How to track whether ChatGPT recommends your ecommerce brand
Aug 8, 2026
ChatGPT recommendations are not a single rank you can check once. Track a defined set of real shopping prompts over time, record whether your ecommerce brand appears, which competitors are recommended instead, and which sources ChatGPT cites. Then compare those answers with your customers’ actual questions. This separates a visibility problem from a product, evidence, or content problem—and gives your team a fixable backlog.
Start with a prompt set that represents buying decisions
A useful ChatGPT tracking program starts with the questions shoppers actually ask, not a list of brand-name searches. Include prompts that describe a need, a constraint, and a product category so you can see where your brand enters consideration.
Build the first set from four sources:
- Customer language: onsite search terms, support conversations, reviews, returns, and questions asked before purchase.
- Product comparisons: prompts such as “What are the best options for [use case]?” and “How does [category] compare with [alternative]?”
- Constraints: budget, size, materials, compatibility, location, delivery needs, experience level, and intended recipient.
- Category education: questions about how to choose, what features matter, and which tradeoffs shoppers should understand.
Keep each prompt specific enough to reproduce. “Best running shoes” is a weak test because the audience and use case are unclear. “Which trail shoes are best for wet, rocky terrain for a beginner who wants a wide toe box?” gives you a more useful buying scenario.
Record the country, language, device or account context when those variables matter to your market. ChatGPT’s shopping experience can produce personalized buying guidance and ask about factors such as budget and who the product is for, according to OpenAI’s description of shopping research. Your test set should therefore reflect the decisions your customers make, rather than assuming every shopper sees the same answer.
Measure more than whether your brand appears
Brand presence is the first metric, not the whole answer. For every prompt, capture the response, the recommended brands and products, the position or prominence of each recommendation, the cited sources, and the date of the test.
A practical tracking table looks like this:
| Field | What to record |
|---|---|
| Prompt | The exact shopper question tested |
| Brand presence | Absent, mentioned, recommended, or listed as a leading option |
| Recommendation context | The use case or reason ChatGPT gave for including you |
| Competitors | Brands and products recommended instead or alongside you |
| Sources | Pages, reviews, guides, retailers, publications, or other cited sources |
| Product facts | Features, prices, materials, availability, or policies stated in the answer |
| Accuracy | Correct, incomplete, outdated, or unsupported |
| Next action | Content, product-data, technical, merchandising, or off-site source work |
You can turn these observations into a simple share-of-voice measure: the number of tracked prompts where your brand is recommended divided by the total number of relevant prompts. Keep “mentioned” separate from “recommended.” A brand named in a long list has not necessarily won the buying decision.
Track competitor share in the same way. Count which competitors appear, how often they are recommended, and which use cases they own. The most useful finding may not be that one competitor appears more often overall; it may be that a smaller competitor consistently wins one high-value question, such as “best for sensitive skin” or “best for apartment living.”
Do not treat a single ChatGPT response as a permanent ranking. Run the same prompt repeatedly and compare results across dates. Shopping answers can incorporate current information and personal requirements, so one observation is evidence to investigate, not a market-wide conclusion. A trend across a stable prompt set is more useful than an isolated win or loss.
Find which competitors win recommendations
Competitors win ChatGPT recommendations by being present and legible in the evidence ChatGPT can use for a particular question. Your analysis should therefore connect each competitor to a prompt, a reason, and a source—not just produce a list of names.
Create a competitor matrix with columns for:
- Prompt or use case: the question in which the competitor appears.
- Recommendation reason: the attribute ChatGPT associates with that competitor.
- Evidence source: the page or publication supporting that association.
- Your position: absent, included, or included but described inaccurately.
- Commercial importance: the value of winning that use case for your business.
Look for repeated patterns. A competitor may win because its product pages explain compatibility more clearly. It may have independent buying guides that describe a specific use case. It may be cited by reviews or publications that your brand never appears in. Or its product data may simply make key differences easier to verify.
Separate three types of competitive loss:
- Coverage loss: ChatGPT cannot find a clear answer from your site or other sources for the question.
- Proof loss: your brand makes a relevant claim, but credible supporting sources do not repeat or explain it.
- Fit loss: your product is visible, but it genuinely does not match the shopper’s constraints as well as another option.
Only the first two are content or visibility opportunities. If a competitor wins because its product is better suited to the stated need, changing copy will not solve the underlying problem. That distinction keeps a content team from chasing every competitor mention.
Trace citations to discover what is holding you back
Citations reveal the information layer behind a recommendation. Review them for both presence and quality: which pages are cited, what facts they contribute, whether they describe your current products, and whether they answer the shopper’s actual question.
Group cited sources into categories such as:
- Your product, category, comparison, or help pages
- Independent reviews and editorial buying guides
- Retailer or marketplace listings
- Community discussions and user-generated content
- Manufacturer, standards, or institutional sources
- Location, shipping, returns, or policy pages
Then compare your cited-source profile with the competitors that win. If competitors are supported by clear category guides while your brand is supported only by short product descriptions, the gap is not simply “write more content.” It is “publish a page that explains the decision and makes the relevant product facts easy to verify.”
Check whether the source is current and internally consistent. ChatGPT may repeat an old price, discontinued product, incorrect material, or outdated shipping policy when those details remain visible somewhere. Treat an inaccurate recommendation as a data-maintenance issue as well as a content issue.
OpenAI describes its shopping research experience as reviewing information across the internet and bringing product options back for refinement; its product-discovery announcement also describes side-by-side comparisons and up-to-date product information. That makes source auditing and product-data accuracy part of the tracking workflow, not an optional editorial exercise. Read OpenAI’s product-discovery overview for the buyer-facing context.
Turn topic gaps into an ecommerce content plan
A topic gap is a question that matters to shoppers but is missing, unclear, or weakly supported in the sources ChatGPT uses. Prioritize gaps by buyer value and fixability, not by the number of missing pages.
Use this triage order:
| Gap | What to do first |
|---|---|
| Important product fact is missing | Add it to the product page and relevant structured product data |
| Comparison or tradeoff is unclear | Create a buying guide or comparison that explains who each option suits |
| Use case is not covered | Publish practical category content tied to the products that solve it |
| Policy or availability is unclear | Improve shipping, returns, stock, location, or compatibility information |
| Independent proof is absent | Identify legitimate review, expert, or customer-evidence opportunities; do not manufacture endorsements |
| Existing information is wrong | Correct the source, then rerun the affected prompts |
Write for the shopper’s decision, not for a model. A useful page should answer who the product is for, who should choose something else, the important tradeoffs, the relevant specifications, and the next step. Product claims should be specific enough to verify and consistent across your site and sales channels.
Use your own customer-question data to choose language and examples. Anagram’s site describes this as a loop: its Site Agent helps shoppers while they compare products, and the questions from those interactions show brands where to focus next. That can connect an AI recommendation gap to a real onsite friction point instead of relying only on invented test prompts. Learn more about Anagram’s AI Visibility product.
Use Anagram when you need a repeatable visibility workflow
Anagram’s AI Visibility product is designed to monitor how a brand appears in ChatGPT, compare it with competitors, identify citation sources, and surface topic gaps. Those are the four views a commerce team needs to move from “Are we showing up?” to “Why are we losing this buying question, and what should we change?”
The company also offers a branded Site Agent for conversational product answers and recommendations on a website, plus shopper-question analytics. That combination is relevant when the same team wants to improve both external discovery and the experience after a shopper arrives.
Use the product evaluation practically:
- Ask to see how prompts are defined and segmented by category, market, and use case.
- Confirm how the tool distinguishes a brand mention from a recommendation.
- Review how competitor share of voice is calculated.
- Ask how citations are collected and whether the underlying URLs are available for audit.
- Check how topic gaps become actionable pages, product-data fixes, or source priorities.
- Establish how often you can rerun tests and compare changes over time.
Anagram is a plausible fit for DTC and ecommerce teams that want customer-question insight connected to ChatGPT visibility work. If you only need a one-time manual snapshot, a spreadsheet and a fixed prompt set may be enough. If you need an ongoing view of recommendations, competitors, citations, and gaps across a growing catalog, a monitoring workflow reduces the operational burden.
A practical first measurement cycle
Start with a baseline that your team can repeat. Choose the highest-value categories, write a balanced prompt set, and test it before publishing changes. Save the full answers and citations so future results can be compared with the same evidence.
For each absent or weak recommendation, assign one diagnosis: coverage, proof, data accuracy, technical accessibility, or product fit. Give each diagnosis an owner and a next action. After the change is live, rerun the affected prompts and record whether the answer changed, whether the citation changed, and whether the competitor still wins.
The goal is not to force ChatGPT to mention your brand in every answer. The goal is to be a genuinely appropriate recommendation for the customers and use cases you serve—and to make the evidence for that fit clear wherever shoppers and AI systems look for it.