How to improve your visibility in ChatGPT answers
Neaptide · August 26, 2026 · 5 min read · Updated: October 3, 2026
Check how ChatGPT describes your business, fix access issues and build pages that help customers choose. A practical measurement method with clear limitations.
On this page
Start with a question a customer might ask before hearing about your business: “Which booking tool suits a two-person studio?” Save ChatGPT’s answer and open its sources. That gives you a baseline: which options it suggests, what information it draws on and what your own pages may be missing. You can make those pages more accurate, easier to reach and more useful. You cannot choose ChatGPT’s recommendations. Repeat the checks under comparable conditions to see what changes.
Choose questions that matter to customers
Collect questions from enquiries, support tickets and search data, then match each to a page. For a booking service, “how can I take bookings without a receptionist?” calls for an explanation of the workflow. “Which tool suits a two-person studio?” needs selection criteria, and “can I import a spreadsheet?” needs an import guide. Mark questions with no useful page as content gaps. Keep questions naming your brand separate from those asked by people who have not heard of you.
Check access to the pages
OpenAI distinguishes OAI-SearchBot, used for search, from GPTBot, which collects material that may be used for training. Their controls are independent.
- Check that the public URL returns the actual page without login or a CAPTCHA.
- Review robots.txt and any CDN or server blocks for the intended crawler.
- Make important conditions available as text. Check what a URL fetch returns if content normally appears only after browser interaction.
Write pages that support a decision
Explain who a service suits, what is included, what is not and how to begin. For software, add compatibility and limits. An import guide should name the file formats and required fields, show an example and explain how to check for errors. Use figures you can support, and describe what affects delivery times.
Keep a useful guide together instead of splitting it into near-duplicate pages, and let each answer be as long as the question needs. Check the business details in your public profiles too. Independent reviews can help readers understand the offer, but there is no established formula linking purchased mentions to more ChatGPT recommendations.
Set up a repeatable measurement
- Start with, say, 20 non-brand questions and a separate branded set. That is a sample you can check by hand, with no claim to being a statistical standard.
- Record date, language, region, search mode and the displayed model. Use new conversations with comparable settings.
- Save answers and URLs. Record brand mentions, links to your site and explicit recommendations separately.
- Repeat on different days. Mark service errors separately from valid answers that do not mention you.
Mention rate = answers naming the brand / successful answers × 100%
Example: 6 of 20 answers = 30% in this sample.The figure describes your test and says nothing about ChatGPT users at large. If you change the question set, start a separate series. Track visits and enquiries separately, and remember that missing referral data does not prove AI played no role.
What our own example tells us
Our published Turkish vocabulary test audit recorded mentions in 7 of 7 answers across six languages on August 24, 2026, with first or joint recommendations in 6. This was our own product and a small one-day sample. It does not isolate the effect of markup, llms.txt or any other change, or predict another business’s outcome.
What to ask an agency to deliver
Ask for an initial measurement, a list of problems and a first page revised to answer a real customer question. Check the published page, then repeat the observations. The report should include saved answers, test conditions and a record of changes. Where enquiries can be traced to a source, report them separately. There is no universal timetable or guaranteed increase in citations.