Tutorials6 min read
How to Check if ChatGPT Recommends Your Website
Test realistic customer prompts, record mentions and citations, and repeat the checks to see whether ChatGPT recommends your website over time.
You have two practical ways to check whether ChatGPT recommends your website: ask a consistent set of customer questions yourself, or use a monitoring tool to collect answers repeatedly. Start with the manual method so you understand what a useful result looks like before paying for automation.
Neither method gives you a universal position in ChatGPT. You are measuring what appeared in a particular set of answers under particular conditions. The goal is to find recurring patterns, not to treat one missing mention as a verdict on your business.
StackMango publishes this guide and provides an AI Rank Tracker. The examples below are a repeatable evaluation method, not a claim that we tested every monitoring platform.
Decide what counts as a recommendation
A brand mention means your company is named. A citation means the answer links to your site. A recommendation means the answer presents your business as a relevant choice. Record all three separately.
“Brand A exists” is a mention. “Brand A is worth considering for this use case” is a recommendation. A footnote linking to Brand A's help page is a citation and may say nothing about whether the product is suitable.
How to Check Your Website's Visibility in ChatGPT Manually
You can build a small manual panel using the access available in your ChatGPT account. Account and usage limits still apply.
Identify 10 to 20 questions that potential customers might genuinely ask ChatGPT.
Open ChatGPT and use a fresh conversation for each test.
Ask a natural question without mentioning the business name.
Record whether ChatGPT mentions your website or brand.
Check whether it provides a direct website link or citation.
Record which competitors appear.
Repeat the tests to identify variability.
Save the results in a spreadsheet.
Say you run a WordPress theme company. Your test panel might look like this:
What are the best lightweight WordPress themes?
Which WordPress themes are good for beginners?
What are the best WordPress block plugins?
Which WordPress themes are suitable for small businesses?
What are the best alternatives to Elementor?
Keep branded prompts such as "Is this brand good?" in a separate reputation panel. Naming the company changes the question and does not test unprompted discovery. What you actually want to know is whether the brand surfaces when you never say its name, because that's the situation a real buyer is in when they open a fresh chat and describe a problem instead of a product.
| Prompt | Brand mentioned | Recommended | Cited | Competitors |
|---|---|---|---|---|
| Best lightweight WordPress themes | Yes/No | Yes/No | Yes/No | Record |
| Best WordPress block plugins | Yes/No | Yes/No | Yes/No | Record |
| Best WordPress themes for beginners | Yes/No | Yes/No | Yes/No | Record |
This gives you a sample of responses, not a complete picture of what all ChatGPT users see. Location, personalization, and the search results available at that moment all shift the answer, so treat your spreadsheet as a snapshot of one set of conditions rather than a verdict on your site.
Keep test conditions as consistent as possible
Use the same prompt wording and record the date, visible model or mode, search setting and relevant location. Start a fresh conversation for each prompt, and avoid earlier messages that reveal which business you want recommended. If account personalization or memory is active, note that as part of the conditions rather than assuming a new conversation removes every influence.
Repeat the panel on several dates. Keep the initial panel unchanged while collecting a baseline, then add new questions as a separately labeled group. Otherwise a better percentage might just mean you replaced difficult questions with easier ones.
A practical starting panel could contain five category questions, five specific use cases and five comparisons. Choose questions from actual sales conversations, support tickets or customer research. A tool's generated suggestions can help you brainstorm, but they are not proof of search demand.
Record the answer as well as the score
Save enough of the response to audit the result later. Alongside each yes or no, record the URL cited, the competing brands, and a short explanation of why you classified the answer as a recommendation.
Use a consistent rule for ambiguous cases. A negative comparison should not count as a positive recommendation merely because it names you. If an answer uses an old company name or a product alias, note the match instead of silently changing your rules halfway through.
For an illustrative panel of 20 completed answers, six mentions equal a 30% mention rate. Three recommendations equal 15%, and two answers linking to your domain equal a 10% citation rate if all 20 were search-enabled and eligible. These figures describe the panel, not all ChatGPT users.
For more detail on denominators and position, read what a ChatGPT rank checker actually measures.
When to automate the checks
Automation becomes useful when keeping the log consistent takes more time than reviewing the findings. Compare tools on the workflow you need: saving exact prompts, keeping raw answers, separating mentions from citations, tracking competitors, retaining history and exporting results.
Do not choose a tool solely because its dashboard looks more confident. Ask whether it collects consumer-interface answers, API responses, an existing prompt database, or some combination. Those methods can serve different purposes and should be labeled in your report.
You can evaluate dedicated products such as Peec AI and Otterly.AI, or the AI reporting available from Ahrefs and Semrush. Their coverage and allowances differ, so use a current plan page rather than assuming that a free checker includes scheduled custom prompts.
For a fair trial, enter the same questions and brands in each candidate. Review the original answers and confirm that the export contains enough context to reproduce your interpretation. Do not compare headline percentages from different prompt panels as though they were measuring the same sample.
Run a panel with StackMango
In AI Rank Tracker, connect a supported provider account, add your brand and competitors, and save your customer questions. Follow the setup instructions to review the estimated cost before the first run.
StackMango collects API answers rather than reproducing every person's consumer ChatGPT session. Account access is free; provider API usage is billed separately. A consumer chat subscription does not supply those API credits.
Begin with one prompt and one engine, inspect the answer, and check that the brand and domain matching work as intended. Then run the rest of the panel and save a baseline before enabling a recurring schedule.
Compare visibility with visits and outcomes
Use website analytics to investigate referral visits and conversions separately. A cited URL is an opportunity for a click, not evidence that a visit happened. Referral data can also miss people who return through another route, so avoid treating it as a complete account of AI influence.
Keep three questions separate: did the answer mention us, did someone visit, and did that visit produce a useful outcome? They require different evidence. Combining them in one report is helpful; merging them into one “AI ranking” score hides the distinction.
What to do when your website is missing
Look for a pattern across prompts before changing your site. Read the sources and competitors that appear repeatedly, then compare what those pages explain with what yours currently provides. Missing product details, unclear positioning and outdated factual information are concrete problems you can investigate.
Choose one improvement, record the date, and repeat the same checks. Keep other settings stable where possible. If the answers change, describe the observation and the limits of the comparison rather than claiming a guaranteed ranking improvement.
The useful habit is a small panel that you review regularly. A manageable set of relevant questions with saved answers teaches you more than a large dashboard whose sampling method you never inspect.
Cover image: conceptual illustration, not evidence of product testing.
