tracking which pages earn citations and what AI-referred traffic results provides the feedback loop that refines the strategy over time. The stores that treat AI visibility as a continuous process, not a one-time project, capture the compounding benefits of being the established source in their category.

Common Mistakes in Ecommerce AI Content Strategy

Even a well-intentioned AI content strategy for ecommerce can falter when brands apply traditional SEO habits to answer engines. The following mistakes surface repeatedly in visibility audits, and each has a corrective action grounded in how AI systems actually retrieve and cite information.

Mistake 1: Optimizing exclusively for Google keywords. Conversational AI queries rarely mirror typed search terms. Buyers ask ChatGPT or Perplexity full questions, such as "What is the best waterproof Bluetooth speaker under $100?" A strategy built solely around short-tail keywords leaves those queries unanswered. The fix involves adding long-tail, question-phrased content that matches how buyers phrase requests to AI assistants, including natural language variations and comparison-oriented phrasing.

Mistake 2: Publishing thin product descriptions. Answer engines require extractable facts to cite a source. A 50-word description with no specifications, materials, dimensions, or differentiators gives an AI nothing to reference. The fix is writing entity-rich summaries that include measurable attributes, unique selling points, and compatibility details, giving answer engines concrete data points to pull into responses.

Mistake 3: Treating all answer engines as identical systems. ChatGPT, Perplexity, Gemini, and Copilot each index content differently and cite from distinct source pools. Crawl data analysis shows meaningful divergence in how these systems access web content. The fix requires testing identical prompts across each platform and comparing which pages surface, then adjusting content priorities accordingly.

Mistake 4: Neglecting structured data markup. Without Product, FAQPage, and BreadcrumbList schema, answer engines struggle to classify content as eligible for citation. Schema.org documentation outlines the required properties for each type. The fix involves implementing structured data across product and FAQ pages, then validating markup through Google's Rich Results Test to confirm error-free implementation.

Mistake 5: Measuring only clicks and rankings. Traditional analytics overlook AI-referred traffic and zero-click visibility, where a brand appears in an answer without generating a visit. Google has acknowledged receiving more visitors from AI systems, yet many brands still ignore this channel entirely. The fix is tracking AI referrers in analytics platforms and monitoring brand mentions within generated answers, not just session counts.

Mistake 6: Creating content without a documented buyer question map. Content produced without a structured list of customer questions lacks the specificity answer engines reward. The fix is building a prompt and topic map tied to each stage of the customer journey before writing begins, ensuring every piece of content answers a real query a buyer would pose.

Checklist for Avoiding These Mistakes

• Audit conversational query coverage. Review existing content against question-based phrasing and add long-tail variations where gaps exist.
• Enrich product descriptions with entity data. Include specifications, dimensions, materials, and differentiators in every product summary.
• Test prompts across multiple engines. Run identical queries through ChatGPT, Perplexity, Gemini, and Copilot to identify which content each engine prefers.
• Implement and validate structured data. Add Product, FAQPage, and BreadcrumbList schema, then confirm markup passes the Rich Results Test.
• Track AI-referred traffic separately. Segment analytics to isolate visits originating from AI platforms and monitor zero-click brand mentions.
• Document buyer questions before writing. Create a question map organized by customer journey stage and reference it during every content brief.

The underlying pattern connects each mistake: treating answer engines as a secondary consideration rather than a primary content audience. Brands that understand why a brand might be invisible in AI answers typically discover the cause traces back to one of these six errors. Evaluating answer engine optimization tools and their selection criteria helps identify which measurement gaps need closing first.

Summary Table: Ecommerce AI Content Strategy Steps and Outcomes

The following table consolidates the complete 10-step process into a single reference. Each row specifies the primary action, the measurable outcome, and the verification signal that confirms successful execution. The table is designed to function as a standalone infographic for internal alignment or stakeholder communication.

The verification signals in the right column provide objective checkpoints. When a step's signal is not met within the stated timeframe, the underlying content or technical implementation requires revision before proceeding to the next phase.

Conclusion

An AI content strategy for ecommerce has moved from experimental to essential. Product pages, category content, and FAQ sections must now be structured for answer engines, not only for Google, because AI-referred traffic represents a discovery channel that rewards clarity, factual density, and machine-readable markup.

The ten-step sequence outlined here functions as a repeatable system: audit current visibility, optimize product and category content, build FAQ pages that answer real queries, implement structured data, and iterate based on performance data. Each step compounds on the previous one, creating a content architecture that answer engines can parse and cite with confidence.

Measurement across multiple engines — ChatGPT, Perplexity, Gemini, and Google — separates visible brands from invisible ones. Single-engine checks provide an incomplete picture, as crawl patterns and citation behavior differ substantially across platforms. Understanding what AI-referred traffic is and how to measure it provides the foundation for tracking progress across this fragmented landscape.

Key takeaways - AI content strategy for ecommerce requires structuring product, category, and FAQ content for answer engines, not just Google. - The 10-step sequence forms a repeatable audit-to-iteration system for sustained AI visibility. - Structured data and factual, comparison-ready content increase citation likelihood. - Cross-engine measurement is essential; single-engine checks miss the full picture. - Iteration based on visibility data, not guesswork, drives compounding results.

Frequently Asked Questions

How do I optimize product pages for AI answers?

Write an entity-rich summary in the first 100–150 words of the product description, add Product and FAQPage schema, and state specifications and differentiators in plain language. AI answer engines extract cited facts from content that clearly defines what a product is, what it does, and who it serves. A concise opening paragraph that names the product category, key materials, dimensions, use cases, and primary differentiators gives crawlers the structured signals they need to attribute a citation. Supporting that summary with Product and FAQPage schema markup further clarifies entity relationships, while the FAQ section captures long-tail questions that AI engines frequently pull into generated answers.

What is AI-referred traffic and how do I measure it?

AI-referred traffic is visits that arrive from cited links inside ChatGPT, Perplexity, and other AI answer engines, and it requires its own detection method because these platforms often lack clean referrer headers. Standard analytics tools typically categorize such visits as direct traffic, obscuring their true origin. To measure AI-referred traffic accurately, ecommerce teams should implement server-side tracking or use platforms that fingerprint AI platform sessions, then segment that data separately from organic search. Monitoring brand mentions in ChatGPT and Perplexity provides the citation-level visibility needed to connect specific AI answers to subsequent site visits.

How long until AI content strategy shows results for an ecommerce store?

Ecommerce stores typically see the first measurable citation shifts within 2–4 weeks, with category and FAQ pages earning citations faster than deep product pages. AI engines prioritize content that answers comparative and informational queries, which is precisely what category overviews and FAQ sections provide. Product pages, by contrast, require more crawl cycles to establish entity trust, especially for newer or less-reviewed items. A 60–90 day evaluation window offers a realistic picture of citation growth, traffic patterns, and which content formats are earning the most AI-referred visits.

Do I need different content for ChatGPT versus Perplexity versus Google?

Each engine draws from different source pools and indexes content differently, so content must be tested across engines rather than optimized for a single platform. ChatGPT relies heavily on its own browsing and partner data sources, while Perplexity maintains a distinct crawler with different indexing priorities. Google's AI Overviews draw from its web index but apply ranking signals that differ from both. The practical implication is that content performing well in one engine may be absent or poorly ranked in another. Ecommerce teams should track citation performance per engine and adjust content emphasis accordingly, as outlined in guidance on ranking in Perplexity answers. Crawl data confirms that AI engines access websites differently than Googlebot, reinforcing the need for engine-specific monitoring.

What schema markup matters most for ecommerce AI visibility?

Product, AggregateRating, FAQPage, and BreadcrumbList markup are the priority schema types for ecommerce citation eligibility. Product schema establishes the core entity — name, brand, SKU, price, and availability — giving AI engines structured facts to cite. AggregateRating adds social proof signals that answer engines frequently reference when comparing options. FAQPage markup feeds directly into conversational answers, while BreadcrumbList clarifies site hierarchy and product categorization. Validation through Google's Rich Results Test ensures the markup parses correctly, though ecommerce teams should note that AI engines maintain their own extraction standards beyond Google's validation criteria.

Sources

• Search Engine Journal — ChatGPT vs Googlebot crawl data analysis
• Search Engine Land — Google acknowledges more visitors from AI systems
• schema.org — Product and FAQPage documentation
• Google Rich Results Test documentation
• Google Search Central — AI Overviews and structured data guidance
• Forrester — AI-driven strategies share of web traffic (cited via Alef's ecommerce AI-referred traffic article)