AI Visibility for Online Marketplaces: Buyer Prompts, Seller Prompts and Wrong Fees

A marketplace has two sets of customers, and each asks AI different questions. Buyers ask about trust and cost. Sellers ask about fees and money. Say you run a hypothetical marketplace for used camera gear. These are the prompts worth testing:

Buyer prompts

• "Is it safe to buy a used lens on [marketplace], and what happens if it arrives broken?"
• "Does [marketplace] charge buyers a service fee?"
• "Best sites to buy used mirrorless cameras with buyer protection"
• "[Marketplace] vs [competitor] for used camera gear"

Seller prompts

• "What fees does [marketplace] take when I sell a camera?"
• "How long do payouts take on [marketplace], and is there a payout fee?"
• "Can I sell internationally on [marketplace]?"
• "Where should I sell used photography equipment for the lowest fees?"

The comparison prompts ("best sites", "lowest fees") decide whether you're named at all. The policy prompts decide whether what AI says about you is true. Track both, and keep them separate. A wrong payout timeline costs you sellers even when buyer recommendations look healthy. For the buyer side, see how online shops show up in AI shopping answers.

Each engine reaches your pages through a different gate, so treat them as separate channels.

• ChatGPT decides per query whether to search. OAI-SearchBot crawls for ChatGPT search, and OpenAI states each crawler setting is independent of the others. Without search, ChatGPT answers from training data. As of October 2026, OpenAI lists a Feb 16, 2026 knowledge cutoff for the GPT-5.6 family and Aug 31, 2025 for the Instant path.
• Google AI Overviews and AI Mode draw on Google's index. Google says a page must be indexed and eligible to show with a snippet to be a supporting link, with no additional requirements. The two surfaces may use different models, so check AI Overviews and AI Mode separately.
• Microsoft Copilot grounds on Bing's index. In Otterly's 2026 experiment, Copilot kept citing two sites after Google deindexed them, because Bing kept them indexed.
• Perplexity and Claude run their own search crawlers (PerplexityBot, Claude-SearchBot), plus user-triggered fetchers that follow different rules.

What this means: a fee question asked in ChatGPT with search off can return an old schedule no matter what you publish today. You control the retrieval-on path. You don't control the cutoff.

Marketplace fees change often, vary by category, and live across several pages. That's the setup for wrong answers. The common causes are all checkable:

1. The model answered from memory. If retrieval didn't fire, the answer reflects training data from before the cutoff. A fee change after that date is invisible to it.
2. The fee table only appears after JavaScript runs. Vercel's crawler study found ChatGPT and Claude crawlers fetch JavaScript files but don't execute them. A client-side fee calculator or help widget can look empty to them.
3. Your own pages disagree. In a hypothetical case, the seller fees page says 10%, an old help article says 8%, and the terms page says "see current schedule". An engine that retrieves the old article quotes it faithfully.
4. The policy sits behind a login. Seller dashboards and payout settings can't be fetched, so engines fill the gap from forums and comparison articles.
5. Third-party pages are out of date. An old "best places to sell" article with your previous fees can be the page an engine retrieves.

The fix is plain. State each fee, the buyer protection window and the payout timeline as a specific sentence in server-rendered HTML on one canonical page, with the date the schedule took effect. A specific fact on a fetchable page gives an engine a sentence it can quote. It doesn't guarantee the engine quotes it. Payout rules overlap with regulated money movement, and the fintech page covers similar ground.

When engines cite other sites instead of yours, being named on those sites is the route into the answer. That's an association in the evidence, not a promised effect. For marketplaces, check:

• Seller forums and communities where sellers compare fees and payout delays. These pages often carry the complaints AI repeats.
• Review sites covering buyer protection and disputes. See whether reviews affect AI recommendations.
• "Best marketplace for X" articles with fee tables. Ask publishers to correct outdated figures.
• Category directories and niche listings, covered in whether directory listings help AI answers.
• Old PDFs and mirrored copies of your seller terms.

Also scan third-party pages for hidden text or injected instructions aimed at AI. Research on LLM search agents shows manipulated content that reaches the result set can change what agents endorse. That's a reputational risk to watch, and using such tactics yourself is illegitimate.

Brand prominence is not a lever

Big marketplaces show up everywhere. In its index of 126M prompts (January to April 2026), Semrush found 36 brands held visibility across all four surfaces it tracked every month, among them Amazon and Walmart. That mostly measures brand size. Prominence is a strong association and the main uncontrolled confounder in GEO studies. No tactic buys it, so benchmark against marketplaces of similar size.

Marketplaces have millions of URLs: listings, categories, filters, help articles. Most citation studies and trackers match at the URL level. Ahrefs found that 37.9% of URLs cited in AI Overviews also appeared in the first 10 results (March 2026). Because that matching is per URL, a large site cited through a listing or help page that doesn't rank for the query counts as a miss.

What it means for you:

• Count citations at the domain level and by page type (listing, category, help, policy), not only by URL.
• Remember that citation isn't use. OpenAI's API returns a list of retrieved sources that is often larger than the list of citations.
• Google's generative AI report in Search Console shows impressions only, with no clicks, CTR or position, as of October 2026. Any "AI rank" for Google has nothing to be checked against.

Marketplaces run on templates, so one wrong directive spreads to thousands of pages. Run these checks:

1. Faceted and filter pages. Many marketplaces add noindex to filter URLs to manage crawling. That's often intentional. The problem starts when it spreads to category pages you want in answers, because Google requires indexing plus snippet eligibility for AI Overviews and AI Mode.
2. Snippet limits. Look for nosnippet, max-snippet:0 and data-nosnippet on listing, category and help templates. Google and Bing document these as controls that keep content out of AI answers. They're often inherited from a template meant to hide something else.
3. Per-bot robots.txt rules. Blocking GPTBot stops training use only. OAI-SearchBot controls ChatGPT search. A blanket AI block removes search eligibility too. SEO bots vs AI crawlers lists who does what.
4. Firewall and CDN. Marketplaces run heavy anti-scraping protection, and it can return 403 to an AI crawler that robots.txt allows. From 15 September 2026, Cloudflare blocks Training and Agent crawlers by default on ad-displaying pages for new domains and existing free-tier customers. See robots.txt vs firewall.
5. Rendering. Compare raw HTML with the rendered page for fee, protection and payout content.

A split policy that allows search but declines training looks like this:

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

Week 1: baseline

Write your eight buyer and seller prompts. Run them on each engine, logged out, and record what each says about fees, protection and payouts. The free checker on this site gives a snapshot for three buyer questions.

Week 2: access

Audit robots.txt per bot. Request a fee page and a category page as each AI crawler, for example with the AI Crawler Checker, and compare the results with robots.txt. Check Search Console's generative AI inclusion control, including inheritance on child properties.

Week 3: facts

Pick one canonical page per fee schedule and policy. Server-render it, date it, and fix or redirect contradicting help articles. Search templates for snippet directives.

Week 4: third parties and re-test

Contact publishers of outdated comparison articles. Re-run the prompts. Expect changes only in answers that used live search. Answers from memory won't move until a newer model ships. Nothing here guarantees a citation.

Should marketplace listings add Product or FAQ schema to get into AI answers? There's no evidence it helps. Google says structured data isn't required for generative AI search. Ahrefs' matched-control test of 1,885 pages found AI Overview citations fell 4.6% after schema was added, and FAQ rich results were deprecated in May 2026. Existing markup does no harm, so there's no need to remove it. It just isn't a lever for AI answers.

Does allowing GPTBot help ChatGPT recommend my marketplace? No. OpenAI documents GPTBot as a training crawler only. ChatGPT search uses OAI-SearchBot, and the settings are independent. You can allow OAI-SearchBot and block GPTBot. If your goal is appearing in ChatGPT search answers, check that OAI-SearchBot is allowed in robots.txt and that your firewall lets it through.

Why does ChatGPT keep quoting our old seller fee? Two likely reasons. If search didn't run, ChatGPT answered from training data with a fixed cutoff, so a newer fee is unknown to it. If search ran, it may have retrieved an outdated help article or third-party page. Publish the current fee once, in server-rendered HTML with its effective date, and fix or redirect the pages that contradict it.

Should we noindex all faceted pages? Noindex on low-value filter combinations is a normal crawl decision. Keep it off category and filter pages that match real buyer questions, because Google only uses indexed, snippet-eligible pages as supporting links in AI Overviews and AI Mode. Review the template rule, not individual pages, since one directive applies to thousands of URLs.

Can a niche marketplace get recommended over the giants? Nobody can promise that. Large brands dominate citation data, and that reflects brand size, not a technique. What you control is whether engines can reach and quote your pages, and whether your facts are consistent. Niche prompts with specific categories or seller needs are where a smaller marketplace has the most realistic chance of being named.