How to check if your Shopify store is in ChatGPT

The most common self-audit method is wrong. A merchant opens ChatGPT, types their brand name, sees themselves mentioned, and concludes they are visible. In practice that tells you almost nothing. ChatGPT Memory retains the stores you have asked about before and biases retrieval toward them. Your session logged in as you is not representative of a first-time buyer prompting from a fresh account. And the three retrieval paths (ACP Shopping, Bing editorial, GPTBot direct crawl) answer different query classes — a brand that is visible on one can still be invisible on the other two.

A defensible self-check tests all three retrieval paths, uses a fresh session with Memory disabled, uses a spread of prompt classes (brand, category, comparison, problem-statement), and runs on a consistent weekly cadence. That is a 15-minute ritual, not a one-off.

Before you run a single prompt, get the environment right. Your goal is to approximate a first-time buyer with no prior relationship to your brand. That means ChatGPT cannot know who you are and cannot bias toward stores you have been researching.

1. 1Open ChatGPT in a private/incognito browser window. This clears cookies, which includes your session identifier. If you are a ChatGPT Plus subscriber, log out — Memory is on by default on Plus accounts.
2. 2In Settings → Personalization → Memory, confirm it is off. On a free account this is the default; on Plus it is not.
3. 3If you sell to a specific geography (e.g., US-only) but are testing from India, use a VPN set to your target market. ChatGPT Shopping routes regionally and will hide stores that do not ship to your IP's country.
4. 4Close every other ChatGPT tab. Two sessions running simultaneously can cross-contaminate suggestions.
5. 5Pick one model (GPT-4o or GPT-5) and stay on it for the entire panel. The two models rank retrieval slightly differently and mixing them will muddle your week-over-week comparison.

The panel below is the one we ship to every Surfient customer as a starter. You should customize the category and problem-statement prompts to your catalog — what matters is the shape of the panel, not the exact wording. Five prompts per class is the minimum useful N; fewer and you cannot tell signal from randomness.

Brand (5 prompts)

'Tell me about {brand}', 'Is {brand} legit', 'What does {brand} sell', '{brand} vs {competitor}', 'Where can I buy {brand} products online'. Answers your retrieval-hygiene question: does ChatGPT know who you are.

Category (5 prompts)

'Best {category} under $X', 'Most popular {category} in 2026', 'Where to buy {category} online', '{category} with {attribute}', 'Cheapest reliable {category}'. Answers your retrieval-reach question: do you surface without being named.

Problem-statement (5 prompts)

'I need {outcome} for {context}', 'What is the best product for {problem}', 'How do I solve {problem}', 'I am looking for a gift that {attribute}', '{persona} is shopping for {occasion}'. Answers your commercial-intent question: do you rank on the money prompts.

Comparison (5 prompts)

'{you} vs {competitor1}', 'Alternatives to {competitor2}', 'Who makes better {category}, {you} or {competitor3}', 'Pros and cons of {competitor4}', 'Honest review of {competitor1}'. Answers your competitive-positioning question: does ChatGPT frame you fairly against rivals.

What to record for each prompt

• Citation: is your store named, linked, or mentioned? (Yes / No)
• Position: where does your mention land in the ordered list? (1, 2, 3… or N/A)
• Sentiment: is the mention positive, neutral, or negative? (+1 / 0 / -1)
• Competitor count: how many other stores appear in the same answer? (integer)
• Source basis: which retrieval path looks like it fed the answer? (ACP / Bing / GPTBot crawl / unknown)

A store that does not appear in any of your 20 prompts has one of three failure modes: the crawler cannot reach you, the retriever cannot parse you, or the ranker cannot rank you. The first one is easy to diagnose and is often the actual blocker. You need a 7-day grep of your server logs for the OpenAI user-agents.

# On a typical Nginx / OLS / Apache log directory
grep -E 'GPTBot|ChatGPT-User|OAI-SearchBot' /var/log/access.log* | wc -l

# If your Shopify store routes through Cloudflare, pull the Firewall Events
# feed from the Cloudflare dashboard instead — the origin logs may not see
# requests that were blocked at the edge.

Expected healthy numbers, at time of writing: 50-500 GPTBot hits per day on a 200-product catalog, 10-80 OAI-SearchBot hits per day, 1-5 ChatGPT-User hits per day (these are live user-triggered fetches). Zero hits on any of the three is a red flag. If you see zero, check your Cloudflare Bot Fight Mode, your Shopify bot-management rules, and your theme's robots.txt.liquid.

A single panel run is a snapshot. What you actually care about is the trendline. ChatGPT's retrieval indices refresh on an observed cadence of 7-14 days, so daily noise will muddy any conclusion you try to draw from hour-to-hour variation. Run the panel on the same day of the week, at roughly the same time, for at least four consecutive weeks before you declare anything 'fixed' or 'regressed'.

The three metrics that matter

Citation count

Out of 20 prompts, how many mention your store by name? 0-20 scale. The headline visibility number.

Weighted position

Sum of (1 / position) for each mentioned prompt. Rewards front-of-list citations; a position-1 mention counts 5x more than a position-5 mention.

Share of AI Voice

Your citation count / total citation count across you plus top 4 named competitors, as a percentage. This is the GEO equivalent of Share of Search.

After one run of the panel and one log grep, you will land in one of four diagnostic buckets. Each maps to a specific fix priority, and getting the diagnosis right saves you weeks.

1. 1GPTBot crawls, panel mentions ≥6/20, position mostly ≥3. You have the basics right but you are ranking thin. Priority: description quality and schema richness. See the 'Writing product descriptions for AI' guide.
2. 2GPTBot crawls, panel mentions ≤3/20, low on category prompts specifically. You have branded titles or an incomplete feed. Priority: feed hygiene (GTIN, variants, GPC) and de-branded titles.
3. 3GPTBot crawls, panel mentions 0/20 even on brand queries. You are enrolled but your content is stale or your structured data is absent. Priority: FAQPage + Product schema on every PDP and a llms.txt refresh.
4. 4GPTBot does not crawl. Everything else is downstream of this. Priority: unblock crawlers at every layer (Cloudflare, Shopify Bot Management, theme robots.txt) before you write another product description.