Before you enrich product copy for AI, check what crawlers actually receive

The tempting AI-shopping project is a copy project. Add "great for gym bags" to the granola bar. Spell out that the sunscreen is reef-conscious and fragrance-free. Map every bullet to a prompt you heard in ChatGPT.

That work helps only if the machine that reads your page actually receives the text.

On November 2, 2026, Search Engine Journal published a recap of seoClarity's webinar "AI Shopping: Get Your Products Found In Google & ChatGPT." Chris Sachs and Aris Vrakas walked through four moves: make product pages readable to bots, enrich content in shopper language, keep feeds and pages consistent, and measure what changed. The order matters. Retrieval first.

Step 1: See what the crawler gets

Vrakas's opening point, as SEJ reports it, was technical and blunt. Product content needs to be available to bots, not only visible to someone using a browser. On JavaScript-heavy pages, shoppers can see details that an automated fetch does not expose.

For a CPG team reviewing a DTC site or a custom storefront, that distinction is the whole first week of work.

Checklist: crawler vs browser on ten priority SKUs

Check	Pass if
Product name	Present in raw HTML
Price and currency	Present without waiting on a script
Availability	Present in HTML or server-rendered JSON-LD
Pack size / net weight	Visible in HTML
Key ingredients or materials	Visible in HTML
Primary image URL	In HTML
Variant options (flavor, size, scent)	Listed in HTML, not only drawn after click
Canonical URL	Stable and self-consistent

Run the same URL through a plain fetch (curl or "view source") and through a normal browser. Where the browser wins and the fetch loses, you have a retrieval gap. Vrakas discussed prerendering and cached HTML as options when rendering is the bottleneck. The SEJ recap's line worth taping above a backlog board: address access before investing heavily in enrichment.

Blend Commerce's audit is a useful neighbour here. Of 171 retailers whose product pages Blend could read, 82% had product schema, but only 18% published a GTIN, 22% shipping details and 20% a returns policy. Being "readable" is not the same as being complete. First make sure the page loads for an automated visitor; then make sure the fields are present.

Step 2: Enrich in the language shoppers use

Once retrieval works, the webinar moved to description quality. The presenters used socks as a running example. Color and fiber composition are useful attributes. They do not answer every question about warmth, thickness, durability or intended use. Conversational prompts often combine several of those needs in one request.

CPG translation is straightforward.

Show the data

Attribute type	Example count on a well-prepared page
Pack facts (weight, count, barcode)	4
Use-case facts (occasion, diet need, storage)	6
Proof points tied to reviews or specs	3

Source: Illustrative figures for this guide, patterned on seoClarity's shopper-language advice in SEJ's Nov 2, 2026 webinar recap. Illustrative figures, not real data. Use to show the mix of pack facts and use-case facts, not as benchmarks.

For a snack bar, pack facts are grams, count and allergens. Use-case facts are "post-workout," "school lunch without peanuts," "no refrigeration." For sunscreen: SPF and active ingredients are pack facts; "under makeup," "for kids who swim," "sensitive face" are use-case facts. Vrakas also discussed using attributable review evidence to support performance claims rather than leaving a model to infer them.

Do not invent claims. Connect real characteristics to the decisions shoppers are already trying to make.

Step 3: Close the loop between feed, page and structured data

Sachs warned against treating the product feed as a separate project owned somewhere else in the company. Information added to a feed should also inform the product page and its structured data.

Vrakas's reason: when a platform revisits a product page, identity, price and supporting details need to be clear and coherent. An enriched feed does not remove the need to make the corresponding page understandable.

Google's product structured data documentation is the baseline for what belongs in machine-readable form on the page. For CPG, we would add a simple ownership rule:

1. Merchandising approves the attribute.
2. The same value is written to the feed, the PDP and JSON-LD (or equivalent).
3. QA fetches the live HTML and the feed export the next day and diffs the fields.

If SEO, paid media and the website team keep three versions of "fragrance free," assistants will pick one at random and shoppers will see another.

Step 4: Measure three different things

The presenters separated:

• Product-grid visibility: how often priority products appear in shopping carousels for relevant queries.
• AI mentions and citations: whether an answer names a product or cites supporting content, and how the brand is described.
• Paid-feed performance: impressions, clicks and conversions from feed-driven campaigns.

These are different signals. An AI mention is not a sale. Sachs's line in the recap: the best SEO teams obsess about testing. Vrakas added that large catalogs can compare product groups, while smaller sites can improve content around a defined topic and compare it with topics left untouched. The useful habit is having a comparison set, not assuming every visibility change proves the optimisation worked.

Signal	Baseline date	Priority SKU set	Comparison set	Owner
Shopping carousel presence
AI mention / citation sample
Paid feed conversions

Fill with your own numbers. Do not paste invented percentages into the sheet.

A two-week CPG plan

Week 1: retrieval. Pick ten SKUs that matter for the next 90 days. Fetch each PDP as a crawler would. Log missing fields. Fix server rendering or prerender for the worst three pages first. Retest.

Week 2: enrichment and consistency. For those ten SKUs, add two use-case attributes that are already true. Push the same values to the feed and structured data. Snapshot carousel presence and ten assistant answers before and after. Leave ten other SKUs untouched as a comparison group.

If your storefront sits behind aggressive bot protection, fix allow-list questions in parallel (tip 47). Enrichment cannot help a page that never loads for the agent.

Common failure modes on CPG sites

We keep seeing the same four retrieval failures:

1. Variant drawers. Flavor and size only appear after a click, so the fetch sees one default SKU.
2. App-only nutrition panels. The browser paints a nice table; the HTML has an empty root.
3. Geo-priced HTML. A US crawler sees a price; a bot from another region sees a login wall or a blank buy box.
4. Consent banners that eat the DOM. If the first paint is a blocker with no product JSON-LD behind it, some fetchers stop.

Fixing those four usually recovers more AI visibility than another round of adjective-rich bullets. Write the shopper-language copy after the HTML already contains the facts.

XENA Intelligence builds tools that ask assistants shopper-style questions about products. Pair that kind of answer sampling with the crawler HTML check from this guide. SearchPanel helps you see the answers; it does not replace viewing source on your own PDP.

A note on retailer-hosted pages

Many CPG brands do not control the PDP HTML. When the live page is a retailer's template, your levers are the feed attributes you submit, the enriched content programs the retailer offers, and any brand site pages assistants can cite as a second source. Run the same crawler-vs-browser check on your own product URLs even if Amazon or a grocer hosts the buy box. Assistants still read brand sites when retailer pages are thin.

Keep a dated spreadsheet of the ten SKUs, the fetch date, and a pass/fail for each HTML field. Without that log, enrichment debates restart every quarter with no evidence.

What "done" looks like

You are done with the first pass when:

• Ten priority PDPs expose name, price, availability, pack size and key ingredients in raw HTML.
• Feed, PDP and structured data agree on those fields.
• You have a dated baseline for grid visibility, AI mentions and paid-feed results.
• One deliberate change has a comparison group.

After that, write better copy. Not before.