Blog about GEO and SEO

Why AI assistants quote competitors instead of your online shop

AI assistants quote the source that is easiest to read, easiest to verify and most often confirmed elsewhere – and for most product queries that is a marketplace, a review site or a competitor with clearer pages, not your shop. If your product pages render only in JavaScript, lack structured data, hide prices behind a cart, or are never mentioned on third-party sites, the model has nothing quotable to work with. The fix is mechanical, not mysterious: make your pages crawlable, make the facts explicit, and make sure other sites confirm them.

How does an AI assistant choose which shop to quote?

Most AI answers about products are built in three steps: the assistant reformulates your question into several search queries, pulls a handful of pages from a search index (Bing, Google or its own), then writes an answer using only the text it managed to fetch. A page that cannot be fetched or parsed simply does not exist in that process.

Within the fetched set, the model prefers sources that state facts plainly: a price with a currency, availability, specifications in a list, a clear "who is this for" paragraph. Vague marketing copy gives the model nothing to cite, so it falls back to a competitor who wrote the numbers down.

Finally, models tend to favour claims that appear in more than one source. If your shop says "fastest delivery in the region" and no other page on the internet repeats it, the model treats it as an unverified marketing claim. If a review blog, a forum thread and a comparison article all mention a competitor's shipping terms, that becomes the "safe" thing to quote.

Why do marketplaces and review sites outrank my shop in AI answers?

Marketplaces win on three mechanics at once. Their product pages are template-consistent, they carry structured data for price, stock and ratings, and they accumulate user-generated Q&A and reviews that read like exactly the content a user asked about.

Review sites and "best of" listicles win because the query itself is comparative. When someone asks "best standing desk under [price]", the assistant looks for pages that already compare several options. A single-product page cannot answer a comparative query, so the assistant quotes whoever did the comparison – and your shop appears only if that listicle mentions you.

There is also a simple volume effect: a marketplace has thousands of pages indexed for the same topic cluster, so its chance of being in the fetched set is mathematically higher. You cannot outspend that, but you can out-specify it on your own narrow range.

Is my shop even readable by AI crawlers?

Many shops are invisible for a boring technical reason. AI crawlers are far less patient than Googlebot: several of them fetch raw HTML and do not execute JavaScript, so a product page that builds its price, description and specs client-side arrives as an empty shell. Behaviour differs by service and changes over time, so test rather than assume.

Run this check on your own product page:

  1. Open the page source with JavaScript disabled (or curl the URL) and look for the product name, price, availability and at least two paragraphs of description in the raw HTML.
  2. Check your robots.txt for blanket Disallow rules and for named AI user agents.
  3. Check your WAF or CDN bot rules – many shops block unknown bots by default, which silently blocks AI crawlers.
  4. Confirm the page returns HTTP 200 without a cookie wall, geo-redirect or age gate for a bot with no cookies.
  5. Check that prices are not loaded from a separate API call after render.

A reasonable starting robots.txt for a shop that wants to appear in AI answers:

User-agent: *
Allow: /
Disallow: /cart
Disallow: /checkout
Disallow: /*?sort=
Disallow: /*?filter=

User-agent: GPTBot
Allow: /

User-agent: OAI-SearchBot
Allow: /

User-agent: ChatGPT-User
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: ClaudeBot
Allow: /

User-agent: Google-Extended
Allow: /

Sitemap: https://example.com/sitemap.xml

Note the nuance: Google-Extended governs Gemini and grounding use, not whether you appear in Google AI Overviews – AI Overviews draw on the normal Google index, so standard Googlebot access is what matters there.

What makes a product page quotable?

Write the facts in a form a model can lift in one sentence. The practical test: can someone answer "what is it, who is it for, what does it cost, when does it arrive, what's the catch" by reading only your page?

SignalTypical competitor pageTypical shop page that gets skipped
PriceIn HTML, with currency and tax noteLoaded by JS, or "request a quote"
Availability"In stock, ships in 1–2 business days""Available"
SpecsLabelled list: material, weight, size, warrantyProse paragraph of adjectives
Comparison"Model A vs Model B" sectionNone
ReviewsVisible text + AggregateRating markupWidget loaded from a third-party script
Returns/shippingStated on the product pageSeparate page only

Add a short FAQ block to each important product and category page, answering the questions support actually receives: sizing, compatibility, delivery times, returns, warranty. These blocks are the single most frequently quoted format in AI answers because they already match the shape of a question.

What structured data does an online shop need?

Structured data does not force a citation, but it removes ambiguity about price, stock and ratings. Put Product with Offer on every product page, BreadcrumbList on every page, Organization on the homepage and FAQPage where you have a real FAQ.

{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": "Oak Standing Desk 140x70",
  "sku": "DSK-140-OAK",
  "brand": {"@type": "Brand", "name": "Example"},
  "description": "Electric height-adjustable desk, 62–128 cm, solid oak top, 80 kg load.",
  "image": ["https://example.com/img/desk-140-oak.jpg"],
  "offers": {
    "@type": "Offer",
    "price": "749.00",
    "priceCurrency": "EUR",
    "availability": "https://schema.org/InStock",
    "itemCondition": "https://schema.org/NewCondition",
    "url": "https://example.com/desks/oak-140",
    "shippingDetails": {
      "@type": "OfferShippingDetails",
      "shippingDestination": {"@type": "DefinedRegion", "addressCountry": "DE"},
      "deliveryTime": {
        "@type": "ShippingDeliveryTime",
        "handlingTime": {"@type": "QuantitativeValue", "minValue": 1, "maxValue": 2, "unitCode": "DAY"},
        "transitTime": {"@type": "QuantitativeValue", "minValue": 2, "maxValue": 4, "unitCode": "DAY"}
      }
    },
    "hasMerchantReturnPolicy": {
      "@type": "MerchantReturnPolicy",
      "applicableCountry": "DE",
      "returnPolicyCategory": "https://schema.org/MerchantReturnFiniteReturnWindow",
      "merchantReturnDays": 30
    }
  },
  "aggregateRating": {
    "@type": "AggregateRating",
    "ratingValue": "4.6",
    "reviewCount": "128"
  }
}

Only mark up ratings you genuinely display on the page. Fabricated review counts are a well-known cause of manual actions in Google and give an assistant a reason to distrust the whole page.

Can an llms.txt file help a shop?

llms.txt is a proposed convention: a plain Markdown file at the root that points AI systems to your most useful pages. Support is not guaranteed by any major assistant, so treat it as cheap insurance rather than a ranking factor.

# Example Shop

> Online shop for height-adjustable desks and ergonomic chairs, shipping across the EU.

## Buying guides
- [How to choose a standing desk](https://example.com/guides/choose-desk): sizes, motors, load, warranty
- [Desk size calculator](https://example.com/guides/desk-size): room and body-height recommendations

## Policies
- [Shipping and delivery times](https://example.com/shipping)
- [Returns within 30 days](https://example.com/returns)

## Categories
- [Standing desks](https://example.com/desks)
- [Ergonomic chairs](https://example.com/chairs)

Why do competitors appear in "best X" answers and I don't?

Because those answers are assembled mostly from third-party pages. An assistant answering "best shop to buy X in [country]" rarely cites a shop praising itself – it cites reviews, comparisons, forum threads and press mentions.

What moves the needle:

  1. Get listed in the category roundups that already rank for your core queries – pitch the author with concrete data, not a banner.
  2. Keep profiles accurate and complete on the review platforms your buyers use, including name, address, categories and product list.
  3. Answer questions in communities where your buyers ask them, with specifics that are correct whether or not anyone clicks through.
  4. Publish your own comparison pages ("X vs Y", "alternatives to Z") that include competitors honestly – these get fetched for comparative queries.
  5. Make sure brand name, legal entity and contact details are identical everywhere, so the model can merge the mentions into one entity.

How do I check whether anything changed?

Track three things, separately. First, citations: ask the assistants a fixed list of 20–40 buyer questions on a schedule and record which domains are cited. Manual checks work; automated tracking saves hours.

Second, referral traffic. Filter analytics for referrers such as chatgpt.com, perplexity.ai, copilot.microsoft.com and gemini.google.com. Volumes are usually small compared with search [check: typical share for e-commerce], but conversion intent is high because the user arrives after an explicit recommendation.

Third, server logs. Look for AI crawler user agents and check which URLs they fetch and what status codes they receive. A crawler hitting 403 or receiving an empty HTML shell explains the absence of citations faster than any content audit.

FAQ

Does blocking GPTBot hurt my visibility in ChatGPT?

It can. GPTBot is used for crawling that feeds training and related systems, while OAI-SearchBot and ChatGPT-User relate to search and on-demand fetching. Blocking all of them removes your pages from consideration; blocking only GPTBot is a middle option if you object to training use.

Will structured data alone get my shop quoted?

No. Structured data clarifies facts for machines, but it does not create demand or authority. It works when paired with readable HTML, substantive page text and third-party mentions that confirm what you claim.

Should I worry that AI answers reduce clicks to my product pages?

Partly. Informational queries lose clicks; transactional ones still need a checkout, and being the cited shop in a recommendation is high-value placement. The practical response is to be the source that gets named, not to hide content from assistants.

How long does it take to appear in AI answers after fixing a page?

It depends on re-crawl frequency and whether the assistant uses a live index or a cached one. Changes that depend on search indexes often surface within days to weeks; mentions on third-party sites take longer [check: typical observed lag].

Do I need separate content for AI and for Google?

No. The same qualities help both: crawlable HTML, clear answers near the top, specific facts, consistent entity data. Writing a second "AI version" of a page adds duplication risk without benefit.

If you want to see which of these gaps apply to your shop, run a free check in Zgadka – it scores your GEO and SEO visibility, shows where crawlers hit walls, and generates the robots.txt, llms.txt and schema.org files you can paste in.


Fazier badge