How to Optimise a Product Page for ChatGPT Shopping

How to Optimise a Product Page for ChatGPT Shopping

Optimising a product page for AI shopping means making sure an assistant can reach the page, read it without executing JavaScript, extract every attribute a buyer might specify, and verify price and availability. Most D2C product pages fail at least two of those. This is the nine-step version, in the order the steps actually depend on each other.

Before the checklist: what you are actually optimising for

Two things worth being clear about, because they shape every step below.

There is no submission process. You are not applying to be included in an AI shopping result. You are making your page readable enough that the systems assembling those results can use it. That is the whole mechanism, and it is why this work overlaps almost entirely with good technical SEO.

Shopping surfaces in AI assistants are changing fast. Feed programmes, merchant integrations and checkout mechanisms differ between providers and change frequently.

VERIFY: Before publishing, confirm the current state of every named shopping programme, feed requirement or merchant integration against the provider's own documentation. Do not name a programme you have not verified exists today, and do not describe eligibility requirements from memory or from another agency's post. Where a specific programme cannot be confirmed, describe the underlying technical requirement instead — that part is stable.

The nine steps below are deliberately written to hold regardless of which specific programme is current, because they address how the page is read rather than how a particular product works.

Step 1 — Confirm AI crawlers can reach the page

Nothing else matters if this fails, and it fails more often than people expect — usually because a security product, a CDN rule or a blanket robots.txt directive is blocking bots nobody audited.

• Check robots.txt for rules affecting AI crawler user agents
• Check your CDN or WAF for bot-blocking rules applied by default
• Check for rate limiting that returns errors to non-browser agents
• Confirm product pages are not noindex
• Confirm they are actually indexed, in Search Console

Decide deliberately which crawlers you allow. Some brands block AI crawlers on principle over training-data use. That is a legitimate position — but it is incompatible with wanting to appear in AI shopping results, and the two decisions need to be made together rather than by whoever configured the CDN.

Step 2 — Make the page render without JavaScript

If your product data loads client-side, a system fetching the raw HTML may see an empty shell.

Test it: disable JavaScript in your browser and load a product page. Are the title, price, description and attributes there? If not, that is what some crawlers see.

The fix is server-side rendering or static generation for product pages. On many Indian D2C stacks this is a platform-level decision rather than a quick change, which is why it belongs early — it may take longer than everything else combined.

Step 3 — Implement complete Product schema

Complete, not partial. Most implementations stop at name and price.

Include: name, description, brand, sku, gtin where you have one, image, offers (with price, priceCurrency, availability, itemCondition), aggregateRating and review where genuine, plus additionalProperty entries for attributes that have no dedicated field.

additionalProperty is the underused one. Material, dimensions, capacity, compatibility, care instructions — each becomes a structured, machine-readable fact rather than a sentence buried in a paragraph.

Validate every page with a structured data testing tool. Schema that is present but malformed is worth roughly nothing, and malformed schema is common.

Step 4 — Put every attribute in the page text

Structured data and visible text should agree, and the text version matters independently.

Every specification, ingredient, dimension, material and compatibility note as readable text on the page. Not in an image. Not in a PDF. Not only in a filter dropdown. Not behind a tab that loads on click.

The test from our post on product imagery applies here: load the page with images disabled and see what a stranger could still learn. Anything missing is a gap.

Step 5 — Write a description that answers purchase questions

Product descriptions written as brand copy give an assistant very little to work with. Descriptions written to answer purchase questions give it a great deal.

Cover, in plain language:

• What the product is and what it is made of
• Who it suits — and, importantly, who it does not
• How it compares to the obvious alternative in your own range
• What it does not do
• What is included

The "who it does not suit" line is unusual enough to be a genuine advantage. An assistant matching a customer's stated constraints uses exclusions as readily as inclusions, and a page that states them is easier to match confidently than one that claims to suit everyone.

Step 6 — State price, availability and policies as data

• Price and currency in schema, matching the visible price exactly
• Availability status that is actually accurate and updates
• Shipping timelines, returns window, warranty terms and country of origin stated in plain text on or linked from the page

Price and availability mismatches between your schema, your page and your feed are a common and self-inflicted problem. A system that finds contradictory data has reason to distrust all of it.

Step 7 — Get review text onto the page

Star ratings are a number. Review text is description — in customer language, which is the language the queries arrive in.

Make sure review content renders in the HTML rather than loading from a third-party widget that a crawler cannot execute. This catches a lot of brands: the reviews are there for shoppers and invisible to everything else.

Mark up genuine reviews in schema. Do not mark up reviews you do not have.

Step 8 — Check your merchant feed matches your page

If you maintain a product feed for shopping surfaces, treat it as a first-class marketing asset rather than an export nobody reads.

• Titles and descriptions in the feed should match the page, not be a truncated legacy version
• Attributes populated properly, including the optional ones
• Prices and availability in sync
• Images meeting the platform's requirements

VERIFY: Feed specifications differ by platform and change. Confirm current requirements against the platform's documentation rather than relying on a general description.

Step 9 — Test what an assistant actually says about your product

The step that closes the loop, and the one that will change your team's mind faster than any argument.

1. Ask an assistant to recommend a product in your category with three specific constraints a real customer would state.
2. Note whether you appear, and what the assistant says about you.
3. Ask it directly to compare your product against two named competitors.
4. Write down every factual error it makes about yours.
5. Trace each error back to the source that should have supplied the fact — page text, schema, feed or reviews.
6. Fix that source. Re-test in four to six weeks.

Step 4 is the useful one. Errors are diagnostic: each one points at a specific gap.

The three checks most brands skip

The JavaScript check. Ten seconds, and it invalidates everything else when it fails.

Schema validation on more than one page. Teams validate the template on one product and assume the rest are fine. Variant products, sale prices and out-of-stock items frequently break the pattern. Check ten.

Feed-to-page consistency. Nobody owns this. The feed was set up once and has drifted since.

What this will not do

• It will not guarantee inclusion in any AI shopping result. There is no submission and no guarantee. You are improving readability, not buying placement.
• It will not fix an uncompetitive product or price. Assistants comparing on stated criteria will surface that faster than a human browsing would.
• It will not produce measurable results quickly or cleanly. Attribution here is poor. The honest measure is the manual test in step 9, repeated monthly.
• None of this is wasted if AI shopping stalls. Every step improves conversion, accessibility and conventional search performance. That is what makes it a safe investment rather than a bet.