When I ask ChatGPT for the best products in our category, our brand never shows up, even though we're a top seller on Amazon. Why?
A top Amazon seller can still be missing from ChatGPT's best-product answers. Here is how we diagnose the gap for clients, and what we change first to fix it.
This is one of the most frustrating calls we get. A brand is a top seller on Amazon, the sales team is proud of it, and then someone asks ChatGPT for the best products in the category and the brand is nowhere. The reaction is usually "the AI is wrong." It is rarely wrong in a way you can complain about. It is working from different evidence than your sales rank.
Why Amazon rank does not carry over
Sales rank measures purchases on Amazon. A "best products" answer in ChatGPT draws on what the engine can retrieve and read across the web, and then favors brands that those sources discuss clearly and consistently. Amazon rank is not part of that. The sources that tend to show up in these answers in many categories are editorial roundups, review sites, forums and video, along with the brand's own site. If those places do not talk about your brand, the engine has little to say. We cannot see inside the engine, so we test hypotheses against data.
Our diagnostic order
1. Define the real prompts and track them
We write the questions a buyer would type: "best [category] for [use case]", "[brand] vs [competitor]", "is [brand] worth it". We track them daily, since ChatGPT answers vary run to run. In Promptwatch we monitor ChatGPT Search through its real interface and watch Visibility Score, Share of Voice and Position in Answer by prompt. Persona stacking lets us run each prompt as different buyer types.
2. Read the winning citations
On the prompts the client loses, we open the responses and list the cited sources. Promptwatch citation analytics separates pages, domains, Reddit, YouTube and LinkedIn citations, and it lists offsite mentions where a brand is named without a link. In our experience this step usually reveals the real problem: competitors are being written about in places the client has no presence.
3. Check crawler access
We look at Agent Analytics, which shows IP-verified AI crawler logs from bots like GPTBot, OAI-SearchBot and ChatGPT-User, with error tracking. If a client's own pages are returning errors to these bots, we fix that first. Crawler logs start on Professional.
4. Check shopping surfaces separately
The text answer and the shopping carousel are different surfaces. Promptwatch covers AI Shopping for ChatGPT shopping carousels and Amazon Alexa product results, with own-catalog upload. Shopping Insights starts on Business ($579/mo). We check whether the client's products appear in those results at all.
5. Check brand identity
If a brand name is a common word, or spelled in several ways, mentions may not be counted or connected. We set a Brand Book and brand aliases in Promptwatch so the counts are fair.
The usual findings
We see the same causes again and again, though we do not put numbers on how often.
- No presence on the sources engines read. Strong marketplace sales, but no coverage in roundups or forums.
- Own site written for buyers who already decided. Product pages with specs but no comparisons or use cases.
- Inconsistent brand naming across site, listings and press.
- Crawler blocks from a blanket AI-bot rule.
- Thin product data. On the Google side we start with Merchant Center feeds and read Google's own documentation for current AI Mode shopping requirements.
What we do about it
First we fix access, since it is quick. Second, we add comparison and use-case content to the client's site, written answer-first. Promptwatch content gap analysis shows which prompts lack coverage, and Content Refresh suggests updates to existing pages. For clients on Business, the Content Agent can plan and draft against gaps, with a review inbox so a human approves every piece.
Third, we work the off-site list: review roundups, communities, video. This is slow, and it takes outreach and real contributions, not tricks. The Action Board in Promptwatch keeps the list prioritized.
Fourth, we align naming and product data across the client's properties.
What we tell the client up front
We cannot make ChatGPT recommend a brand. We can show which prompts the brand loses, who wins them, and which sources the winners rely on. We also tell clients not to treat Amazon rank as a bridge. It is a strong commercial signal and a weak AI one.
When to expect movement
We re-measure after two weeks, then monthly. Single-day swings are noise. Off-site work takes longer than onsite fixes, and clients should hear that early.
If you are in this position, start with the tracking and citation step. It turns "the AI ignores us" into a list of specific pages, sources and fixes, which is something a team can actually work through.