D2C Product Recommendations in ChatGPT, Gemini and Perplexity: India Testing Playbook

Design a repeatable test and state its limitations

Choose one category, India as the market, a fixed language, 5–10 products including your brand and named competitors, and a dated 15-prompt set. Run each prompt three times per engine in fresh sessions; record account state, location, model, shopping/research mode and whether live web search is enabled. Do not mix personalized and clean-session results. Capture product and brand named, order of appearance, cited or linked URL, stated price/availability, recommendation rationale and any unsupported claim. Results are a snapshot, not a stable ranking. Publish the run date, exact prompts and method whenever you share a score.

Use a 15-prompt category map—not one ‘best product’ query

Build five prompt groups: 1–3 broad category discovery; 4–6 use-case or audience needs; 7–9 constraints such as price, size, ingredient, material or delivery; 10–12 comparisons and alternatives; 13–15 India-specific availability, occasion or climate needs. Example forms: ‘Best [category] in India for [use case] under ₹X’; ‘Compare [brand/product] with [alternative] on [three criteria]’; ‘Which [category] avoids [excluded feature]?’ Keep prompts neutral and identical across engines. Score recommendation share = prompts naming the brand ÷ eligible prompts; top-three share; source share; product-attribute accuracy; and repeatability across the three runs. Review outputs manually because a named product with wrong price or variant is not a win.

Build a product-page truth layer buyers can verify

Each product page should expose, in visible crawlable HTML, the stable product name, brand, model/SKU, category, intended user or use case, price and currency, availability, variants, GTIN/MPN where applicable, materials or ingredients, dimensions or sizing, instructions, limitations or warnings, delivery promise and return policy. Use original images with descriptive alt text and show the source/date for ratings or testing claims. Keep price, stock, variant and policy data consistent across the page, checkout, marketplace and feeds. Avoid unsupported superlatives such as ‘best’ or ‘dermatologist recommended’; a claim needs a named source and scope.

Synchronize Merchant Center, Product/Offer markup and OpenAI feeds

For Google, maintain a Merchant Center feed and valid Product/Offer structured data; Google recommends both because they help it understand and verify product information: https://developers.google.com/search/docs/appearance/structured-data/product. Use stable IDs, correct brand and GTIN/MPN where assigned, variant attributes, high-quality images, price, availability, shipping and returns, and reconcile feed-versus-page mismatches. For ChatGPT, OpenAI announced in March 2026 that Agentic Commerce Protocol supports product discovery feeds and that Shopify Catalog products are integrated; availability and merchant paths can change, so verify current eligibility rather than assuming India coverage: https://openai.com/index/powering-product-discovery-in-chatgpt/. Feed or markup eligibility does not guarantee a recommendation.

Audit corroboration and comparison evidence

Create a source ledger for every material external mention: URL/platform, product/variant, claim, rating or conclusion, review count, date, country, independence and whether the page is current. Separate verified-customer reviews, marketplaces, editorial tests, professional recommendations, creator demonstrations and community discussion; they are not equivalent evidence. Fix incorrect marketplace data and request factual corrections where appropriate, but never manufacture reviews or seed undisclosed endorsements. Publish fair comparison pages only when criteria, method, date, products tested and commercial relationship are visible. A buyer answer in page copy may help users; FAQ markup itself is not an AI-recommendation lever.

Run a 30-day experiment and measure directional change

Days 1–5: freeze the 15 prompts, complete three runs per engine and log sources and errors. Days 6–12: audit five priority product pages, schema and feeds against the truth-layer fields. Days 13–20: correct high-impact conflicts in price, availability, variant, identifiers, claims and policies; validate Google Merchant listings and rich-result reports. Days 21–25: add one evidence-led comparison or buying guide and close legitimate corroboration gaps. Days 26–30: rerun the original prompts under the same settings. Report recommendation share, top-three share, attribute accuracy, repeatability, cited-source mix and identifiable referral/conversion activity. Do not claim causation from one before/after run. Request a 15-prompt category visibility check at /free-plan and compare transaction readiness with /blog/agentic-commerce-india-ai-shopping-agents-ready.

What each engine result can—and cannot—tell you

Treat ChatGPT, Gemini and Perplexity as separate test surfaces. They can use different search, shopping, merchant and citation systems, and their behavior changes over time. A linked product card, cited web answer and unlinked conversational mention are different result types; report them separately. OpenAI notes that shopping research may still make errors about price and availability and advises checking merchant sites: https://openai.com/index/chatgpt-shopping-research/. Repeat the fixed study monthly or after a material catalog change, not daily, and keep an update log.

Frequently asked questions

How do I get a product recommended by ChatGPT, Gemini or Perplexity?

There is no guaranteed tactic. Start by testing a fixed prompt set, then make product facts accurate, complete, crawlable and consistent across pages, feeds and legitimate third-party sources. Correct technical eligibility and evidence gaps, rerun under the same conditions and report the limitations.

Why is an AI assistant recommending competitors instead of my brand?

Possible reasons include missing or conflicting product data, weaker fit for the prompt constraints, limited availability, stale price or stock, thin independent evidence, inaccessible pages or simple run-to-run variation. Inspect the linked sources and factual rationale rather than assuming a single ranking factor.

Does Product structured data guarantee AI recommendations?

No. Valid Product/Offer markup can help systems understand page facts and can support eligible Google product experiences, but it does not guarantee a rich result, AI citation or recommendation. Mark up only visible, accurate content and keep feeds and pages synchronized.

How should AI product visibility be measured?

Use a disclosed prompt set, repeated runs and metrics for recommendation share, top-three share, source mix, attribute accuracy and repeatability. Track identifiable referral and conversion data separately. A change after page edits is directional evidence, not proof that the edits caused the change.