AI Visibility for Ecommerce and DTC Brands

Shoppers don't ask AI for a brand. They ask for a product that fits a constraint, then check the brand that comes back. Say you run a hypothetical DTC store selling merino running socks. Your buyers might type:

• "What are the best merino running socks under $25 a pair?"
• "Is [your brand] legit? Do they ship to Canada?"
• "Compare [your brand] and [competitor] for runners with blisters."
• "What is [your brand]'s return policy if the socks don't fit?"
• "Does [your brand] still make the trail crew sock?"
• "Where can I buy [your brand] cheapest right now?"
• "Best sustainable sock brands that deliver to the UK"

Two things follow. First, category prompts decide whether you enter the shortlist at all. Second, brand prompts decide whether the shortlist holds, because a wrong price or shipping answer ends the sale. Test both kinds. For the mechanics of the first kind, see how an online shop shows up in AI shopping answers.

One caution: engines don't run your prompt as typed. OpenAI's help page says ChatGPT search rewrites a query into one or more targeted queries before it searches. Any prompt list is a sample, not the real input.

Each engine reaches the web differently, so a store can be visible in one and absent in another.

• ChatGPT. Search fires at the model's discretion. Without it, answers come from training data with a fixed cutoff (OpenAI's model docs list February 16, 2026 for GPT-5.6 and August 31, 2025 for the Instant path). A product launched after the cutoff cannot appear in an unsearched answer.
• Google AI Overviews and AI Mode. Google says a page must be indexed and eligible for a snippet to appear as a supporting link, with no extra requirements.
• Gemini, Perplexity and Claude. Each uses its own crawlers and retrieval path.
• Microsoft Copilot. Grounds answers in the Bing index.

Otterly's 2026 experiment shows why per-engine checks matter. After Google deindexed its test sites, Google's AI surfaces almost stopped citing them, while Copilot kept citing them because Bing kept them indexed. Check Search Console and Bing Webmaster Tools separately. If you sell through API-driven tools as well as the apps, note that API answers differ from consumer app answers.

When an engine cites other sites instead of yours, being named on those sites is the route into the answer. The evidence shows a strong association here, not a guaranteed effect. For a DTC brand, the usual sources are:

• Marketplace listings where you or resellers sell the product. Old prices and discontinued variants often live on here.
• Review platforms and retailer review pages, which can carry outdated return or shipping complaints.
• Editorial roundups and gift guides that rank products in your category.
• Forums and community threads where buyers compare brands.

Run your test prompts, record which URLs each engine cites, and treat that list as your outreach and correction map. Correct facts you control (your own marketplace listings), and ask editors to update stale ones. If you're weighing a head-to-head page of your own, read whether comparison pages help AI recommend a product first.

AI platforms get specific facts wrong about specific brands. For stores, four categories carry direct revenue risk:

1. Prices. A training-data answer can quote last year's price or a past sale price.
2. Shipping regions. An assistant may say you don't ship to a country you added, or that you do ship somewhere you stopped.
3. Return windows. A 30-day policy quoted as 14 days, or free returns quoted when they now carry a fee.
4. Discontinued products. Assistants recommend a retired SKU, and the shopper lands on a 404.

The fix starts on your own site. State each fact once, in plain words, on a page that can be fetched: "Standard shipping to Canada costs $8 and takes 5 to 7 business days." A specific sentence gives the engine something it can quote. Then remove contradictions: the same return window stated differently on the FAQ, the footer and a product page is a likely source of wrong answers. When you retire or launch a product, the product launch checklist for AI accuracy covers the pages to update.

Access is where the strongest evidence sits. These checks remove blocks. None of them guarantees a citation, but failing any of them can keep you out entirely.

JavaScript-only product data

Vercel's crawler study found that ChatGPT and Claude crawlers fetch JavaScript files but don't execute them, so they can't read client-side rendered content. Independent testing in June 2026 found the same for OAI-SearchBot, PerplexityBot and Claude-SearchBot. Googlebot renders. If price, stock or shipping text loads after hydration, compare the raw HTML with the rendered page. The ecommerce product data check for AI crawlers walks through it.

robots.txt, per bot

Rules are per bot. OpenAI's crawler page says GPTBot governs training and OAI-SearchBot governs ChatGPT search, and "each setting is independent of the others." A blanket AI block can remove you from ChatGPT search without anyone noticing.

User-agent: OAI-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

CDN and firewall rules

OpenAI's help page says eligibility needs both robots.txt permission and a host or CDN that allows its published searchbot IPs. A bot-protection rule that returns 403 overrides robots.txt. From 15 September 2026, Cloudflare sets Training and Agent crawlers to blocked by default on pages that display ads, for new domains, new sites on existing accounts and all existing free-tier customers, while Search stays allowed. Review your bot settings even if you changed nothing. The AI Crawler Checker requests a page as each AI crawler and shows which ones your firewall turns away.

Snippet controls

noindex, nosnippet, max-snippet:0 and data-nosnippet keep content out of AI answers. Themes sometimes apply them to shipping banners or policy blocks by accident. Search your templates for them.

Product, Offer and FAQ markup is not a lever for AI citations. Google says structured data isn't required for generative AI search. Ahrefs tracked 1,885 pages adding schema against matched controls and found AI Overview citations fell 4.6% after the change, with no major lift on other platforms. FAQ rich results were deprecated in May 2026.

One boundary matters for smaller stores: every page in the Ahrefs study was already heavily cited, so whether schema helps a new page enter the answer is untested. Don't rip out markup you have, since Google says existing markup does no harm. Just don't spend your AI budget on it, and be wary of anyone promising citations from it.

Days 1 to 7: baseline. Write 10 to 15 prompts across category, comparison and brand-fact questions. Run them in each engine logged out, and record who's named, what facts are stated and which URLs are cited. The free checker on this site gives a quick snapshot for three buyer questions. For a repeatable method, see how to track visibility for individual products.

Days 8 to 14: access. Diff raw and rendered HTML on your top product, shipping and returns pages. Check robots.txt per bot. Test each AI crawler against your CDN, and review Cloudflare settings if you use it. Confirm index status in Search Console and Bing Webmaster Tools.

Days 15 to 21: facts. List every wrong answer from week one. Fix the source on your site with one clear sentence per fact, in server-rendered HTML. Remove contradictions and redirect retired SKUs to a live alternative.

Days 22 to 30: third parties. Update your marketplace listings. Contact the roundups and retailers that engines cited with stale facts. Re-run the same prompts and compare. Expect some answers not to move yet, since retrieval is discretionary and training data changes only with new models.

Can I track AI visibility for individual SKUs? Yes, as a sample. Write prompts that name the product or its use case and record whether each engine mentions it and states its price and availability correctly. Keep the prompt set fixed and run it logged out so results are comparable over time. Remember the numbers reflect your chosen prompts, not the rewritten queries engines actually run, so treat them as a trend signal rather than a measurement of all shopping traffic.

Why does ChatGPT quote an old price for my product? Usually because the answer came from training data rather than a live search. ChatGPT decides per query whether to search, and unsearched answers reflect a fixed knowledge cutoff. If it did search, the old price may sit on a marketplace listing or review page it cited, or your current price may load only after JavaScript runs. Check the cited URLs and your raw HTML first.

Should I block GPTBot to protect my product content? That's a business choice about training use. Blocking GPTBot does not remove you from ChatGPT search, because OpenAI says OAI-SearchBot controls search and the settings are independent. Allowing GPTBot also isn't documented to help you get cited. What matters for ChatGPT search visibility is that OAI-SearchBot is allowed in robots.txt and can actually reach your pages through your CDN.

Does publishing thousands of AI-written product guides help? The evidence points the other way. In Otterly's 2026 experiment, Google deindexed two sites that each published about 1,000 AI-generated posts, with no manual action shown in Search Console. Google's spam policy treats generating many pages without adding value as scaled content abuse. Copilot kept citing the sites through Bing, but that didn't offset losing Google and its AI surfaces.

Do hidden instructions or AI-only text on product pages work? Don't use them. Hidden text, prompt injection and cloaking content for AI crawlers are manipulation, and research on LLM search agents shows engines differ widely in how they handle injected content. The risk is penalties and misrepresentation, not a durable gain. The defensive move is to watch third-party pages that mention your brand for injected instructions that could misdescribe you.