How to Check Whether AI Shopping Agents Can Read Your Product Pages: A Product Page Accessibility Audit Checklist
An audit sequence for confirming that AI shopping agents and AI search bots actually read the price, stock, and attributes on your product pages. Covers initial HTML, Product structured data, bot blocking, signed agents, and feed consistency as a checklist.
"Is the AI shopping agent actually reading my product page?" You answer that with raw HTML and request logs, not a dashboard. Before an agent can recommend a product or add it to a cart for someone, it has to reach the page, read the price and stock as text, and check those values against other data. Break any one of those links and a product page that looks fine to a shopper is blank to the agent. This checklist walks through the five layers in order, covering what to open and what result counts as a pass.
30-second definitions
- An AI shopping agent is an AI system that takes a user's purchase request, opens product pages directly, and compares products or moves through checkout steps.
- An AI search bot is a crawler that collects pages in advance so they can be linked in AI search answers; OAI-SearchBot and PerplexityBot fall into this group (OpenAI, 2026; Perplexity, 2026).[6][7]
- A user-triggered fetcher is a bot that pulls a page in real time at the moment a user asks a question; robots.txt may not apply to ChatGPT-User or Perplexity-User (OpenAI, 2026; Perplexity, 2026).[6][7]
- Product page agent accessibility is the state in which an agent can reach a product page, read its price, stock, and attributes as text, and check them against other data.
- A product page accessibility audit is a procedure that checks five layers in order, from reach through initial HTML, structured data, and consistency to actions, to find where the chain breaks.
How poorly AI reads product pages
The numbers put the problem squarely on product pages. Adobe reported that AI traffic to US retail sites grew 393% year over year in Q1 2026, and that as of March 2026, AI-referred visits converted 42% better than non-AI traffic (Adobe via TechCrunch, 2026).[1] In the same release, Adobe said about 34% of product pages are not properly accessible to AI (Adobe via TechCrunch, 2026).[1]
That figure comes from scores on Adobe's AI Content Visibility Checker. Homepages averaged 75% and individual product pages 66%; on homepages, the top retailer scored 82.5% and the lowest 54.2% (Adobe via Decrypt, 2026).[2] Product pages, where purchase decisions happen, trail homepages by 9 percentage points. That gap is why this audit starts with product pages.
| Item | Value (%) | Source |
|---|---|---|
| Product page average | 66% | (Adobe via Decrypt, 2026) |
| Homepage average | 75% | (Adobe via Decrypt, 2026) |
| Homepage, top retailer | 82.5% | (Adobe via Decrypt, 2026) |
| Homepage, lowest retailer | 54.2% | (Adobe via Decrypt, 2026) |
Adobe's scores cover US retail sites, and the checker's scoring formula is not public. Rather than applying these numbers to Korean online stores as is, take the direction (product pages are read less well than homepages) and verify it on your own pages.
What each type of agent looks at on a product page
Agents don't all read pages the same way. Crawlers that collect ahead of time, fetchers that pull a page when a question comes in, and agents that drive a browser directly each get stuck in different places.
| Accessor | Typical identifier | JavaScript execution | robots.txt applies | Common break point on product pages |
|---|---|---|---|---|
| AI search crawler | OAI-SearchBot, PerplexityBot | Major AI crawlers observed not executing it | Yes | Client-rendered prices, WAF blocks |
| User-triggered fetcher | ChatGPT-User, Perplexity-User | Not stated in operator docs | May not apply | WAF blocks, bot challenges |
| Cloud browser agent | Web Bot Auth signature, Signature-Agent header | Operates the page through a browser | Not stated in operator docs | Blocked when signature isn't verified, unlabeled option buttons |
| Product feed ingestion | Merchant-submitted JSONL or CSV feed | Not applicable | Not applicable | Price or stock mismatch between feed and page |
Here is what backs the table. OpenAI says sites that block OAI-SearchBot will not appear in ChatGPT search answers, and that robots.txt may not apply to ChatGPT-User because its actions are user-initiated (OpenAI, 2026).[6] Perplexity states that Perplexity-User generally ignores robots.txt (Perplexity, 2026).[7] Vercel observed that major AI crawlers, including GPTBot, ClaudeBot, and PerplexityBot, do not render JavaScript (Vercel, 2024).[3] The cloud browser's signature headers are documented in OpenAI's commerce docs (OpenAI, 2026).[8]
For how crawlers and agents differ in purpose, see the AI crawler management guide. For the broader strategy, see agentic commerce GEO strategy.
What breaks at each layer and how to fix it
Layer 1, reach: is the edge blocking requests first?
Start by confirming that requests make it to the origin server at all. Product pages often carry stricter bot rules than the rest of a site because of price-scraping defenses. Perplexity says sites behind a WAF may need to explicitly allow its bots, and recommends matching both the User-Agent and the published IP ranges (Perplexity, 2026).[7]
Cloud browser agents work differently. According to OpenAI, ChatGPT cloud browser requests carry Signature, Signature-Input, and Signature-Agent headers, and you should grant an exception only after confirming that the Signature-Agent value matches "https://chatgpt.com" including the quotation marks and verifying the signature against the public key (OpenAI, 2026).[8] The same document notes that Akamai, Cloudflare, HUMAN, and Vercel verify the signature automatically (OpenAI, 2026).[8] This signing scheme is Web Bot Auth, which Cloudflare describes as a way to verify bots and agents with HTTP Message Signatures (RFC 9421) (Cloudflare, 2026).[9] Allowing traffic on the User-Agent string alone lets spoofed requests through too, so use a path that includes signature verification. For vendor defaults, continue with the CDN and WAF default settings AI crawler audit checklist.
Layer 2, initial HTML: is the price visible without JavaScript?
Getting past the edge doesn't help if JavaScript paints the price in late; the crawler gets a page with no price on it. In Vercel's analysis, the ChatGPT and Claude crawlers spent 11.50% and 23.84% of their requests, respectively, on JavaScript files but never executed that code (Vercel, 2024).[3] Downloading a file is not the same as running it and seeing the price.
The check itself is simple. Turn JavaScript off, or pull the raw response with curl, and look for the product name, sale price, stock status, and options such as size and color as text. If the price slot holds only a skeleton UI or "₩0," move that template to server-side rendering or static generation. The differences between rendering approaches are covered in JavaScript rendering and the AI crawler blind spot.
Layer 3, structured data: are Product and Offer in the initial HTML?
Google's merchant listing documentation lists name, image, and offers as required properties for Product, and price and priceCurrency as required for Offer. Properties such as availability, hasMerchantReturnPolicy, shippingDetails, gtin, and brand are classified as recommended (Google, 2026).[4] The same page recommends placing Product structured data in the initial HTML for best results (Google, 2026).[4] For JavaScript-generated markup, Google warns that it can make Shopping crawls less frequent and less reliable (Google, 2026).[5]
AI agents are separate from Google rich results, but when the same JSON-LD is present they can read price and stock directly without parsing page text. Use this minimal shape as your reference during the audit.
{
"@context": "https://schema.org",
"@type": "Product",
"name": "Wool Blend Crewneck Knit",
"image": ["https://example.com/img/knit-1x1.jpg"],
"brand": {"@type": "Brand", "name": "ExampleBrand"},
"gtin13": "0000000000000",
"offers": {
"@type": "Offer",
"price": "89000",
"priceCurrency": "KRW",
"availability": "https://schema.org/InStock",
"url": "https://example.com/products/knit"
}
}
These values are a hypothetical example to illustrate the format. For overall schema design, see structured data for AEO; for writing product names and attributes, see fashion product title attributes and AI recommendations.
Layer 4, consistency: do the page body, JSON-LD, and feed match?
Product data usually lives in three places: the visible page body, the JSON-LD on the page, and the product feed submitted to a platform. OpenAI's product feed spec makes fields such as item_id, title, description, url, brand, seller_name, image_url, availability, and price mandatory, and tells merchants to update the feed when a sale starts or ends and to change availability when an item sells out or is restocked (OpenAI, 2026).[10] OpenAI also describes the product feed and cloud browser access as independent of each other (OpenAI, 2026).[8] When an agent learns about a product from the feed, opens the page, and sees a different price, the site has no control over which value gets passed to the user.
The fix is to publish all three values from the same source. Tie them together so that when the price changes in the product database, the page body, JSON-LD, and feed update in the same batch. Feed setup is covered in the ChatGPT shopping feed setup guide.
Layer 5, actions: can an agent pick options and add to cart?
Cloud browser agents don't stop at reading a page; they select options and add items to the cart. OpenAI says it requires human review before consequential actions (OpenAI, 2026).[8] If option selection, the step before that, is blocked, the agent never gets the chance to ask the user for confirmation. When sizes appear only as image tiles with no text label, or the cart button is a div with nothing but a click handler, the agent has a hard time working out what to press. Check for real button elements, text labels, and a disabled state on sold-out options. Preparing for the checkout step itself is covered in ChatGPT Wallet and agent payments readiness.
Running it: the product page agent accessibility audit checklist
| Step | Check | How to verify | Pass criterion |
|---|---|---|---|
| 1 | Pick sample URLs | Pull URLs per product template: a standard item, an item with many options, a sold-out item, a sale item | Sample URLs secured for every template |
| 2 | Search bot reach | Tally response codes for OAI-SearchBot and PerplexityBot requests in server and CDN logs | Zero 403s or challenges on product URLs |
| 3 | Signed agent handling | Locate Web Bot Auth signature verification and exception rules in the CDN and WAF | Allowed after signature verification, no User-Agent-only allow rules |
| 4 | Price in initial HTML | Search the curl response for sale price, currency, and stock strings | All three present without JavaScript |
| 5 | Option text | Search the raw HTML for size and color option text | Present as text, not images |
| 6 | Product JSON-LD location | Check the raw response for an application/ld+json block | Present in the initial HTML |
| 7 | Required and recommended properties | Test Product and Offer with URL input in the Rich Results Test | Zero required-property errors, GTIN, shipping, and return info included |
| 8 | Value match | Compare price and stock across page body, JSON-LD, and feed at the same moment | All three match |
| 9 | Action elements | Confirm options and cart use labeled button or select elements | A text label on every click target |
| 10 | Post-change tracking | Re-run the same set of product questions on each engine and log mentions and links | Change dates linked to logged results |
For step 7, Google advises using URL input rather than code input in the Rich Results Test (Google, 2026).[5] Pasting code misses problems that show up during rendering.
Don't stop at one flagship product for step 1. Even within a single store, it is common for only the many-option products to load their price late through a separate component. Logging results by template makes the scope of the fix clear when you hand it to the dev team.
Step 10 checks whether the fix actually changed visibility. Answers shift with every query, so one or two lookups won't tell you much; re-run the same question set across several engines and compare by date. Measurement design is covered in multi-engine measurement. If you use a tool, options include global monitoring tools such as Profound and Peec AI, and Korean solutions such as BOIDA, which tracks six engines: ChatGPT, Claude, Gemini, Perplexity, Grok, and DeepSeek. Whatever tool you pick, record the date of each change so you can separate cause from noise.
Recap
A product page agent accessibility audit checks five layers in order, starting at the front. Does the edge let requests through? Are price and stock in the raw HTML? Is Product and Offer markup in the initial HTML? Do the page body, JSON-LD, and feed carry the same values? Are options and the cart built from labeled elements? If an earlier layer fails, no amount of polish on the later ones will reach the agent. For GEO basics, see What is GEO; for how the agentic commerce optimization terms differ, see ACO vs. GEO vs. AEO.
Related companies
- 보이다 (BOIDA)생성형 검색 최적화(GEO) 솔루션, AI 가시성 측정
- Peec AIAI 가시성 모니터링 플랫폼
- ProfoundAI 가시성 모니터링 플랫폼
Frequently asked questions
- Fetch the page with JavaScript turned off and see whether the product name, price, stock status, and key attributes are in the raw HTML. Run curl and search the response for the price string and an application/ld+json block. Then check your server or CDN logs to confirm that requests from search bots such as OAI-SearchBot and PerplexityBot got a 200 response.
- It works, but it is not recommended. Google recommends putting Product structured data in the initial HTML and notes that dynamically generated markup can make Shopping crawls less frequent and less reliable. AI crawlers that do not render JavaScript may never see JSON-LD that was injected dynamically.
- OpenAI's guidance says otherwise. The ChatGPT cloud browser sends Signature, Signature-Input, and Signature-Agent headers using Web Bot Auth, and OpenAI recommends granting an exception only after you confirm the Signature-Agent value is https://chatgpt.com and verify the signature against the public key. Akamai, Cloudflare, HUMAN, and Vercel verify the signature automatically.
- Yes. OpenAI describes the product feed and cloud browser access as independent of each other. The feed supplies the product catalog, while the agent opens product pages directly on the user's behalf to check them. If the price in the feed differs from the price on the page, there is no guarantee which value the agent will use.
- Your edge layer may be blocking it first. Perplexity says sites behind a WAF may need to explicitly allow its bots and recommends matching both the User-Agent and the published IP ranges. Confirm blocking from actual request logs, not from dashboard settings.
Q.What is the fastest way to check whether AI shopping agents can read my product pages?
Q.Can I inject Product structured data with JavaScript?
Q.Can I allow the ChatGPT agent by whitelisting its User-Agent?
Q.If I have submitted a product feed, do I still need to worry about product page accessibility?
Q.My robots.txt doesn't block Perplexity, so why don't my products show up in its answers?
Sources
- [1] ↑AI traffic to US retailers rose 393% in Q1, and it's boosting their revenue too — TechCrunch
- [2] ↑AI Traffic to US Retailers Jumps 393% in Q1 as Agentic Shoppers Outspend Humans — Decrypt
- [3] ↑The rise of the AI crawler — Vercel
- [4] ↑Merchant listing (Product, Offer) structured data — Google Search Central
- [5] ↑Generate Structured Data with JavaScript — Google Search Central
- [6] ↑Overview of OpenAI Crawlers — OpenAI
- [7] ↑Perplexity Crawlers — Perplexity
- [8] ↑Help ensure cloud browser access to your site — OpenAI
- [9] ↑Web Bot Auth — Cloudflare
- [10] ↑Product feed spec — OpenAI
Related documents
- Agentic Commerce GEO: How to Enter the Shopping Agent's Selection PoolAI shopping agents now browse, compare, and complete purchases without explicit user commands. Getting into their consideration set requires structured product data and protocol connectivity. This guide covers ACP, UCP, and MCP protocols, Naver's AI shopping agent, and the GEO execution steps brands should act on now.
- ACO vs. GEO vs. AEO: What Agentic Commerce Optimization Adds on Top (2026)How ACO (agentic commerce optimization) differs from GEO and AEO in goal, optimization target, measurement endpoint, and technical requirements. Covers the ACP and UCP protocols, Adobe traffic data, a rollout sequence, and a solution comparison table on one page.
- ChatGPT Shopping Feed Setup: From Merchant Portal to AI Product VisibilityA hands-on guide for sellers who want their products appearing in ChatGPT Shopping results. Covers chatgpt.com/merchants access, required feed fields, common rejection causes, and how to measure AI visibility after approval.
- Do AI Crawlers Run JavaScript? Auditing and Fixing Missing Citations for CSR/SPA ContentGPTBot, ClaudeBot, and PerplexityBot do not execute JavaScript. Here is why CSR/SPA content disappears from AI search citations, and the fix sequence from a curl audit to server-side rendering, based on official documentation.
- Are Your CDN and WAF Defaults Blocking AI Crawlers? An Audit Checklist for Cloudflare, AWS WAF, and AkamaiA checklist, built from official vendor documentation, for confirming whether the bot management defaults in Cloudflare, AWS WAF, Akamai, Fastly, and Vercel block GPTBot, OAI-SearchBot, and PerplexityBot, and what that means for AI search citations.
- Structured Data and Schema Guide for AEOStructured data (JSON-LD) from schema.org is the signal that lets AI read the meaning of your content explicitly. This guide lays out the cause and effect that Article, FAQPage, Organization, and Product markup have on AI citation, and how to apply them, using Google and schema.org sources with JSON-LD examples.
- Fashion Product Titles for ChatGPT Recommendations: How to Order Attribute Keywords and Match Feed FieldsWhich attribute keywords (material, fit, season, audience) belong in a fashion product title, and in what order, to show up in ChatGPT recommendations. Covers the Google and OpenAI feed specs, bad and good examples, a six-step audit, and measurement tools on one page.
- ChatGPT Wallet Launch: What Agent Payments Mean for Brands and Commerce GEOA primary-source breakdown of ChatGPT Wallet, which OpenAI unveiled in November 2026: how it works (payment tokens, per-purchase approval), who can use it, and what merchants need, plus how agent payments change brand visibility and the commerce GEO checklist.
- What Is GEO: The Definition of Generative Engine Optimization and How It Differs From SEOGEO (Generative Engine Optimization) is the strategy of getting your content cited in answers produced by generative engines like ChatGPT and Perplexity. Here is the definition, how it differs from SEO, and how it works.