Optimising Product Data for Google & AI Results

As OpenAI tools like ChatGPT begin surfacing product listings and shopping results directly in their interfaces, the accuracy, quality, and optimisation of product feeds are now more critical than ever.

A comprehensive audit, like the one outlined below, ensures that businesses are fully aligned with best practices for Google Shopping and feed management, helping them maintain visibility, accuracy, and competitive performance across both search engines and emerging AI-powered platforms.

The increasing reliance on AI to interpret and surface e-commerce content means feeds must be:

• Technically precise (e.g., correct use of GTINs, sale prices, and categorisation),
• Visually appealing (high-quality images free from disallowed elements),
• Content-rich (optimised titles and descriptions tailored for both search relevance and AI parsing),
• And strategically structured (with logical use of custom labels and item groupings).

How AI shopping surfaces actually read your feed

Before you run the audit, it helps to know what you are actually optimising for. Google Shopping and AI assistants do not read your products the same way.

Google Shopping matches structured attributes in your feed against a shopping query, then ranks eligible offers. AI shopping surfaces work differently. ChatGPT’s shopping results are organic: there is no bidding, no keyword matching and no paid placement. Recommendations are driven by semantic reasoning over product metadata, and rankings lean on data completeness, pricing transparency, review signals and clear descriptions rather than ad spend. ChatGPT can pull that data from two places: pages it crawls, and structured product feeds shared through OpenAI’s merchant programme, which accepts refreshes as often as every 15 minutes.

The practical upshot is that your Google Merchant Center feed is doing double duty. Google itself is the source of truth for its own AI Overviews and AI Mode, and the same clean, complete feed is increasingly what AI assistants read when they surface products. In other words, feed hygiene is no longer just a Shopping tactic; it is the foundation of AI product visibility too. (If you are also trying to measure that traffic, see how to track LLM referral traffic in GA4.)

That is why the prompt below audits classic Shopping best practice first, then layers on the AI-specific tweaks.

Google Shopping best practice vs AI/ChatGPT tweaks

Feed Field	Google Shopping Best Practice	Extra Tweak for AI/ChatGPT Surfacing	Example
Title	Front-load brand, product type and key attributes. You can submit up to 150 characters, but Google only displays around 70, so put what matters first.	Write it as a human would read it aloud, not as keyword soup. If the title is generated with AI, submit it in the structured_title attribute or Google may disapprove it.	Charlie Cotton Oxford Shirt - Men’s Slim Fit, Blue, Size M
Description	Key features in the first 160 characters; no HTML, promos or symbols.	Use the full natural-language description (ChatGPT’s feed accepts up to 5,000 plain-text characters) covering materials, fit and use cases an AI can quote.	”Breathable slim-fit Oxford shirt in 100% cotton, ideal for smart-casual wear…“
product_highlights	List 4–6 concise highlights, each up to 150 characters (Google accepts 2–10).	Frame each as a buyer benefit an assistant can lift verbatim into an answer.	”Wrinkle-resistant cotton”, “Machine washable at 30°C”
GTIN	Provide a valid GTIN for every variant that has one. Google reports products with GTINs get up to 40% more clicks.	The same GTIN lets AI match and de-duplicate your offer across sources, improving the odds you are the cited result.	5012345678900
Reviews & ratings	Not a core Shopping title field, handled via product ratings.	ChatGPT explicitly weights review_count and star_rating, so populate them where the feed allows.	4.7 stars, 320 reviews
Price & availability	Accurate, VAT-inclusive and consistent with the product page.	Keep it near real-time; AI feeds can refresh every 15 minutes, and stale prices get filtered out.	£39.00, in stock

A worked before/after example. Take a weak title like Blue Shirt. It tells Google almost nothing and gives an AI assistant nothing to reason about. Applied to the standard above, it becomes Charlie Cotton Oxford Shirt - Men's Slim Fit, Blue, Size M: brand, product type, fit, colour and size are all front-loaded, it reads naturally, and it stays inside the 150-character limit. That single change is the difference between a listing that gets skipped and one that gets surfaced. For more feed-level wins, see our Google Shopping feed hack on duplicate product feeds.

Shopping Audit Prompt

Instructions:

Step 1: Select “Deep Dive Analysis” in ChatGPT
Step 2: Copy and paste the prompt below
Step 3: Add your XML feed (you may attach it as a Word doc)

---

Prompt:

I would like a comprehensive, in-depth Google Shopping audit of the product feed (XML) I will send you shortly, focused on the (LOCATION HERE) market.

The audit should specifically focus on best practices in Google Shopping and Feed Management.

Please analyze the product feed and use best practices from Google Shopping Ads, particularly from Channable and DataFeedWatch.

---

Structure of the Report

Present your analysis in a table format:

• Vertical (Rows): per element

• Horizontal (Columns): Element, Analysis, Best Practice, Strengths, Recommendations, Impact (High/Medium/Low)

---

Audit Sections:

Titles [title]

• Are titles clear and is key product info front-loaded for search relevance?

• Do they include brand, size, color, model number, and avoid being too long or vague?

• Are the full 150-character limits used effectively?

Descriptions [description]

• Are descriptions clear and detailed, with key features in the first 160 characters?

• Are they free of links, promos, symbols, HTML, or unnecessary capitalization?

• Do they include relevant features (material, color, size, age group, etc.)?

Images [image_link]

• Are images free of borders, branded backgrounds, or promotional elements?

• Are they high quality (at least 1500×1500), with proper focus and scale?

• Are image URLs functional and only one image_link used per product?

Product Prices [price, sale_price & sale_price_effective_date]

• Are feed prices consistent with the product page and up-to-date?

• Are prices formatted correctly with VAT included, no hidden costs?

• Are sales prices correctly entered and timed?

Categorization [google_product_category & product_type]

• Is google_product_category correctly mapped to the most specific option?

• Is product_type structured hierarchically and not overly generic?

• Are breadcrumb paths clear and keyword stuffing avoided?

Product Identifiers [brand, gtin, mpn & ean]

• Are GTINs or EANs valid and applied to each variant?

• Is the brand field accurate and consistent in language/script?

• Is MPN used correctly per manufacturer and per variant?

Groupings [item_group_id]

• Is the same item_group_id used for variants of a parent SKU?

• Is it not used where there are no variants?

• Are variant attributes (e.g. color/material) included?

Product Details [product_detail & product_highlight]

• Are these fields used for extra information not covered elsewhere?

• Are sub-attributes properly structured (section_name, attribute_name, attribute_value)?

• Are 4–6 highlights listed per product, max 150 characters each?

Custom Labels

• What custom labels are used, and for what purpose (e.g., margin, season, inventory)?

• Are labels logically segmenting products by margin/performance?

• Do labels reflect inventory or performance (e.g. “Low Stock”, “Underperforming”)?

Other Relevant Attributes [color, gender, material, size, condition]

• Is color listed using a primary color and up to 2 secondaries (e.g., “blue/white”)?

• Is material listed accurately, up to 3 items (e.g., “cotton/polyester/elastane”)?

• Are sizes clearly defined, standardized, and consistent?

---

Quick Wins & Growth Opportunities

List at least 10 easy optimizations that deliver quick improvements.

For each:

• State priority

• Ease of implementation

• Potential impact

---

Report Requirements:

• Must include over 150 Google Shopping Ads checkpoints

• Must be actionable and concrete (no fluff or generic tips)

• Should include tables, visual aids, example screenshots, and comparisons

• Written for an experienced e-commerce or Google Ads professional

• Must be ready-to-implement by a team immediately

• Well-structured per page and per audit section

---

This audit should result in a complete Google Shopping Ads advisory document, serving both strategic and tactical improvement needs — effectively a blueprint for an optimized product feed.

How to export your feed to run this audit

Step 3 assumes you can hand the model a feed. Here is how to get one from the most common platforms:

• Shopify: connect the Google & YouTube channel to sync your catalogue to Google Merchant Center, then pull the primary feed from there. Alternatively, a feed-management app can generate a downloadable XML or CSV file directly.
• WooCommerce: use a product-feed plugin to generate a Google Shopping feed as XML or CSV, then download the file. You can point the audit at the exported file rather than the live URL.
• Google Merchant Center: open Products, select your primary feed, and download or copy the current feed file. This is the cleanest source because it is exactly what Google (and increasingly AI surfaces) already reads.

If you do not have a full feed to hand, export a representative sample of 20–50 SKUs that spans your best and worst listings. That is enough for the model to spot systemic problems.

What a completed audit row looks like

Before you run the prompt, it helps to know what a “good” filled-in row should look like, so you can judge the output. For the Titles element you should get back something like this:

Element	Analysis	Best Practice	Strengths	Recommendations	Impact
Titles [title]	62% of titles lead with an internal SKU code, not the product. Brand missing on 40%.	Front-load brand, product type and key attributes within ~70 characters.	Colour and size attributes are consistently included.	Rewrite templated titles to “Brand + Product Type + Key Attribute + Variant”. Move SKU codes out of the title.	High

If a row comes back vague (“titles could be improved”), push the model to quantify and cite the specific SKUs, exactly as the prompt demands.

Model choice and file limits

A few practical notes so the prompt does not fail on a real feed:

• Use a reasoning-capable model with browsing or a large context window. A quick, cheap model will skim rather than genuinely audit 150+ checkpoints.
• Mind the upload limits. Large XML feeds can exceed a single chat’s file-size cap. If the model truncates or complains, split the feed into chunks (for example a few hundred SKUs at a time) and audit each, or paste a representative sample rather than the whole file.
• Plain text beats Word. Attaching the raw XML or a CSV usually parses more reliably than wrapping it in a Word document, which can strip or reflow the markup.

Feed hygiene is the foundation, but it is only half the job. Once your products are surfacing, make sure the campaigns behind them are structured to convert — see why running Performance Max without Shopping leaves money on the table and our small e-commerce Google Ads strategy.

FAQ

What does “feeding the AI machine” actually mean for my product data? It means treating your product feed as the primary input that both Google and AI assistants read to decide whether, and how, to surface your products. The cleaner and more complete the feed, the more likely an assistant is to pull your product into an answer. In practice it is the same Google Merchant Center feed doing the work across Shopping, Google’s AI results and tools like ChatGPT.

Does ChatGPT use my Google Shopping feed? Largely, yes. Google Merchant Center has become the de facto source of truth that many AI shopping surfaces read from, and independent analyses suggest a large share of ChatGPT’s product results trace back to Google Shopping data. You can also share a feed directly through OpenAI’s merchant programme. Either way, the quality of your structured product data is what determines whether you get cited.

Are ChatGPT shopping results paid ads? No. ChatGPT’s shopping results are organic, with no bidding, keyword matching or paid placement. Products are ranked on data completeness, pricing transparency, review signals and description quality — which is exactly why the feed audit matters more than any ad-side lever here.

How do I optimise product titles for AI search versus Google Shopping? For Google Shopping, front-load brand, product type and key attributes within roughly 70 visible characters. For AI surfacing, keep that structure but make it read naturally, since assistants reason over the language rather than keyword-matching it. If a title is generated with AI, submit it in the structured_title attribute so Google does not disapprove it.

What’s the single highest-impact feed fix? For most catalogues it is titles and GTINs. Rewriting weak, code-led titles into “Brand + Product Type + Key Attribute + Variant” format lifts both Shopping performance and AI legibility, and Google reports products with valid GTINs get up to 40% more clicks while also being easier for AI to match across sources.