What ChatGPT shopping research does
OpenAI introduced shopping research in November 2025 for purchases involving comparisons and constraints. Rather than asking for a single product name, you can explain the job the product must do and work through alternatives.
For example, a shopper might need a compact desk lamp for a shared bedroom, with a warm light setting and a firm budget. Brightness alone is not enough. The comparison should account for size, glare, controls, and whether an advertised feature applies to the exact model being sold. Those details make the research more useful than a generic list of popular lamps.
The current shopping-research guide describes selecting products already shown in chat and choosing Research. ChatGPT asks follow-up questions, explores alternatives, and produces a buyer's guide. You can remove options or refine constraints during the process. Memory may inform personalization when enabled; you can change that in your settings.
How to get a useful comparison
Start with requirements, then distinguish what is essential from what would merely be nice. A product can be an excellent general recommendation and still fail your particular use case.
An illustrative request could be:
Compare desk lamps for a small shared bedroom. The base must fit on my measured shelf, and I need a warm setting for evening reading. Show the exact model, dimensions, power source, and merchant link. Flag specifications you cannot verify instead of guessing.
Add your actual measurements and budget. Then inspect why each option was selected. If the guide prioritizes appearance over fit, correct that priority. If two stores use different names for what looks like the same lamp, ask for the manufacturer model number before comparing their prices.
Keep the process interactive. Removing an unsuitable option is useful feedback, but it does not prove the remaining products satisfy every requirement. Use a short checklist to review the final candidates yourself.
What to verify before trusting a deal
OpenAI's shopping-results documentation says prices can lag behind merchant changes, generated labels are not guarantees, and reviews or ratings are not verified by OpenAI. The initial merchant price also need not be the lowest available.
| Check | Why it can change the decision |
|---|---|
| Exact item and variant | A discount may apply to another size, color, capacity, or model generation. |
| Total purchase cost | Delivery, tax, mandatory fees, and required accessories can outweigh the headline saving. |
| Availability | A listed item may not ship to your address or arrive when needed. |
| Source behind a feature | A review summary may omit conditions that matter for your intended use. |
| Seller and returns | The seller, warranty coverage, and return conditions are part of the purchase, not an afterthought. |
Open the relevant merchant and manufacturer pages. If a specification is unclear, ask the seller before relying on it. When two sources conflict, preserve that uncertainty rather than averaging them into a confident answer.
That is particularly important for labels such as “best price” or “budget-friendly.” Ask what is being compared. A good value within a shortlist is not necessarily the cheapest suitable item in the whole market.
Research and checkout are separate decisions
Shopping research helps choose an item. Checkout is the transaction offered for that item and merchant. Current OpenAI guidance still describes Instant Checkout for some eligible products and merchants; other results take shoppers to the merchant's site. Neither route follows automatically from appearing in a comparison.
Before authorizing a purchase, confirm the final item, seller, delivery details, and total shown in that checkout. Do not treat an earlier chat answer as the final order confirmation. The same caution applies if you change variants after following a recommendation.
What shopping research means for merchants
OpenAI says product recommendations are separate from ads and are not determined by its partnerships. Merchants can improve the information available about their products, but cannot assume a listing or feed guarantees placement.
Keep product names, variants, specifications, policy pages, and purchase links consistent. When customers arrive with a comparison, your own site should help them verify it. An unexplained difference between a product card and checkout creates work for the shopper and support team.
The useful goal is not to control every third-party summary. It is to make the facts verifiable and give visitors a reliable next step when the summary is incomplete. Measure actual referral and support behavior before claiming an AI discovery feature increased sales; a large audience does not establish results for an individual store.
Your storefront assistant needs evidence too
An assistant on your site can work with your catalog and policies, but that does not eliminate mistakes. It needs appropriate source access, current information, and a clear boundary when information is missing.
For example, the eesel Shopify integration distinguishes public storefront crawling from a fully authorized Shopify connection. The latter supports live product and variant lookups alongside synced knowledge. Source sync and live lookup are different: a stored product description is not a current stock check.
This is an illustrative storefront example, not a measured customer result. For your store, verify the exact variant and evidence behind each recommendation. An attractive carousel alone does not establish fit, availability, or a completed purchase.
Review catalog comparisons with eesel CLI
Suppose shoppers keep asking whether two lamp variants fit a narrow shelf. The merchandising owner has corrected a dimension in Shopify, but wants to check whether the store teammate still repeats the old description.
A person, script, or coding agent such as Claude Code, Codex, or Cursor can use eesel CLI to inspect that teammate. It is the same workspace and configuration as the dashboard; its JSON output lets a coding agent examine the results and identify what needs checking.
With Node.js 18.17 or newer, confirm the workspace and intended existing teammate. The name below is an example:
npx @eesel/cli whoami
npx @eesel/cli status --agent "Store product guide"
npx @eesel/cli integrations --agent "Store product guide"
npx @eesel/cli instructions --agent "Store product guide"
npx @eesel/cli activity --agent "Store product guide"
Ask the coding agent to compare the configured sources and recent answers with the owner's approved product record. It should report what the output establishes and what it does not. A connection marked active is not proof that the corrected dimension was retrieved, and a recent conversation may have used a different variant.
If synced knowledge is stale, the owner can review and authorize a re-sync through the documented Shopify setup. If a live lookup is needed, check that the full connection and relevant read action are permitted. Do not enable order-changing actions merely to answer a product question. Read back the source state after an approved change, then test the answer itself.
Test a known specification and a missing one
CLI chats are billed work. Obtain approval for the test cost and data, and use a non-production configuration with consequential actions disabled or absent. Supply approved test product records first; these illustrative prompts do not upload records or connect Shopify themselves.
npx @eesel/cli new --name "lamp-variant-fit" --agent "Store product guide"
npx @eesel/cli chat "Compare the two approved test lamp variants against the supplied shelf measurement. Identify the exact variant and dimensional evidence for each recommendation. Do not create or change an order." --agent "Store product guide"
npx @eesel/cli new --name "missing-lamp-dimension" --agent "Store product guide"
npx @eesel/cli chat "The approved record omits the lamp base width. Explain whether you can confirm shelf fit, identify the missing information, and suggest how the shopper can verify it. Do not guess or change an order." --agent "Store product guide"
Compare both responses with the approved records. A fresh conversation changes context, not permissions. Check whether the answer names the correct variant, distinguishes documented facts from uncertainty, and avoids inventing a measurement.
Then separately test the actual storefront with permission: product links, card labels, selected variants, and any route to checkout. Terminal JSON cannot demonstrate that a card renders correctly or that a purchase link preserves the selected variant.
Compare costs for the work you need
OpenAI's 2025 launch offered shopping research across Free, Go, Plus, and Pro and described a temporary holiday usage promotion. That announcement is not a promise of unlimited access today; check the feature and current allowance in your account.
A merchant-operated teammate is a different purchase. The current eesel pricing guide uses task-based pricing: light work is free, regular tasks are $0.40, and heavy tasks are $4 on the default pay-as-you-go plan. An upfront annual commitment can discount usage by up to 25%; Enterprise adds $1,000 per month plus usage. Review your actual plan and configured spend cap, including what happens when new work pauses at that cap. Estimate the task mix and review results rather than repeating a retired $299 subscription price or promising a fixed bill regardless of usage.
Keep shopping answers verifiable with eesel CLI
The dashboard example shows why sources and actions need separate checks. Use eesel CLI to inspect and test the same teammate, then review the shopper-facing result in your own store. Try eesel to build that review around your approved catalog and policies.
Sources
Frequently asked questions
What is ChatGPT shopping research?
It is an interactive product-discovery experience that helps compare options against your needs, budget, and trade-offs, then produces a personalized buyer's guide.
How do I start shopping research?
The current help guide describes selecting one or more products already in a chat and choosing Research. Answer its follow-up questions and refine the options as the research runs.
Does ChatGPT always find the lowest price?
No. A displayed price or value label is not proof of the lowest market price. Compare the exact variant, seller, taxes, shipping, availability, and relevant return terms before purchasing.
Does a product recommendation mean I can buy inside ChatGPT?
No. Some eligible products and merchants may offer Instant Checkout, while other results link to a merchant's site. Check the purchase route actually offered for the selected item.
Can a merchant pay to guarantee a shopping recommendation?
OpenAI says product results are selected independently and are separate from ads. Providing product data does not guarantee a particular recommendation or ranking.
Will my own store assistant always have accurate stock?
No. Accuracy depends on its source freshness and permitted lookups. In eesel's Shopify integration, knowledge sources are synced, while supported lookup actions query Shopify live. Public storefront crawling alone does not provide those live actions.
What can eesel CLI check after a shopping comparison?
A person, script, or coding agent can inspect the same teammate's connections, instructions, and activity as the dashboard. Owner-approved paid test chats can check product fit and missing information in an isolated configuration with consequential actions disabled, followed by separate storefront verification.
