AI search for online shops: how to become visible in ChatGPT, Gemini, Claude and Perplexity

How AI search engines find their sources

For current answers with source links, AI search engines draw on a search index that a crawler has built beforehand, or fetch a page live when a user asks for it. Providers therefore split their bots by purpose.

Which indexes sit behind them: according to Google, AI Overviews and AI Mode use the regular Search index. A page must be indexed and eligible to show with a snippet; there are no further technical requirements. OpenAI, Anthropic and Perplexity describe their own search crawlers. Whether they also use third-party search indexes is not disclosed in their crawler documentation.

robots.txt: control training and search separately

You can exclude your content from model training and still appear in AI search; according to OpenAI, each setting is independent. If you block OAI-SearchBot, however, your site is not shown in ChatGPT search answers, at most as a plain navigational link. According to OpenAI and Perplexity, changes take effect after about 24 hours.

• Allow search: do not block OAI-SearchBot, Googlebot, Claude-SearchBot or PerplexityBot.
• Block training (optional): sections for “User-agent: GPTBot”, “User-agent: ClaudeBot” and “User-agent: Google-Extended”, each with “Disallow: /”.
• Subdomains: each one needs its own robots.txt, as Anthropic stresses.
• Firewall and CDN: bot protection that turns crawlers away acts like a block. OpenAI and Perplexity publish IP lists of their bots so you can allow them. Anthropic considers blocking by IP address unreliable, because the bots then cannot read robots.txt.

Two traps: Google-Extended is not a pure training switch; according to Google, it also covers grounding in Gemini Apps and the Vertex AI API. It does not affect Google Search: the only way out of AI Overviews and AI Mode is nosnippet, data-nosnippet, max-snippet or noindex, and those also affect classic results. Also, according to OpenAI and Perplexity, user-requested fetches do not necessarily follow robots.txt; for Claude-User, Anthropic says the opposite.

Technical foundations

HTML that reads without JavaScript

Google renders JavaScript, but still calls server-side rendering or pre-rendering a great idea because not all bots can run JavaScript. OpenAI, Anthropic and Perplexity do not document this for their crawlers. A measurement by Vercel (secondary source, December 2024) found no JavaScript rendering by the major AI crawlers apart from Google's infrastructure and Applebot. Product name, price, availability, description and FAQ therefore belong in the HTML the server delivers.

Structured data

According to Google, there is no special schema for AI Overviews and AI Mode; structured data must, however, match the visible text. For shops, Google recommends Product (ProductGroup for variants), Review, Organization with contact details and return policy, and BreadcrumbList.

Product data, feeds and sitemaps

Google counts an up-to-date Merchant Center and Business Profile among the basics that also apply to the AI features in Search. OpenAI publishes a product feed specification through which merchants supply products for search and shopping in ChatGPT; required fields include title, description, product URL, brand, image, price and availability. For Bing and Copilot, Microsoft advises an XML sitemap with the true modification date (lastmod) and IndexNow. IndexNow submissions are shared with all participating search engines but do not guarantee indexing.

Content that gets cited

None of the providers asks for special optimisation for AI search. Google recommends helpful, reliable content and author bylines where readers expect them. For citations in Copilot, Microsoft advises clear headings, tables and FAQ, claims backed by data and sources, and regular updates. For shops, that means:

• Direct answer first: the key statement sits in the first paragraph, not after a long introduction.
• Concrete data: dimensions, material, compatibility, delivery time and returns as text or a table, not only in an image.
• Evidence: certification marks, standards or test results with a link to the original source.
• Author and date: a real author and the date of the last content review. Google calls changing dates just to make pages look fresh a warning sign.
• Your own knowledge: experience from customer service and returns, or your own comparisons that exist nowhere else.

Research (secondary source): in the paper “GEO: Generative Engine Optimization” (KDD 2024), source citations, quotations and statistics raised visibility by 30 to 40 percent on the authors' own metric in their test setup; keyword stuffing barely helped. This is a lab result, not a promise for real AI search engines.

Mentions on other websites

AI answers rarely rest on a single page. In AI Mode, Google fans a query out into several related searches across subtopics and data sources, and Microsoft recommends describing brands, products and terms consistently across text, images and video. This leads to a recommendation, not an official ranking rule: make sure your shop is mentioned correctly and consistently where your customers do their research, such as trade media, product tests, comparison sites and marketplaces. Mentions bought or generated in bulk are no shortcut.

llms.txt, put honestly

llms.txt is a proposal by Jeremy Howard dated 3 September 2024: a Markdown file that summarises a website's most important pages for language models. It is not an official standard. Google states explicitly that AI Overviews and AI Mode need no new machine-readable files or AI text files, and the crawler documentation of OpenAI, Anthropic and Perplexity does not mention it. That AI providers offer an llms.txt for their own developer documentation does not show that their search products read it. It does no harm, but it belongs at the bottom of the list, behind readable HTML, a correct robots.txt and maintained product data.

Measuring visibility in AI search

• Google Search Console: clicks and impressions from AI Overviews and AI Mode count towards the “Web” search type in the Performance report, without a separate breakdown.
• Bing Webmaster Tools: the AI Performance report (public preview since 10 February 2026) shows citations in Microsoft Copilot, in AI summaries in Bing and with selected partners, per page and with a sample of the search phrases.
• Web analytics: visits show up under referring domains such as chatgpt.com or perplexity.ai. Mentions without a click do not appear there.
• Server logs: check fetches by the search crawlers and match IP addresses against the published lists, because bot names can be faked.
• Spot checks: regularly ask the same purchase questions in ChatGPT, Gemini, Claude and Perplexity. Answers vary, so evaluate several runs.

Checklist

With the ecomDNAI AI visibility audit, you see whether ChatGPT, Claude and Gemini with web search mention your brand and cite your pages, plus ready-to-paste blocks for robots.txt and structured data, for example. The SEO Audit checks, among other things, access for AI crawlers, and Keyword Visibility shows next to your Google positions whether AI search engines mention your brand.