🔎 Focus: eCommerce
🔴 Impact: High
🟠 Difficulty: Mid
Sponsored by Ahrefs
Does AI Recommend Your Brand?
Track and grow your brand’s visibility across AI answers, YouTube, and Reddit.
Turn SEO into AEO and reach new audiences.
Dear Tech SEO 👋
I am testing the impact of structured data Chat GPT responses. I am not talking about Schema, which is only one form of structured data…
I am talking about the-often-forgotten Google Merchant Feeds…
Merchant Feeds As GEO Asset
AI hallucination is fun until is on eCommerce.
Imagine AI giving you wrong prices, wrong titles, wrong features of products you want to buy.
Guess what is the most simple method to avoid hallucination: structured data.
Do not think only about Schema. Structured data is a plain CSV or XML or anything that resembles a database (and HTML table?).
Now LLMs are opening up to Merchant Feed. In fact, you can apply to send your merchant feed to Chat GPT: https://chatgpt.com/merchants/
Form to send your merchant feed
The form to send your merchant feed says things like:
… so your products can appear in results …
… Shopify or Etsy, your catalog is already integrated …
Will this influence organic mentions and citations?
I went down a rabbit whole to figure the best way to use Merchant Feeds to improve our eCommerce GEO results: where clicks matter
Ready? Let’s goooooooo
My First Clue: The ?srsltid= Parameter
I was minding my own prompts when I saw a linked source by Chat GPT that contained an auto-tagging parameter: the infamous ?srsltid=
Here an article from 2022 talking about the subject: https://searchengineland.com/auto-tagging-added-to-google-merchant-center-free-listings-379590
The auto-tagging parameter is to “better tracking your shopping campaigns through Google Search by adding a result ID to your URLs“.
In short, the auto-tagging will be triggered on Shopping Ads or “Free Listings“.
“Why is an Ads Tracking Parameter on my ChatGPT response?” I asked myself.
In the screenshot below you can see the link with the ?srsltid= parameter. The car image below is just part of the Chat GPT response at the time:
Spotted the ?srsltid= parameter on Chat GPT responses and the reason of my 73-hour investigation
The ?srsltid= parameter is exclusive of Google Ads. Specifically Google Shopping Ads.
Therefore a Google Ads link in Chat GPT responses indicates that Open AI might be scraping SERP Google results as a regular user.
Open AI scraping live SERPs might take too long or might require Open AI using SerpApi which would make responses too slow. Live SERP scraping indeed might take too much, maybe that is why I could not find many more examples on other prompt results giving links with the ?srsltid parameter.
Instead of ?srsltid I found even more interesting results…
My 2nd Clue: the UTM Parameters
I tested more transactional queries, for example:
best laptop for work apple
Sunday usual experiment day
Since the launch of Chat GPT’s dynamic UI, we get more responses that are far more than text. I found out that even static responses will trigger live searches.
For example, after I got the static recommendation for my query, the actual shops listings appeared after I clicked each of the recommended products:
Note there is no actual information on the right sidebar: results are being retrieved real time:
Shop listing: quite similar to what Google Shopping offers
Product information: reviews of the products from multiple sites
The product title (link) triggers an on-demand live verification and hydration flow via OpenAI’s background task engine:
Quickly checked the Network tab and spotted the process that happens after I click a product title:
1. Task Orchestration:
After you click, there is a planning of tasks controlled by this Orchestrator. Let’s say it is the project manager making sure each task gets completed before the user get the lists of products/shops.
2. Live Commerce Hydration:
Retrieves multiple stores across the web in real time to stream in live pricing and stock levels. It assigns the label "Best price" which is also the priority for the listing presented to the user. This script is the responsible to get the actual links to the product pages. It also checks if an instant in-chat checkout is available (more of this later). Then it gives the final listing of shops/products to be presented to the user.
3. Security Clearance:
Runs captchas (math puzzles , bot checks) in your browser to verify you're a real person, protecting OpenAI from scrapers. Real time searches are quite expensive so Open AI does not want to waste it in bots (I feel bad for prompt tracking).
4. Sidebar Synthesis:
This one transform all data retrieved about the product reviews into a fancy response in the sidebar. The script calls a “gpt-5-5-instant“ to be able to process all information on the go without consuming too expensive tokens into the new information (I guess).
After this beautiful process is completed we finally get the “ranked“ options. It is in that final ranking where I spotted the dependency of ChatGPT on Google Merchant. As example, the listed URL from Amazon.nl has some juicy UTM? parameters:
Amazon Google Shopping Link on ChatGPT response
Parameters as follows:
source=ps-sl-shoppingads-lpcontext
Internal traffic source. Paid Search (ps), Sponsored Listing (sl). In short, coming from the Google Shopping URL
ref_=fplfs
Internal referral tag (ref_). Tracks which specific UI widget or layout feature generated the click (in this case, First Party Landing Feature Set)
psc=1
Purpose: Product Specific Code (variant selector). 1 forces the page to load a specific product variant pre-selected (tried I few but could not get anything changed)
smid=A17D2BRD4YMT0X
Seller Merchant ID. Identifies the exact seller account listing the item. The product ID for Amazon
utm_source=chatgpt.com
I call “dependency“ on Google Marchant because despite the script backend-api/search/product_update tries different sources for product data, the links indeed are the ones from Google Shopping ads.
My 3rd Clue: The Ranking Parameters
How is the final list built?
Here the list of parameters I found:
Most critical ranking factor on the listing is if the store adopted Agentic Checkout.
Agentic checkout is a fancy way to say embedded checkout: the purchase happens on ChatGPT not at the merchant’s website.
Here is an example when the sidebar triggered while my VPN was in the US:
Top result is Best Buy: instead of the CTA “Visit“ you find “Open“.
Top ranked results must have embedded checkout in the LLM (ChatGPT)
You can buy straight from your live conversation which I find quite convenient 👀
Best Buy allowed users to buy straight from ChatGPT
The Agentic Commerce Audit (Early-Bird)
Is your catalog not being shown on ChatGPT? We help you to solve it.
⚡ Early-Bird: €2,500 (Standard: €5,995) · Limited to 5 brands
Top 5 Ranking Parameters for eCommerce GEO
Fun fact, you can bypass all citations and recommendations given in any given response to finally get the click from the sidebar list given by Chat GPT. In this case, none of the websites listed were used to build the initial response.
I picked my top 5 ranking factors (after the Agentic Checkout). Below I will list all factors I found (Google Doc link):
1 - Calculated Landed Cost (price_details.total)
Sticker price alone does not win the slot. ChatGPT sorts offers by total delivered cost ($base + shipping + tax$) tailored to the user's detected location. Delivering the lowest landed cost triggers the "Best price" UI tag and awards the internal precedence bump ("debug_info.p": "101").
The price (cheapest) is a critical factor on the final product listing given by Chat GPT
All product information (including shipping) will be taken into account for the final display
⚠ I found quite interesting that ChatGPT gave me a MacBook for 53.89 EUR (!). It is meant to be rented not bought. I am not against this just that I did not expected. Surprised positively.
2 - Real-Time Stock Verification (availability)
Items that are backordered, out of stock, or fail real-time ping verification are immediately dropped or hidden beneath the fold. Note: even if the stock has to be confirmed the price will allow the product to be ranked 1st.
3 - Regional Entity Matching (merchant_hints & Geo-Alignment)
Before querying individual stores, the initial query planner injects anchor retailers into merchant_hints based on the user's country and language. When my VPN was in the US, the sidebar refreshed and gave me US stores.
Best
When landed prices are nearly identical, the algorithm breaks ties using domain authority. Direct manufacturers (Apple, Sony) receive preferential "maker_status" routing, followed by Tier-1 national retailers with structured feed verification, penalizing unverified third-party marketplace resellers.
5 - Fulfillment Velocity & Delivery SLA (delivery_estimate / shipping_speed)
How fast the product reaches the doorstep. Guaranteed 24–48 hour local fulfillment acts as a vital quality filter, routinely beating out cheaper cross-border options burdened with week-long shipping estimates.
The full list of factors I found (so far):
Quick eCommerce GEO Checklist
Write Constraint-Driven Descriptions, Not Fluff
Always include (GTIN/MPN) (Product Knowledge Graph)
Structure Variants Under Shared Parent IDs (
item_group_id)Provide explicit
min_handling_timeandtransit_timeestimatesIntegrate with the Agentic Commerce Protocol (ACP) (if active in your region)
Structure for All-Inclusive Cost, Not Single Item Price (
shipping [price])Stock & Pricing Via API Feed. CSV or XML for static values (titles, description)
Apply to send your merchant feed to Chat GPT: https://chatgpt.com/merchants/
I will keep investigating and dropping soon more of my findings.
The Agentic Commerce Audit (Early-Bird)
Is your catalog not being shown on ChatGPT?
4 Weeks Turnaround, we run a full diagnostic on your feeds, full-site technical check up, crawling, ranking and timeout bottlenecks, check current gaps and give you the exact technical roadmap to lock down top-3 placement.
⚡ Early-Bird: €2,500 (Standard: €5,995) · Limited to 5 brands
Let’s meet in REAL LIFEY
That’s it for today, practical and simple.
Until next time 👋
—