Visibility in the Age of AI
Everything you know about ecommerce is changing fundamentally
You are no longer selling only to people. Agents can research, compare and act on the same product information, shaping which brands enter consideration before a shopper reaches a website.
Agents
58.7%Humans
41.3%Visibility started with keywords
To find a product, a customer compressed a need into a short query, scanned ranked results and compared the options themselves. Brands competed to appear high enough to enter consideration.
CloudStep cushioned walking shoes
£89.00
Northline
TrailLite waterproof walking shoes
£112.00
Northbound
CityFlex supportive walking shoes
£96.00
Common Ground
MetroWalk memory-foam shoes
£84.00
Fieldwork
AI agents research, compare and build the shortlist
AI turns the request into product tests.
Checks product pages, reviews, price and availability against those tests.
Bose
QuietComfort Headphones
24h with ANC
£229.95
JBL
Tune 770NC Headphones
44h with ANC
£49.99
Sony
WH-CH720N Headphones
35h with ANC
£75
Above £150 budget
Shortlisted
Below 40h requirement
JBL Tune 770NC enters the shortlist.
Customer chooses JBL Tune 770NC and continues to checkout.
JBL
Tune 770NC Headphones
44h with ANC
£49.99
Under the hood, several factors shape an agent’s answer
An AI agent’s answer depends on the selected model, the information supplied in the conversation and capabilities such as live web retrieval. The same prompt can produce different recommendations.
The following outputs come from real agent runs on platform.openai.com using GPT 5.6 Luna, GPT 5.6 Terra and GPT 5.6 Sol.The shopper asksa question
They describe what they need, how they will use it and their budget.
Model answer
For 20 km/week on roads, I’d choose a comfortable, cushioned daily trainer in an actual wide width, not simply a roomy standard shoe. Good UK options under £150 include:
Shoe Why it’s a good choice Typical UK price Brooks Ghost 16 Wide Safest all-round pick: comfortable, reliable cushioning, smooth on roads £130–£150 Saucony Ride 17 Wide Soft, well-cushioned and versatile for easy and moderate runs £120–£145 New Balance 880v14 Wide / 2E Stable, durable and often particularly good for broad feet £120–£145 ASICS Gel-Cumulus 26/27 Wide Plush, comfortable cushioning without feeling overly soft £130–£150 Mizuno Wave Rider 28 Wide Firmer, responsive feel and good durability £130–£150 My recommendation
- Buy the Brooks Ghost 16 Wide if you want the safest, most universally comfortable option.
- Choose the New Balance 880 Wide/2E if your feet are broad through the forefoot.
- Choose the Saucony Ride 17 Wide if you prefer a softer, more cushioned feel.
Fit tips
- Look for “Wide,” “2E,” or “4E”. A standard-width shoe with a stretchy upper may still be too narrow.
- Try shoes on later in the day, wearing your running socks.
- Leave roughly a thumb’s width in front of your longest toe.
- Your heel should feel secure while your toes can spread naturally.
- Don’t automatically size up to solve width; try a wide version first.
Prices and available widths vary by men’s/women’s model and retailer, so check the exact width before ordering. If you have never used zero-drop shoes, I’d avoid Altra as a first choice.
Model answer
For your 20 km/week of road running and £150 budget, I’d buy the Saucony Ride 18 Wide.
Best choices
-
Saucony Ride 18 Wide: best overall
- £78 currently on Saucony UK, reduced from £130.
- Neutral, everyday road shoe with a good balance of cushioning and responsiveness.
- Best fit if you want one shoe for most runs without a “max-cushion” feel. (saucony.com)
-
Brooks Ghost 17 Wide: safest comfort pick
- £94.50 currently, reduced from £135.
- Soft, dependable cushioning and balanced support; particularly suitable for easy runs and general training.
- Check the exact colour/size carefully, as availability varies and some wide versions are listed as preorder or sold out. (brooksrunning.com)
-
HOKA Clifton 10 Wide: if you prefer very cushioned shoes
- Around £98 at one UK running retailer, with other listings under £150.
- Lightweight, cushioned and road-oriented, but the rocker-like shape and HOKA feel are more distinctive, so try them on if possible. (achillesheel.co.uk)
-
New Balance Fresh Foam X 880v15: if you need a broader, more conventional fit
- The 2E Wide version is currently listed around £104 by UK price comparison data.
- A sensible daily trainer, especially if New Balance’s width options fit your foot shape better than Saucony or Brooks. (pricespy.co.uk)
-
ASICS GT-2000 14 Wide: only if you want extra guidance
- £98 currently, reduced from £140.
- A road shoe with a stability/guidance design, so choose it if you know you prefer or need that feeling, not simply because your feet are wide. (asics.com)
My recommendation
- Buy the Saucony Ride 18 Wide if you want the best value and a versatile neutral shoe.
- Choose the Brooks Ghost 17 Wide if softness and comfort matter most.
- Choose the HOKA Clifton 10 Wide if you like substantial cushioning and a rocker-style ride.
When trying them on, run or walk in them with your usual socks. You should have roughly a thumb’s width in front of your longest toe, no pressure over the little-toe side or big-toe joint, and a secure heel without tightening the laces excessively. Don’t assume a regular shoe in a larger size is equivalent to a proper Wide/2E version.
Change the model or live web access,and the answer can change
Compare the final answer after each change. First turn live web retrieval on and off for the same model; then keep web access fixed and choose another model.
- No matching modelsgpt-5.6-solFrontier model for complex professional workgpt-5.6-terraGPT-5.6 model that balances intelligence and costgpt-5.6-lunaGPT-5.6 model optimized for cost-sensitive workloadsgpt-5.5A new class of intelligence for professional workgpt-5.5-pro-2026-04-23gpt-5.5-progpt-5.5-2026-04-23gpt-5.4-pro-2026-03-05gpt-5.4-progpt-5.4-miniFaster, cost-efficient version of GPT-5.4gpt-5.4-nanoFastest, most cost-efficient version of GPT-5.4gpt-5.4-2026-03-05gpt-5.4Intelligence at scale for agents & professional workgpt-5.3-codexgpt-5.3-chat-latestgpt-5.2-pro-2025-12-11gpt-5.2-progpt-5.2-codexgpt-5.2-chat-latestgpt-5.1-codex-minigpt-5.1-codex-maxgpt-5.1-codexgpt-5.1-chat-latestgpt-5.1-2025-11-13gpt-5.1gpt-5-pro-2025-10-06gpt-5-progpt-5-nano-2025-08-07gpt-5-nanoFastest, most cost-efficient version of GPT-5gpt-5-mini-2025-08-07gpt-5-miniSmaller, faster version of GPT-5 for well-defined tasksgpt-5-codexgpt-5-chat-latestgpt-5-2025-08-07gpt-5gpt-4.1Smartest non-reasoning modelgpt-4.1-miniSmaller, faster version of GPT-4.1gpt-4.1-nanoFastest, most cost-efficient version of GPT-4.1gpt-4.1-nano-2025-04-14gpt-4.1-mini-2025-04-14gpt-4.1-2025-04-14o4-mini-2025-04-16o3-2025-04-16o3Smartest reasoning modelo3-minio3-mini-2025-01-31o1-pro-2025-03-19o1-proA version of o1 that uses maximum compute for more reliable responseso1-2024-12-17o1chat-latestgpt-4ogpt-4o-minigpt-4o-search-previewFast, up-to-date answers with clear and relevant citations from the webgpt-audio-mini-2025-12-15gpt-audio-mini-2025-10-06gpt-audio-mini
Shoppers encounter afragmented model landscape
One prompt can change under different configurations. Shoppers can also encounter many models, versions and capability levels. This catalogue shows that breadth; selecting an item does not change a response.
Some signals that may help your brand become part of the answer
Agent-led discovery is already a paid channel
Businesses can already buy placements inside some agent journeys. In one documented ChatGPT car rental flow, Avis ads appeared while the shopper researched Hertz.
This single case does not prove that the ad changed the recommendation or that every platform works the same way. It shows that paid placement now sits alongside earned recommendations.
As visibility becomes valuable, bad actors follow
Black-hat tactics may lift visibility briefly by manipulating what AI systems see or how they behave, but they put customer trust and the brand at risk.
Image 1 of 9
Looking ahead, shopping becomes agentic
One possible direction, not a forecast for every purchase, is that trusted agents coordinate more of the journey after a customer expresses the outcome.
Scenario boundaryUCP can carry checkout handoffs; AP2 can add verifiable authorization. The multi-person agent coordination shown is illustrative, not defined by either protocol. The business remains the merchant of record.
Shopping meant doing every step
On a computer, one person searched, compared, collected preferences and checked out.
Manual pizza ordering. One person finds local delivery, checks restaurant menus, asks each friend what they want, compares prices and delivery times, chooses a restaurant, configures four pizzas, reviews the basket and fees, enters delivery and payment details, places the order and receives confirmation..
The screen changed, not the work
On a phone, the person still searched, chose and entered every detail.
Manual mobile shopping. One person scrolls restaurant menus, messages friends, configures four pizzas, enters checkout details and pays..
Order pizza for me, Maya, Theo and Sam.
Shopping starts with the outcome
One person asks for pizza for the group. Trusted agents coordinate the preferences and order.
The customer says, “Order pizza for me, Maya, Theo and Sam.” The shopping agent confirms it is working, retrieves its customer’s Margherita preference, waits while the group’s personal agents resolve Ortolana, Capricciosa and Diavola, finds a suitable restaurant, checks the menu and delivery fit, then confirms: “I placed the order with a nearby pizza restaurant. Four pizzas: Margherita, Ortolana, Capricciosa and Diavola. Total: £64.80 including delivery.” The customer replies, “Thank you,” and the agent closes with, “No problem.”. The shopping agent asks Maya’s personal agent for her preference. It checks her saved favourites and exclusions, selects an Ortolana and responds.. The shopping agent asks Theo’s personal agent for his preference. It reviews his recent orders and dietary profile, selects a Capricciosa and responds.. The shopping agent asks Sam’s personal agent for his preference. It checks his usual order and spice preference, selects a Diavola and responds..
The merchant stays in control
The restaurant checks the order, authorization, menu availability and kitchen capacity before accepting.
The restaurant receives a new order request, reviews the four pizzas and delivery details, confirms payment authorization status, checks menu availability and kitchen capacity, accepts the order, then sends the ticket to the kitchen.. Checkout request received. Buyer authorization verified. Merchant accepts under its rules.
Coordination continues after checkout
The accepted order moves through preparation and delivery as one coordinated handoff.
The agent follows the order through fulfilment. One coordinated handoff carries the whole group order.
The outcome replaces the process
Everyone gets a personalised choice without one person managing each step.
Four individual preferences resolve into one order. No one has to manage the transaction step by step.
The shift is already underway. Your business needs to be ready for what comes next
This will impact your business
AI will change how customers find, compare and buy products. The impact will differ by business, but no business can ignore it. Founders need to stay involved, prepare their teams and decide where to act first.
There is no fixed playbook
Everything is moving fast. A tactic can become obsolete within days. Keep watching, testing and adapting instead of relying on one playbook.
This is an opportunity to grow
This change is not only a risk. It can help you reach new customers, reduce friction and create better buying journeys. Businesses that understand the shift and adapt can use it to grow.
Get the full artifact
Join the CognitionHub Network for the complete AI visibility artifact and the conversations behind it.
Sources and limitations
- Google, AI Mode updates from I/O 2025
Supports query fan-out and related research. Availability varies by market and account.
- Amazon, Interests
Explains translation of everyday language into queries and attributes.
- Amazon, Rufus usage and shopping behaviour
More than 250 million customers had used Rufus in 2026 by 13 May; the purchase relationship is associative, not causal, and Rufus was renamed Alexa for Shopping.
- Similarweb, Zero-Click Marketing: The SEO Strategy For A World Where Most Searches Skip Your Site
Published 10 June 2026 from Similarweb’s own clickstream panel; the article reports a 68.01% zero-click rate and presents 68% as the headline figure. It measures general Google search behaviour, not commerce journeys, purchases or the isolated effect of AI Overviews. The article does not publish a complete sample size or confidence interval.
- Google Search Central, Optimizing your website for generative AI features on Google Search
States that Google’s generative AI Search features use core Search ranking and quality systems, sets the indexed-and-snippet-eligible threshold, describes Merchant Center visibility, and says the features can show material from blogs, videos and forums. The guidance applies to Google Search only and does not guarantee inclusion.
- Google Search Central, generative AI performance reports
Announced the 3 June 2026 launch and limited rollout. Current Search Console Help pages for [Search](https://support.google.com/webmasters/answer/16984139?hl=en) and [Discover](https://support.google.com/webmasters/answer/16983858?hl=en) describe impression trends, page, country and date dimensions, plus device for Search. They do not list clicks, queries, conversions or causal-attribution metrics, and the surface definitions differ.
- OpenAI, product checkout specification
Merchants retain order, payment and compliance systems.
- Juozas Kaziukėnas, Hertz and Avis observation
Evidence of one public interface state, not universal behaviour or causal influence.
- Google Search spam policies
Defines Google’s boundary around attempts to manipulate generative responses in Search.
- CrowdReply
Describes the service’s positioning, not independent evidence of effectiveness.
- Dawood Khan, CrowdReply case-study post
Includes the aged-account description; performance claims are unverified.
- CrowdReply refund policy
Refers to submitted comments, threads and moderator removals.
- Reddit Rules
Requires authentic participation and prohibits spam and content manipulation.
- Federal Trade Commission, endorsement guidance
Local legal requirements must be checked for other markets.
- OpenAI, Understanding prompt injections
Defines prompt injection and gives an indirect recommendation example.
- OWASP GenAI Security Project, LLM01:2025 Prompt Injection
Defines direct and indirect prompt injection.
- NIST, Strengthening AI Agent Hijacking Evaluations
Defines agent hijacking in the context of indirect prompt injection and unintended action.
- Google, AI Mode shopping and virtual try-on
Supports Shopping Graph scale, refresh rate and current shopping capabilities.
- Shop with Google, The Devil Wears Prada 2 cinema campaign
Supports how Google’s virtual try-on was presented through a fashion-led cinema collaboration; it does not establish audience adoption or effect.
- Google Ads, Discovering & Shopping Made Easy with Google AI
Demonstrates Google’s promoted shopping journey across AI Mode, virtual try-on and YouTube; it is advertising, not observed customer behaviour.
- StatCounter Global Stats, worldwide search-engine market share
June 2026 share of pageviews referred by search engines to StatCounter-instrumented sites; not searches, users, zero-click answers, product discovery or assistant activity.
- Google I/O 2026 opening keynote
Reports Gemini, AI Mode and AI Overview monthly audiences; the products and audiences overlap and are not shopping-specific.
- OpenAI, May 2026 publisher-partnership announcement
Reports more than 900 million weekly active ChatGPT users worldwide; not shopping-specific usage.
- Microsoft, fiscal 2026 first-quarter earnings call
Overall consumer and professional Copilot usage.
- Google Search Central, Creating helpful, reliable, people-first content
Says Google’s automated ranking systems are designed to prioritise helpful, reliable information created for people. It does not define a product-verification stage or a universal credibility score.
- Google Search Central, AI features and your website
States that a supporting page must be indexed and eligible for a Search snippet, that standard Search requirements apply, and that meeting them does not guarantee crawling, indexing or serving. It applies to AI Overviews and AI Mode in Google Search.
- Google, new controls and insights for website owners
Confirms that Google began rolling out the new Search Console control and insights to a subset of UK website owners on 3 June 2026; availability is mutable.
- Google, Universal Cart and agentic shopping
Reports more than 60 billion Shopping Graph listings in May 2026; it does not establish recommendation quality.
- Semrush, most visited websites worldwide
June 2026 estimates of website visits derived from clickstream data and modelling, not unique users, searches or product-discovery activity; excludes app-only, API and most embedded-assistant use.
- Google, how people use AI Mode
Based on Google Trends and query data; growth rates describe query categories, not their total share or commercial outcomes.
- OpenAI, API models
and [OpenAI, API usage tiers](https://help.openai.com/en/articles/6643435). OpenAI lists multiple models and says API usage and spend can advance accounts to higher usage tiers; availability and limits vary by account.
- Anthropic Economic Index, March 2026
Sample of one million Claude conversations from February 2026; task and personal-use classifications are not comparable with other platforms’ taxonomies.
- Microsoft Research, how people use Microsoft 365 Copilot Chat
Analysis of approximately 5.5 million enterprise sessions; describes workplace use, not consumer or shopping intent.
- Sensor Tower, State of AI 2026
Estimates that the ChatGPT mobile app reached one billion monthly active users in May 2026; app activity excludes web-only, API and embedded-product use.
- Reuters, ChatGPT app reaches one billion monthly active users
Corroborates Sensor Tower’s May 2026 ChatGPT estimate and 56 million Claude monthly active app users; the figures are modelled estimates, not company disclosures.
- El País / EFE, Q2 2026 AI-assistant monthly active app users
Reports Sensor Tower’s like-for-like app estimates for ten assistants; users may overlap across apps, and app activity excludes web-only, API and embedded-product use.
- Microsoft, fiscal 2026 third-quarter earnings call
Reports that Bing reached one billion monthly active users in April 2026; Microsoft does not publish a directly comparable intent distribution in the same disclosure.
- Yang et al., The Adoption and Usage of AI Agents
Analyses hundreds of millions of anonymised Perplexity Comet interactions; 57% of agentic queries were productivity/workflow or learning/research, but Comet use is not all Perplexity use.
- Bing Webmaster Tools, AI Performance
Lists grounding-query intent classes such as informational, commercial, navigational, research and local, but does not publish the share of Bing usage in each class.
- IBM Institute for Business Value and NRF, Consumers are using AI before shopping begins
Survey of more than 18,000 consumers across 23 countries in Q3 2025; the reported behaviours are self-described rather than observed transactions.
- Pew Research Center, ChatGPT use by age in 2026
Survey of 5,119 US adults, fielded 17–23 February 2026. Measures whether a respondent had ever used ChatGPT, not frequency, shopping use or purchase intent.
- Sony, WH-CH720N
Supports the stated noise-cancelling capability, call design and up-to-35-hour battery claim; price and availability can change.
- Bose, QuietComfort Headphones
Supports the stated noise-cancelling capability, call capability and up-to-24-hour battery claim; price and availability can change.
- JBL, Tune 770NC
Supports the stated adaptive noise cancelling, calls and up-to-44-hour battery claim with ANC on; price and availability can change.
- Bose QuietComfort
(£229.95) and [JBL Tune 770NC](https://uk.jbl.com/TUNE770NC.html) (£49.99), plus a current market observation for [Sony WH-CH720N](https://www.whathifi.com/headphones/wireless-headphones/sony-wh-ch720n) (around £75). Prices, promotions, stock and retailers can change.
- Google Research, Conversational Recommendation as Retrieval: A Simple, Strong Baseline
Represents conversations as queries and items as documents in a retrieval study. It does not disclose the complete decision process of a live shopping assistant.
- Amazon Science, Shopping Queries Data Set: A large-scale ESCI benchmark for improving product search
Labels query-product relationships as Exact, Substitute, Complement or Irrelevant for product-search research. It does not disclose Amazon’s complete live ranking system.
- Google Search Central, Introduction to Product structured data
Says structured data can make product information eligible for richer Search appearances and that using it with a Merchant Center feed maximises eligibility and helps Google understand and verify data. Enhancements remain discretionary.
- Google Merchant Center, Availability attribute
Requires availability to remain consistent across product data, landing pages, structured data when present and checkout; a mismatch can trigger product disapproval. The page also instructs merchants to provide up-to-date availability and price data.
- Google Merchant Center, store pages
Explains how businesses can express brand identity and how Google combines the merchant voice with customer reviews and shopping-experience information.
- Google Search Central, discussion forum structured data
Describes forums as places where people collectively share first-hand perspectives; eligible markup does not guarantee appearance.
- OpenAI and Reddit partnership
Confirms OpenAI access to Reddit’s real-time, structured content to help its products understand and surface community discussion, especially on recent topics.
- Google, expanded Reddit partnership
Confirms Google’s structured access to Reddit content for display, training and other uses; it does not imply that every post enters every model or answer.
- OpenAI, Buy it in ChatGPT
Says product results are organic and relevance-led; when multiple merchants sell the same product, merchant selection may consider availability, price, quality, primary-seller status and checkout support.
- Cloudflare Radar, worldwide bot-versus-human requests to HTML content
Shows 58.7% likely automated and 41.3% likely human HTTP requests to HTML content for the last seven days, captured 28 July 2026. Cloudflare classifies bot scores 1–29 as likely automated and 30–99 as likely human. The automated share includes crawlers, monitors and other bots; it is not a measure of AI-shopping-agent traffic, customers or sales.
- Google Developers Blog, Universal Commerce Protocol
Explains how UCP supports commerce flows and checkout handoffs. It does not define the illustrative multi-person coordination shown here or establish current adoption.
- Agent Payments Protocol, AP2 specification
Defines checkout and payment mandates for authorised agent transactions and compatibility with UCP. It does not define the illustrative multi-person coordination shown here or establish current adoption.
- Krasakis et al., A New Taxonomy of Web Search: A User-Centered Framework for Search Intent in the AI Era
Published in CHI 2026. Eighty-six US, English-fluent, experienced users reviewed 247 recent search sessions and 253 recent chatbot sessions in February 2025. Of the 86 participants, 85 used Google and 75 used ChatGPT. The small Prolific sample is not representative of all users, and the figures compare task categories within the reviewed sessions rather than platform-wide intent prevalence.
- CognitionHub, Shopping with Google AI Mode: From a Prompt to a Complete Look
Shows the illustrated Google AI Mode shopping journey embedded in this artifact. The interface and availability can change; the video demonstrates the experience rather than measured adoption or commercial outcomes.
- Coveo, Commerce Relevance Report 2026
Arlington Research surveyed 4,000 adults in the US and UK from 14–19 December 2025, with quotas for age, gender and region. Respondents could select multiple places where they typically start a product search, so the results describe overlapping starting surfaces rather than exclusive shares.
- OpenAI Academy, AI fundamentals
Published 10 April 2026. Explains models, applications, pre-training, post-training and model trade-offs; it is not a universal description of every provider.
- OpenAI, How ChatGPT and our language models are developed
Explains next-token prediction, post-training, variable outputs and weights that reflect learned patterns rather than a searchable copy of training examples. It describes OpenAI models and does not rule out memorisation failures.
- OpenAI, Improving instruction hierarchy in frontier LLMs
Published 10 March 2026. Describes system, developer, user and tool-message authority for OpenAI models; other providers may implement different instruction systems.
- OpenAI, Memory FAQ
Explains that saved memories can be added to the context used for later ChatGPT responses. It describes ChatGPT, not every application.
- OpenAI Academy, Workspace agents
Published 22 April 2026. Describes agents as systems that use instructions and approved tools to move through a process, with bounded decisions, guardrails and approvals.
- Google Developers Blog, Why we built ADK 2.0
Published 1 July 2026. Explains explicit orchestration, control flow and failure handling in production agent systems.
- Google Search Central, Optimizing your website for generative AI features
Updated 10 July 2026. Explains retrieval-augmented generation, crawlability, merchant feeds, structured-data limits and the absence of any guarantee of crawling, indexing or serving.
- OpenAI, Powering Product Discovery in ChatGPT
Published 24 March 2026. Explains supported merchant feeds and catalogue connections that can improve the accuracy and completeness of product representation in ChatGPT; it does not promise recommendation.
- OpenAI, What is ChatGPT: FAQ
and [OpenAI, Powering Product Discovery in ChatGPT](https://openai.com/index/powering-product-discovery-in-chatgpt/), published 24 March 2026. Describe ChatGPT’s general assistant capabilities and its conversational product-discovery experience. Features and limits vary by plan and region; provider descriptions do not measure population-wide shopping behaviour.
- Anthropic, What are some things I can use Claude for?
updated 16 March 2026; [Upload files to Claude](https://support.claude.com/en/articles/8241126-upload-files-to-claude), updated 22 April 2026; and [Use research on Claude](https://support.claude.com/en/articles/11088861-use-research-on-claude), updated 2 June 2026. Describe Claude’s general, file and multi-search research capabilities. Availability depends on account and plan; the sources do not establish shopping adoption.
- Google, Use Gemini Apps
[Google, Get help with your shopping in Gemini Apps](https://support.google.com/gemini/answer/16730149?hl=en-CA), and [Google, Shopping made easier with the Gemini app](https://blog.google/intl/en-ca/products/explore-get-answers/shopping-made-easier-with-the-gemini-app/), published 13 March 2026. Describe general Gemini capabilities and shopping features that use Google’s Shopping Graph. Shopping availability is limited by country, account, language and surface.
- OpenAI developer documentation, Using GPT-5.6
and [GPT-5.6 Luna](https://developers.openai.com/api/docs/models/gpt-5.6-luna). Identify GPT-5.6 Sol, Terra and Luna as models with different capability, cost and workload positions. Model availability and product exposure can change.